Walk the average mid-sized consultancy, specialised distributor, or professional services partnership and you will find the same artefacts of seriousness. An AI steering group. A list of use cases ranked by “impact” and “feasibility.” A pilot chatbot on the service desk. A copilot licence for the analysts. A workshop with sticky notes on the wall. A slide that says transformation.

What you will rarely find is a new unit of sale.

The firm still captures value through consultant-days, service-hours, person-months, reactive tickets, and bespoke statements of work. AI has been invited to decorate that unit — to draft faster, summarise more, search better — without being asked to replace it. That is how you end up with a programme that looks modern and a commercial model that is still a horse, only now the horse has a dashboard.

The question this book answers

Why does “find an AI use case” keep producing trinkets — and what exactly should the firm be hunting for instead?

After this book you should be able to define your successor offer; run it through eight ordered gates (friction → suppressed promise → constitution → commercial boundary → delivery physics → unit economics → transfer → strategic migration); price it against a stable unit of value rather than hours; and say precisely which old commercial unit it is built to compress.

The public pattern: adoption without a new commercial object

None of this is a private complaint. McKinsey’s 2025 global survey on the state of AI still finds most organisations in experimentation or piloting, with only about a third reporting that they have begun to scale AI across the enterprise.¹ MIT’s NANDA initiative, in The GenAI Divide: State of AI in Business 2025, reports a sharper commercial picture: despite tens of billions of dollars of enterprise GenAI investment, the vast majority of organisations show no measurable P&L return, and only about five percent of custom enterprise AI tools reach production with sustained impact.² For professional services specifically, the same report’s disruption framing is blunt: efficiency gains appear; client delivery remains largely unchanged.²

Those figures do not prove this book’s framework. They prove the environment in which the wrong object keeps getting funded. Use-case inventories and pilot funnels can grow while the commercial unit through which the firm sells — and through which an attacker will disintermediate it — stays frozen. Independent coverage of the MIT NANDA findings has stressed the same commercial reading: most generative AI pilots are not delivering measurable business results even where model quality is not the bottleneck.³

What counts as a trinket here

A merely good AI project might cut report-writing time, draft quotes faster, search manuals, or classify tickets. Useful. Worth doing as operations spend. Relative to the prize, it is still a trinket if success leaves the labour-priced unit intact.

The Terminal Value Doctrine already named the altitude problem: when the cost of cognition collapses, value migrates; defending the old layer is how excellent operators become irrelevant. AI-Constituted Services named the category problem: most “AI-native” labels fail a simple counterfactual — remove the AI and the same offer still exists, only slower.

What the corpus had not yet defined as a first-class commercial design object is the successor: not only a service that could not exist without machine-scale cognition, but the offer built to replace the firm’s current unit of sale. “Successor” adds the replacement relation — and therefore the defensive urgency — that “constituted” alone does not name.

The prize is not an AI capability. It is the AI-native successor offer.

Honesty about the evidence base

Where this book goes — and stops

Part I names the wrong object and the four ideas compressed into “finding the car,” then defends the successor-offer definition clause by clause.

Part II states the discovery question — suppressed promises, not use cases — and the vocabulary of successor units.

Part III is the qualification machinery: why gate order matters; eight gates with failure modes; scarce-expert elasticity; the destruction test.

Part IV is the proof: two design specimens, then a worked negative case.

Part V positions the market (knowledge-intensive services; problem shape, not industry), the firm hierarchy (offer → governed delivery system → foundry), and the Monday operating moves.

It will not re-teach the foundry’s six functions, the remove-the-AI checklist’s internals, offer-build engagement mechanics, either specimen’s full product anatomy, founder-knowledge succession machinery, three-kernel runtime design, or enterprise portfolio Kill/Fix/Double-Down. Those organs exist elsewhere. This book owns the successor offer as a commercial design object — and the gates that qualify it.

“Don’t optimise the horse. Find the car.” That line has done real work inside strategy rooms. It contrasts horse optimisation — AI that makes the existing labour model slightly faster — with car construction: a different commercial architecture for a world where cognition is cheap.

As client-facing language it is weaker. It needs explanation. It sounds like technology strategy. Worse, as a design instruction it currently compresses four different ideas into one slogan. Teams nod, then disagree about what they just agreed to build.

The four ideas

They belong together. They are not the same test.

A firm can discover a new technical possibility and still sell it as days. It can remove friction and still measure success in utilisation. It can productise the brochure without productising delivery. It can narrate defence while every “product” remains a hero-led project with a nicer name.

Where the successor offer sits

Two prior frames meet at the object this book owns.

Car Discovery — from the Terminal Value Doctrine’s project altitude work — asks what future product, channel, business model or value architecture should replace the old one. It is the strategic question: which car should exist?

AI-Constituted Services asks what commercial service could not exist in recognisably the same form without machine-scale cognition. It is the category question: what shape can that future take when the machine is constitutive, not decorative?

Not every “car” is a service. Not every constituted service is important enough to become the company’s successor. For knowledge-intensive service businesses — this book’s market — the overlap is the prize.

Car Discovery
  (what future commercial model should exist)
            ∩
AI-Constituted Services
  (what commercial shape requires machine-scale cognition)
            ↓
Successor-offer candidate
  (still must pass ordered gates: boundary, physics,
   economics, transfer, strategic migration)

That last line matters. Overlap produces a candidate, not a graduation certificate. Chapters 6–10 are the gates that keep charismatic technical demos from being declared the firm’s future.

Why the compression was tempting

Compression is efficient in conversation. Partners say “find the car” and mean “stop funding horse optimisation.” The danger is operational. Delivery leaders hear possibility and build a prototype. Sales leaders hear productisation and invent a SKU. Strategy leaders hear defence and commission a threat deck. Innovation leaders hear friction and ship a widget that customers like and finance cannot price.

Each group can claim progress. None of them has necessarily changed the unit of sale.

Worked misreads of each idea alone

Possibility without friction. A firm builds an impressive multi-source reconciling engine for a problem customers do not pay to remove. The demo wins internal awards. Pipeline stays empty. Possibility is real; demand is not.

Friction without productisation. A firm correctly diagnoses that customers hate the pre-sales dance, then “fixes” it with a free workshop and a chatbot. Friction falls slightly. The unit of sale remains days. The firm has improved hospitality, not commercial architecture.

Productisation without constitution. A firm launches three named packages with fixed prices. Under the hood, every package is still a senior reconstructing the world. Fixed price becomes a bet on heroics. When exceptions hit, change requests return, or margin dies. The brochure was productised; delivery was not.

Defence without transfer or migration. A strategy offsite names the AI-native threat and funds a “defensive product.” Engagement one is founder-led. Nothing compresses the old unit firm-wide. The threat deck was accurate; the response was theatre.

These misreads are not straw men. They are the default ways a compressed slogan gets implemented by functions that each own only one of the four ideas. The successor offer is the discipline of holding all four until the object is one commercial design — then still running the ordered gates in Part III.

They belong together. They are not the same test.

A better object for the market

“Car Discovery” remains useful internally. The client-facing, board-legible object is the AI-native successor offer: a commercially bounded customer promise that was previously uneconomic or impossible to sell, made viable through machine-scale cognition, governed delivery and retained human authority — and defined by the labour-priced unit it is built to compress.

That sentence will be dismantled clause by clause in the next chapter. For now, hold only this: if your AI initiative cannot say which of the four ideas it is serving — and whether it sits at their overlap — it is not yet pointed at the prize.

That is the commercial design object this book mints. The rest of this chapter defends each clause. Later chapters turn the defence into gates, metrics and specimens. If a clause is optional in your mouth, the object will dissolve back into “AI project.”

Bounded

Providers cannot own whole business outcomes. “Your project will succeed,” “your margin will improve,” “your uptime is guaranteed” often depend on factors the supplier does not control: the customer’s operators, weather, upstream vendors, political decisions inside the buyer’s organisation.

A successor promise terminates in a bounded commercial object: a verified decision, a maintained state, a protected period, an assessed estate, an enrolled fleet, or a response commitment. The vocabulary expands in Chapter 5. The principle is already load-bearing: bound the promise so a buyer can recognise completion and a supplier can keep it without lying about control.

Named

A buyer should understand what enters, what the firm does, what state or deliverable comes out, what remains uncertain, what the customer must decide, what is included, what triggers additional work, and what it costs. Without that legibility, you may have excellent customised consulting. You do not yet have a successor offer.

Naming is not branding theatre. It is the condition under which the firm can measure win rate, exception density, delivery margin and transfer without every deal being a unique snowflake. Commercial legibility is a property of the unit of sale, not a marketing afterthought.

Priced — against a stable unit of value

The general rule is not “always one fixed fee.” The general rule is: priced against a stable unit of customer value or supplier commitment, not primarily against elapsed human effort.

That allows fixed-price readiness reviews, annual fleet-continuity subscriptions, per-project activation fees, reserved-stock charges, per-machine onboarding, and premium response tiers — provided the unit is stable enough to configure, band and explain.

Newly economical because of machine-scale cognition

The promise was previously uneconomic or impossible to sell honestly because breadth, frequency, specificity, coordination or uncertainty would consume expert labour at a rate no buyer would fund and no firm could staff. Machine-scale cognition absorbs that load. Humans retain review, authority and consequence.

This is the constitutive claim. AI is not a feature badge. It is the economic substrate that allows the offer to exist in productised form. AI-Constituted Services owns the category test; this book inherits the consequence for commercial design.

Compresses an existing labour-priced commercial unit

This is the word successor doing real work. A candidate that leaves the consultant-day, reactive ticket, unpaid discovery SOW, or ad hoc expert reconstruction untouched may still be a useful AI product. It is not the successor offer for the firm’s unit of sale.

Compression does not mean vandalising current revenue overnight. Harvest, migrate and construct can run in parallel. It does mean the firm knows which revenue is funding the future and which revenue the future is intended eventually to compress. Chapter 10 makes that explicit as a destruction test.

Converts expertise into governed delivery machinery

The firm’s best knowledge must participate in substantially more customer outcomes without requiring the same experts to reproduce the whole analysis each time. Old shape: the scarce expert reads, remembers, joins, decides and explains every case. Successor shape: the system reads and joins; compiled judgment shapes every case; authorised humans handle material dispositions.

That is not “replace the expert.” It is convert expertise from variable labour into a compounding productive asset. The governed delivery system that makes this safe is a distinct object — Chapter 13 places it in the hierarchy without re-deriving its anatomy.

Meters residual scarcity

AI does not repeal physical economics, regulated authority, or genuine judgment under consequence. Stock, freight, branches, working capital, licensed sign-off, and scarce dispositions remain. The successor offer meters what is still scarce and prices it — it does not pretend marginal cost is zero.

A capital-equipment distributor still has warehouses. A consultancy still has principals who must own certain findings. Honesty about residual scarcity is what keeps fixed envelopes from becoming death marches.

More transferable and profitable across engagements

Engagement one can be impressive for the wrong reason: the founder, the principals, the late nights. Engagement two is the honesty test: can ordinary capable staff deliver materially more because engagement one improved shared infrastructure?

If successive engagements do not improve transfer and contribution, the product is consulting with software around it. Chapter 8 specifies the transfer gate; Chapter 9 specifies scarce-expert elasticity; Chapter 12 walks a candidate that fails transfer on purpose.

Spoken naturally

Find the new thing the business can sell because AI can now carry the breadth — and make sure it is valuable enough to replace the hours AI is going to commoditise.

That is stronger than “find an AI use case.” It also clarifies expertise: help knowledge-intensive service businesses discover and construct their AI-native successor offer — not generic AI strategy, not process automation theatre, not merely productising current consulting under a new logo.

What the definition excludes

It excludes copilots that leave the invoice unit unchanged. It excludes “AI strategy” retainers whose deliverable is advice about AI. It excludes internal productivity tools with no customer promise. It excludes constituted technical demos that cannot be named, bounded or transferred. It excludes high-margin projects that do not compress a threatened labour unit.

Exclusion is how a definition earns its keep. If everything counts, nothing is the prize.

Company-level proposition

Once the object is sharp, the firm-level proposition becomes sharp too: help knowledge-intensive service businesses discover their successor offer, construct the machinery that can safely deliver it, and install the foundry that produces the next one. Discovery, construction and foundry are different commercial stages (Chapter 13). Conflating them produces vague “AI transformation” programmes that never name the unit of sale.

Ask a room of service professionals “where could AI help?” and you will get a list of existing tasks. Drafting. Searching. Summarising. Routing. Classifying. The list is not worthless. It is the wrong hunting ground for a successor offer.

Existing tasks are where the current unit of sale already lives. Improving them improves the horse. The prize lives in the market’s catalogue shadow: the services nobody sells because, under human-labour economics, nobody could price them honestly.

That is the central discovery question of this book. It is not “where is AI cool.” It is “where is valuable commercial silence being enforced by expert scarcity.”

Five ways expert labour suppresses a promise

Breadth. Full coverage instead of sampling. The market sells a review of a handful of systems, accounts, or machines because reviewing the whole estate would require an army of seniors. Machine-scale cognition can read and join at breadth humans cannot staff.

Frequency. Continuous or high-cadence attention instead of annual or incident-driven attention. Always-investigated instead of escalation-only. Continuous supportability state instead of rebuild-on-breakdown.

Specificity. Customer-specific, configuration-specific, project-specific preparation as a standard product rather than an expensive exception. Under labour economics, specificity is what you charge extra for and still under-deliver.

Coordination. Many-to-many matching across manuals, histories, stock, suppliers, branches, prior cases and customer constraints — the join work seniors currently perform in their heads and inboxes.

Uncertainty. Typed, evidence-backed uncertainty as a sellable product, rather than buried assumptions inside an unreviewable statement of work. Buying a reduction in uncertainty is a different commercial object from buying a speculative build.

AI-Constituted Services maps this hunting ground as economically suppressed services — the shadow catalogue of every industry. The Cognition Scarcity Audit is the enterprise-internal sibling: fund the analysis you never do because scarcity made it rational not to. This book’s version is commercial and outward: which promise is the market silent on?

You can often nominate friction from outside

Structural friction is frequently visible without a confidential interview. Industry structure, business model, website, catalogue, public case studies and the shape of proposals in a sector all leak the same pattern: customers translating stable intent into the interactions the provider’s systems demand.

For a mid-sized data consultancy, you can often see from the outside the commercial structure that produces unpaid pre-sales cognition, multi-senior SOW authorship, and large bespoke transformation quotes that buyers distrust. For a capital-equipment distributor, you can often see a contact surface that routes hard technical reality into scarce experts and reactive parts transactions after project risk is already live.

The defensible boundary is this: outside evidence can nominate the problem shape; internal evidence must earn the word “biggest.” You can infer the equation from outside. You need internal evidence to solve for the coefficients — hours, margin, segment intensity, competing frictions. That is not a weakness in the method. It tells you what the first paid access should test.

The capability sentence

Force nominations into one disciplined format:

We could promise [bounded customer promise]
at [stable commercial unit]
if a machine carried [breadth / frequency / specificity /
                      coordination / uncertainty].

Half a dozen honest sentences beat forty brainstormed use cases. Then select: named buyer with budget and a decision they cannot currently purchase; native evidence the client already possesses; a path to compress a labour-priced unit the firm actually lives on.

What this fights

It fights the empty-page workshop: “tell me your problems and I’ll suggest AI ideas.” Organisations are often the least reliable originators of the initial diagnosis. Structural friction has been normalised as “how this industry works,” distributed across sales, delivery and finance, visible only in fragments, protected by incentives, and described in the vocabulary of the current process.

Sales hears slow approvals. Architecture hears weak requirements. Delivery hears overselling. Finance hears margin variance. Each account can be correct while none names the system manufacturing all four outcomes.

Industry structure → company expression → internal residual

A useful hierarchy keeps outside nomination honest without pretending coefficients are known:

INDUSTRY STRUCTURE
What friction normally follows from this market and business model?
        ↓
COMPANY EXPRESSION
How does this firm’s website, catalogue, organisation and public
activity appear to instantiate or resist that structure?
        ↓
INTERNAL RESIDUAL
What is genuinely different here — scale, politics, exceptions,
unit economics, hidden systems and strategic priorities?

Vertical-of-One remains correct: company-specific workflows, politics, systems and approval chains ultimately determine AI fit. It does not require beginning with an empty page. Industry structure gives priors. Company expression tests them. Internal residual earns the word “biggest.” Diagnosis still happens; the methodology of outside-in pitch selection is not this book’s organ — only the principle that the firm need not wait for clients to invent the suppressed promise for them.

Examples of silence (shape, not recipes)

In professional data and analytics services, silence often sits around full-estate certainty before large implementation: buyers want to know what is knowable, but the firm only knows how to sell investigation as days or bury uncertainty in a transformation SOW.

In capital-equipment distribution and technical aftermarket businesses, silence often sits around prepared continuity: customers want project risk reduced before the failure, but the commercial unit still activates after the failure — parts, callouts, hero diagnosis.

In account-heavy consultancies, silence often sits around decision packs compiled from history the firm already holds: the firm has the exhaust of prior work, but sells net-new analysis as if each engagement starts from zero.

None of these is a mandatory product for every firm. Each is an illustration of catalogue shadow. Your capability sentences will differ. The question form should not.

Hunt suppressed promises, not use cases.

The next chapter converts those promises into commercial units the market can buy — and a vocabulary that replaces days and hours without pretending the supplier owns the whole business outcome.

“We sell outcomes, not hours.” It sounds mature. It is often a category error wearing a premium.

True business outcomes — project completion on a construction site, realised margin in a client’s P&L, fleet uptime across weather and operator behaviour — depend on factors the supplier does not control. Sell those as unbounded guarantees and you either price yourself out of the market, destroy margin, or quietly rewrite the promise later through change requests and fine print.

The safer formulation is not hours versus outcomes. It is hours versus a bounded state, decision, deliverable or commitment.

The successor-unit vocabulary

Boards and founders need language for what replaces the labour unit. This is that vocabulary.

These are commercial units, not feature names. A verified decision is something a named buyer can act on. A maintained state is something that continues between incidents. A protected period is time under a defined commitment. An assessed estate is coverage with typed uncertainty. An enrolled fleet is a bounded set of assets under a continuity envelope. A response commitment is preparedness and pathway — not omnipotent control of every field outcome.

For a capital-equipment distributor, the unit should probably not be “uptime guaranteed.” It could be one fleet enrolled for twelve months; one project given a readiness and continuity plan; one defined set of critical stock reserved; one response commitment maintained for a particular operating period. That is commercially bounded without pretending the firm controls the entire project.

The deep transition

Old:  expertise × hours
New:  compiled expertise × customer context × bounded result

The customer is not buying fewer hours. They are buying the provider’s willingness and ability to absorb controlled complexity — and to show evidence of what is known, unknown and decided. That is why productisation can support a premium rather than a discount.

Low friction is not low price. Predictability is a premium attribute when the supplier possesses machinery capable of holding the risk.

Buy Certainty First develops that premium logic for fixed-price evidence products: transferring estimation risk to a supplier who can actually hold it is valuable; a weak supplier overprices, underprices, narrows the mirror, or recovers through change requests.

Ten layers: old model versus successor offer

Read the table as a migration map, not a moral judgment. Relationships still matter. Scarce people still matter. The successor does not abolish them; it stops treating them as the only economic architecture while an entrant builds the next layer.

Commercial legibility as a side effect that is not a side effect

Once the unit of sale is stable, the firm can distinguish account suitability, sales execution, pricing integrity, delivery efficiency, expert review load, exception density, downstream client value and reusable learning. With broad bespoke consulting, those variables are mixed together. Win rate is noisy because every proposal, client, scope and delivery method is different.

Changing the unit of sale also changes the business’s ability to understand itself. That is commercial legibility: the offer becomes a measurement instrument for the company, not merely something to sell. It implies two commercial lanes — legacy RFQs and utilisation on one side; named offer, eligibility, banded price, typed states, transfer and lifecycle evidence on the other — so the successor is not judged solely by the accounting logic of the model it is intended to replace.

Reading the vocabulary against a real conversation

Imagine a buyer who says: “We need better uptime.” If you sell hours, you staff more callouts. If you sell unbounded outcomes, you write a guarantee you cannot keep. If you sell a successor unit, you enrol a fleet, define a protected project period, and price a response commitment with explicit exclusions. The buyer’s language stays emotional; your commercial unit stays bounded.

Imagine a buyer who says: “We need a dashboard programme.” If you sell days, you estimate a build. If you sell a successor unit, you may first sell an assessed estate: what is knowable, unsupported, contradictory or missing before anyone funds the programme. The dashboard may never be purchased — and that can still be a successful paid certainty engagement.

The vocabulary is not wordplay. It changes what sales is allowed to promise, what delivery is allowed to optimise, and what finance is allowed to call revenue quality. Pricing theory has long insisted that product and price are co-designed for a segment — a free product and a high-price product are different products even when the underlying machinery looks similar.⁴ Successor units force that co-design: the bounded promise and the stable commercial unit arrive together, or neither is real.

Myth vs reality

Innovation programmes love unordered lists. Eight tests on a slide. Tick the ones that sound true. Celebrate the pilot. The list feels rigorous because it is long. It is soft because nothing depends on sequence, and because failure modes are not named. The same pattern shows up in AI portfolios that measure activity — use cases, pilots, demos — without a commercial qualification path: adoption can look high while nothing about the unit of sale changes.²

This book reorders eight qualification tests into ordered gates. Order is load-bearing. Later gates assume earlier truths. Skip the order and you will optimise architecture for a promise nobody will buy, or scale a hero product that cannot transfer, or fund a constituted demo that weakens nothing of the old unit of sale.

The eight gates at a glance

The ordering argument

Without friction and a suppressed promise, constitution builds a cathedral nobody needs to buy. Technical impressiveness is not demand. Gate 3 before Gates 1–2 produces AI theatres that pass the remove-the-AI test in a lab and fail in a pipeline.

Without constitution, commercial productisation productises non-constituted work. You get a named SKU for something humans already sold — only now with a model in the brochure. Gate 4 before Gate 3 is how “AI-powered consulting packages” become rebundled days.

Without a commercial boundary, physics and economics optimise an illegible blob. You cannot honestly price or staff a promise that has no stop state, no exclusions, and no recognition of completion. Gates 5–6 before Gate 4 produce false precision: spreadsheets about a thing sales cannot describe.

Without delivery physics, unit economics is fantasy. Contribution models that ignore stock, freight, licensed authority, latency and branch capacity will approve offers the field cannot keep. Gate 6 before Gate 5 is how finance green-lights a death march.

Without unit economics, transfer multiplies a bad deal. Making a loss-making offer easier for ordinary staff to deliver is not a strategy. Gate 7 before Gate 6 scales pain.

Without transfer, strategic migration is a story about the first three logos. Heroes can “migrate” revenue in a deck while the firm’s scarce experts remain the production system. Gate 8 before Gate 7 confuses sales narrative with institutional change.

Without strategic migration, you may have a constituted, transferable, economic product that is still not the successor — a side product that never compresses the labour unit the entrant is attacking. Migration is the replacement relation. It is last because it is the claim about the firm’s future, not the first technical hurdle.

A technically constituted service can still fail on demand, boundary, physics, economics, transfer or strategy.

Why this is not pedantry

Most AI portfolios fail by celebrating the wrong gate. Engineering celebrates constitution. Sales celebrates a named package. Finance celebrates a pilot margin that ignored scarce disposition cost. Strategy celebrates a threat narrative. Delivery celebrates a demo the founder still runs.

Ordered gates force a single conversation: where did this candidate actually fail? That question is more useful than “is it innovative?” Failure modes are design tools. They tell you whether to repair, demote, or stop.

Two honesty rules that sit across the gates

Fixed price is a strong signal, not a requirement — it appears most naturally at Gate 4 (boundary) and Gate 6 (economics), but the underlying requirement is a stable unit of value.

Outsized profitability is a selection criterion, not an intrinsic property — Gate 6 is where “impossible without AI” can still die as a business.

How teams abuse unordered tests

Watch a typical portfolio review. Someone presents eight positive adjectives: strategic, AI-native, scalable, fixed-price, high-margin, defensible, transferable, customer-loved. Nobody can name which claim has evidence. Nobody can name which claim, if false, would kill the candidate. The list becomes a mood board.

Ordered gates convert mood into sequence. You are not allowed to claim “scalable” before transfer. You are not allowed to claim “successor” before migration. You are not allowed to optimise the model architecture before the commercial boundary exists. That feels slower in week one. It is faster across a year of capital allocation.

Gate map as a single wall artefact

Put one wall (or one page) per candidate:

• Candidate name and one-sentence bounded promise.
• Eight rows: gate, status (pass / fail / conditional), one-line evidence, owner of next proof.
• A red cell is not shame; it is the work plan.
• A green cell without evidence is a lie — mark it yellow until proof exists.

That wall is more useful than a roadmap of model features. Features serve gates; gates do not serve feature backlogs.

Chapters 7 and 8 make each gate runnable. Chapter 9 meters scarce-expert elasticity as the honesty metric behind transfer. Chapter 10 turns migration into an explicit destruction protocol. Chapter 12 runs a candidate until it fails.

This chapter makes the first four gates runnable. Treat each as a protocol you can run on a real candidate next week — not as a slogan. Passing Gates 1–4 produces a sellable, constituted candidate. It does not yet produce a successor. Physics, economics, transfer and migration remain.

Gate 1 — Friction

Purpose

Establish that the candidate removes structural friction worth removing: customer attention, senior labour, margin, or trust burned by the commercial model itself — not merely a cosmetic inconvenience.

Pass criteria

• You can name who feels the friction (buyer role, staff role, both) and how often the pattern recurs.
• The friction is produced by the business model or system architecture, not only by one bad account.
• Removing it would change a material economic or trust variable — unpaid senior hours, proposal variance, emergency reconstruction, client effort at the front door — not only satisfaction with a chat interface.

Evidence you can gather

Lost proposals sampled by cause class; unbilled senior hours tagged by reconstruction, reconciliation, exception hunting; customer complaints that repeat; account histories where the same dance happens every year. Outside nomination can name the shape; internal coefficients still need measurement.

Failure mode

When it fails

Do not “AI-ify” the convenience layer and call it strategy. Tolerate it as operations spend if useful. Return to the discovery question in Chapter 4.

Gate 2 — Suppressed promise

Purpose

Establish that the candidate is a valuable promise the market does not currently make — because breadth, frequency, specificity, coordination or uncertainty would consume too much expert labour — not a rebrand of something already sold economically without AI.

Pass criteria

• You can write the capability sentence: promise + commercial unit + what the machine must carry.
• A competent human-only version is not already on the price list at a viable fee and margin.
• A named buyer has a decision they cannot currently purchase in this form.

Evidence

Catalogue gaps; services clients ask for that the firm refuses or underprices; work done only on samples or escalations; unpaid pre-sales that should have been a product; competitor silence in the same category.

Failure mode

When it fails

Reclassify as AI-enabled improvement to an existing offer. Fund accordingly. Do not claim successor status.

Gate 3 — Constitution

Purpose

Establish that machine-scale cognition is an existence condition for the promise in recognisably this form — not merely a speed or cost improvement to a service that snaps back to staffing.

Pass criteria

• Remove the AI: does the same offer still exist, only slower or more expensive? If yes, you are in AI-enabled (or at best AI-dependent economics), not succession territory as constituted.
• Remove the AI: do fixed envelope, full coverage, specificity and evidence promises collapse? Then constitution is plausible.
• There is no honest pre-AI version of this commercial product sitting on the shelf waiting to be re-staffed at any price.

The full taxonomy and remove-the-AI test live in AI-Constituted Services; this book does not re-teach their internals. What it requires is an honest classification written down, one line per candidate, that a sceptic can attack.

Failure mode

When it fails

Keep the productivity gain if real. Stop calling it the firm’s car. Search again for suppressed promises that fail the snap-back test.

Gate 4 — Commercial boundary

Purpose

Establish that the promise is named, bounded, priced, buyable and recognisable as complete — a commercial product, not a SOW-shaped blob with a brand name.

Pass criteria

• Buyer can understand: inputs, work, exit state or deliverable, residual uncertainty, customer decisions required, inclusions, exclusions, price triggers, stop states (including legitimate “build nothing” where appropriate).
• Price is configured against a stable unit of value — fixed fee, band, subscription, activation — not primarily negotiated as open-ended hours.
• Eligibility is explicit: who this is for, and who it is not for.

Evidence

A one-page commercial envelope a non-technical buyer can read; sample contract language; a sales conversation that does not require inventing scope live; comparison against the old SOW path for the same decision.

Failure mode

When it fails

Compose the commercial transaction before more architecture. Buy Certainty First’s insistence that the purchase be designed before the stack is the right instinct here.

What Gates 1–4 together establish

Real friction
  + suppressed promise
  + constitutive role for machine cognition
  + commercially legible boundary
        ↓
A candidate worth building and selling once
        ↓
Still not a successor until physics, economics,
transfer and strategic migration pass

The most common false positive in the market is a candidate that fails Gate 1 or 2 but is celebrated for a clever demo. The second most common is a candidate that passes 1–3 and fails Gate 4 — excellent technical composition sold as bespoke consulting forever. Name the failure. Repair or demote. Do not scale the story.

A short working session for Gates 1–4

Ninety minutes, one candidate, three people maximum (commercial, delivery, someone who can kill sacred cows):

1. Ten minutes: write the bounded promise and the old unit it might compress.
2. Fifteen minutes: Gate 1 — list evidence of structural friction; if only anecdotes of inconvenience, stop.
3. Fifteen minutes: Gate 2 — write the capability sentence; attack it as “already sold without AI.”
4. Twenty minutes: Gate 3 — run remove-the-AI in writing; one line classification; no brand language allowed.
5. Twenty minutes: Gate 4 — draft the one-page envelope: eligibility, inputs, outputs, exclusions, stop states, price unit.
6. Ten minutes: assign the next proof owner for the first red or yellow cell.

If the room cannot finish the envelope, you do not have a product candidate. You have a project idea. That is a useful result. It is cheaper than a six-month build that discovers the same fact later.

Gates 1–4 ask whether the candidate deserves to exist as a commercial object. Gates 5–8 ask whether the firm can keep it, afford it, transfer it, and use it to replace the unit of sale. Most false “successors” die here after a flattering pilot.

Gate 5 — Delivery physics

Purpose

Establish that the firm can actually keep the promise given physical logistics, authority, regulation, stock, latency and network capacity. Machine cognition improves the cognitive and coordination curve; it does not repeal physics.

Pass criteria

• Every commitment in the commercial boundary maps to an operational pathway that exists or is funded to exist.
• Physical scarcity (parts, freight, branch coverage, working capital) is explicit in the offer design, not wished away by the AI narrative.
• Authority is placed: who may approve which class of action; what cannot be autonomous; what regulation or safety constraint forbids.
• Latency and hours-of-cover claims match real staffing and escalation design.

Evidence

Operational maps; stock and supplier lead-time reality; branch capability matrices; legal/regulatory constraints written as offer exclusions; shadow runs against historical incidents without pretending field physics changed.

Failure mode

When it fails

Shrink the promise to what physics allows, or fund the physical substrate explicitly and price it. Do not let the AI story outrun the trucks.

Gate 6 — Unit economics

Purpose

Establish attractive contribution after real costs — not “impossible without AI” as a substitute for a business.

Pass criteria

Write the contribution shape (magnitudes are firm-specific; inventing them is forbidden):

customer willingness to pay
− machine and infrastructure cost
− human disposition cost (scarce experts only, costed honestly)
− sales and onboarding cost
− physical delivery capacity cost
− liability and commitment risk
− exceptions and failure remediation
= contribution margin (attractive?)

• Willingness to pay is anchored to customer value or risk exposure, not to “hours we used to bill.”
• Scarce disposition is not free because the founder loves the product.
• Exception rates have a measurement path; they are not assumed zero.

Failure mode

Remember the correction from Chapter 3: outsized profitability is a selection criterion, not an intrinsic property of AI productisation.

Gate 7 — Transfer

Purpose

Establish productisation that heroes cannot fake: ordinary capable staff deliver materially more of engagement two because engagement one improved shared infrastructure.

Pass criteria

• Engagement two is designed before engagement one is celebrated.
• Shared infrastructure improved: explicit boundaries, usable playbooks that change ordinary work, tests, evidence structures, pricing rules, escalation classes, training, support ownership, reusable machinery.
• Scarce-expert disposition density falls or stabilises at a designed level while quality holds — not “the same principals, more logos.”
• A hero can be absent for a meaningful period without the offer collapsing.

Five Postures already treats machine versus senior-consultant effort across successive engagements as a load-bearing metric: if the ratio does not improve, the product may simply be consulting with software around it. Chapter 9 sharpens that into scarce-expert elasticity. Chapter 12 is a full worked failure on this gate.

Failure mode

When it fails

Do not scale into the installed base. Repair transfer or demote. Demotion is success when it prevents a zombie product from consuming senior time forever.

If the same expert must reappear at the same density, the offer has a name but has not yet become a product.

Gate 8 — Strategic migration

Purpose

Establish the replacement relation: success shrinks a labour-priced unit, grows a successor unit, and migrates valuable assets — self-disintermediation with intent.

Pass criteria

• You can name the old commercial unit that becomes smaller if this succeeds.
• You can name the new unit that becomes larger.
• You can name which assets migrate (relationships, history, judgment, access, evidence pathways).
• Harvest / migrate / construct can run in parallel without pretending the old unit is already dead — but the strategic intent to compress is explicit.

Failure mode

Chapter 10 turns this gate into a board-fillable destruction protocol. Migration is last because it is a claim about the firm’s future architecture, not a feature of the first demo.

Closing the set

Constitution without transfer is not succession. Transfer without economics is not a business. Economics without physics is not keepable. Migration without the prior gates is a strategy slide. Run them in order. Write the failure. That discipline is the difference between a portfolio of stories and a firm that can actually change its unit of sale.

Interaction effects between late gates

Physics ↔ economics. Reserved stock and response capacity can make a continuity offer keepable and still destroy contribution if capital and obsolescence are not priced. Passing Gate 5 while ignoring Gate 6 produces operationally brave, financially soft products.

Economics ↔ transfer. Early packs can look profitable because scarce dispositions are under-costed. Transfer work often raises measured disposition cost before it lowers it — instrumentation reveals the true denominator. Firms that panic at that moment and stop instrumenting choose narrative over control.

Transfer ↔ migration. You cannot migrate the firm’s economic core onto a product that only three people can deliver. Migration claims without transfer are how strategy decks outrun operations. Conversely, beautiful transfer of a side product that threatens nothing of the old unit is craft without succession.

All four without early gates. A transferable, economic, physical, “migratory” chatbot still fails if there was no structural friction and no suppressed promise. Late-gate excellence does not redeem early-gate emptiness. That is why order is not optional bureaucracy.

Decision table after Gates 5–8

Every AI product claims leverage. Few measure it in a way that would survive a sceptical finance partner. Utilisation of the old pyramid is the wrong meter. Demo counts are the wrong meter. “Hours saved” on tasks that were never the unit of sale is the wrong meter.

The honest metric for a successor offer is scarce-expert elasticity.

Numerator discipline

Count only paid, bounded commercial units: readiness reviews sold, fleets enrolled, estates assessed, protected periods under commitment, project readiness packs activated. Units the customer bought as the successor object.

Do not count free pilots as volume. Do not count demos. Do not count “influenced” pipeline. Do not count every AI-assisted task inside a legacy day-rate project as if the unit of sale had changed. Inflating the numerator is how consulting with software around it looks like a product on a dashboard.

Denominator discipline

Count only genuinely scarce expert work — material dispositions that require authorised judgment under consequence. Not every human touch. Not junior assembly the machine should have absorbed. Not project management theatre. Not the hours a principal spent polishing a slide deck because the evidence surface was weak.

If the denominator secretly includes all labour, you are back to a labour productivity metric. If it secretly excludes the founder’s night work, you are lying about scarcity. The denominator is a design claim about where authority still lives.

Lineage — reference, do not re-derive

Five Postures already tracks machine effort versus senior-consultant effort across successive engagements as a product health metric. If that ratio does not improve, the “product” may simply be consulting with software around it.

Scarce-expert elasticity is that honesty sharpened for the successor offer: the numerator is the new commercial unit, not generic project revenue; the denominator is scarce dispositions, not all senior calendar time misallocated. This book does not re-derive the foundry’s full metrics system. It requires the ratio and the discipline.

What improvement looks like without invented numbers

You do not need a universal benchmark. You need direction and comparison across engagements of the same offer:

• Engagement n+1 should deliver more paid bounded units per scarce disposition than engagement n, or hold units while disposition intensity falls, without quality collapse.
• Exception classes should become typed and fewer of them should require the same principal.
• Ordinary staff should close more of the path to the bounded result.

Shape, not fabricated percentages. If the firm cannot yet measure the ratio, that is itself a finding: the offer is not instrumented enough to claim scalability.

Relationship to the transfer gate

Transfer (Gate 7) is the qualitative form: can ordinary staff carry more because infrastructure improved? Elasticity is the quantitative honesty check on the same phenomenon. A story of transfer without an improving ratio is a story. A ratio without transfer design is a spreadsheet waiting to be gamed. Use both.

If the ratio does not improve across engagements, you are still selling consulting with software around it.

What to put on the wall

For each successor candidate, define before first sale:

1. What counts as one paid bounded unit.
2. What counts as a scarce expert disposition (classes, not vibes).
3. How both will be logged per engagement.
4. What “improve” means for the next three engagements.
5. What demotion looks like if the ratio stagnates.

That wall is more useful than a utilisation target borrowed from the pyramid the successor is meant to compress.

Example instrumentation shapes (no invented magnitudes)

For a readiness-style certainty pack, the numerator might be paid packs completed in a band. The denominator might be principal disposition hours on material findings only — not junior assembly, not project administration. Track ratio pack-to-pack. Track percentage of findings disposed without principal intervention. Track whether engagement-two independence rises.

For a continuity-style enrolment, the numerator might be fleets enrolled or project activations under commitment. The denominator might be founder or senior specialist dispositions per hundred technical cases, or per project activation. Track whether ordinary staff resolve more cases correctly with fewer interruptions because shared infrastructure improved.

These are shapes. Your firm must define classes that match its authority design. The mistake is either refusing to define classes (so everything is “senior work”) or defining them so loosely that the metric always looks good.

Anti-patterns that fake elasticity

• Counting free pilots in the numerator.
• Moving scarce work off the books as “sales support” or “R&D.”
• Redefining dispositions mid-stream to protect a narrative.
• Comparing unlike units (a tiny estate pack versus a national fleet) without banding.
• Declaring victory after one heroic engagement.

Elasticity is a discipline before it is a number. Firms that will not accept demotion when the ratio stagnates will eventually accept margin collapse or brand damage instead.

Self-disintermediation sounds violent. It is not vandalism of current revenue. It is defensive clarity: if an AI-native competitor could destroy part of your business, build that competitor inside your own company first — while harvest, migrate and construct run in parallel.

Gate 8 asked whether strategic migration is real. This chapter is the protocol artefact a partnership can fill in without a consultant in the room.

Why “destroy” is the right verb

Softer verbs let the old unit off the hook. “Complement,” “enhance,” “sit alongside” are how firms fund side products that never threaten the utilisation machine. The entrant is not planning to sit alongside your day rates politely. The entrant is planning to make the labour-priced unit look absurd for the decision the customer actually needs.

Naming destruction is how you refuse that fate without pretending cashflow from the old model can be switched off on Monday. You still harvest. You still fund construction from today’s book. You refuse to confuse harvest with destiny.

Harvest, migrate, construct — altitude only

Five Postures applies the same triptych to professional services: harvest the cash-generating advisory and implementation book; migrate durable assets into firm memory and machinery; construct a separate-but-connected portfolio of AI-constituted products whose metrics are not utilisation of the old pyramid.

This book does not re-teach organisational design for that motion. It requires that any successor candidate complete the destruction test so construct is not a slogan.

Worked shapes (not production metrics)

Mid-sized data consultancy

Old unit reduced:
  unpaid discovery, proposal labour, vague SOWs, consultant-days

New unit increased:
  paid evidence-backed certainty; governed implementation options

Assets that migrate:
  account history, delivery patterns, senior judgment,
  evidence pathways, trust and procurement access

Capital-equipment distributor

Old unit reduced:
  reactive parts enquiries, emergency context reconstruction,
  ad hoc expert diagnosis, one-off repair coordination

New unit increased:
  enrolled fleet continuity, project readiness,
  reserved recovery capacity, maintained supportability

Assets that migrate:
  installed base, configuration and service history,
  founder and senior judgment, supplier network,
  branch capability, customer project relationships

Chapter 11 develops these as design specimens. Here, notice only the structure: shrink / grow / migrate. If any line is blank, Gate 8 is not passed.

What does not count as destruction

• A copilot that makes the same days slightly cheaper to deliver.
• A chatbot that deflects tier-1 tickets while the economic core remains reactive labour.
• A fixed-price package that still consumes the same senior density and the same SOW theatre after signature.
• An internal efficiency tool that never becomes a customer promise.

Useful, possibly. Successor, no.

Name what you destroy — or admit you are not building the successor.

How to run the test in a partner meeting

1. Write the candidate’s bounded promise in one sentence.
2. List the firm’s top three labour-priced units by revenue and by senior consumption.
3. Force a mapping: which of those units shrinks if this candidate wins at scale?
4. If partners refuse to name one, the candidate is a side product — fund it as such or kill the successor claim.
5. List assets that must migrate for the new unit to compound; assign owners for migration work, not only for the demo.
6. Record what will still be harvested for cash while construction runs.

That meeting is uncomfortable. It is supposed to be. Comfortable AI portfolios are how profit pools are lost politely.

Cannibalisation anxiety — answer it cleanly

Partners will ask: “Won’t this eat our billable work?” Sometimes yes. That is the point of succession with intent. The alternative is an external party eating it without leaving you the higher layer.

Three clarifying questions reduce panic to design:

• Which billable work is low-quality revenue that already destroys margin or trust?
• Which billable work is harvestable cash that should fund construction for a defined period?
• Which customer value moves upward so total firm value can rise even as a particular unit shrinks?

If the candidate only destroys high-quality, high-trust revenue without capturing a more valuable layer — certainty, preparedness, continuous context, accountability, evidence, decision quality, response capacity — then Gate 6 and Gate 8 should both be under pressure. Destruction without capture is vandalism. Capture without destruction of the threatened unit is often a side product. The successor does both with intent.

Installed base as sensor and launch channel

A service incumbent’s advantage over a cold AI startup is not model access. It is the installed base: repeated requests, abandoned proposals, common exceptions, delivery failures, unpaid senior effort, latent adjacent demand, trust and procurement access. That base is the sensing instrument for which old unit is already rotting — and the first launch channel once a successor is composed.

Destruction testing without installed-base evidence is abstract. Destruction testing with account histories is specific: “we rebuild this SOW shape repeatedly and lose margin on half the wins” is a migration target; “maybe a chatbot for finance” is not.

Doctrine chapters stayed clean so they can be lifted and reused. This chapter is where concreteness is supposed to live — still without shipping product anatomy that other organs own. You will see enough to map the gates. You will not get Continuity’s full disposition pools or the Data Readiness Review’s full commercial protocol. Those are destination systems, not this book’s surface.

Specimen A — Mid-sized data consultancy

What the old unit looks like

Customers arrive with imperfectly framed reporting or data needs. Several disciplines contribute to scope. Delivery effort is uncertain. Substantial proposals and SOWs are authored by multiple seniors. The number is hard to review because the join that produced it lives in heads. Buyers increasingly resist large, undifferentiated advisory and implementation quotes that do not match the decision they actually need to make first: what is knowable, unsupported, contradictory or missing before we fund the build?

That structure almost necessarily creates unpaid pre-sales cognition, pricing before sufficient understanding, incomparability between proposed scopes, contingency padding or underquoting, and rediscovery during delivery. The firm may measure the activity as healthy engagement. Economically it is often wrong friction manufactured by the commercial model.

Suppressed promise

Evidence-backed certainty before implementation: a bounded readiness product that makes the estate legible — with typed uncertainty — so implementation is an option purchased against observed reality rather than a speculative SOW authored as a private join algorithm.

We could promise evidence-backed certainty across a defined estate band at a configured commercial envelope if a machine carried breadth of systems, documents and history that seniors cannot re-join economically on every bid.

Light gate map

Successor unit

Assessed estate / verified decision pack — not “a smaller consulting project.” The customer buys a reduction in uncertainty and a decision-ready artefact under a configured envelope. Implementation, if any, is a separate purchase.

Kill conditions (shape)

Estates cannot be made legible enough to type uncertainty; almost every finding still requires the same principal density; buyers will not pay for certainty separate from build; exception tails destroy contribution; engagement two cannot transfer. A product that cannot fail these tests is merely an attractive story.

Specimen B — Capital-equipment distributor

What the old unit looks like

Value capture runs through equipment, parts and service transactions. Supportability knowledge — configurations, supersessions, failure shapes, project exposure — concentrates in scarce experts, including the founder. Customers experience the firm most intensely after something has already gone wrong. The contact surface and commercial unit push hard technical reality into reactive labour and component sales rather than preparedness.

A company can own trucks, warehouses and branches and still be a knowledge-intensive service business if the scarce economic core is expertise, customer-specific judgment and coordination under uncertainty.

Suppressed promise

Maintained fleet and project continuity: prepared supportability state, readiness for a defined project period, and response commitments — without pretending the distributor controls the entire project outcome.

We could promise prepared support for these machines, during this project, under these commitments, at a subscription-plus-activation envelope, if a machine carried configuration history, document breadth, matching and continuous preparation that expert labour cannot staff per fleet.

Light gate map

Successor unit

Enrolled fleet / protected period / response commitment. The governed delivery system that makes the promise keepable is a separate object — the operating system behind the continuity product. Competitors can copy a brochure line; they cannot immediately copy compiled history, judgment, network and write-back. Anatomy of that system is out of scope here; existence of the hierarchy is not.

Kill conditions (shape)

Machine and part identities unreliable; history too incomplete; almost every line still requires founder-level judgment; customers will not pay for reserved readiness; stock capital and obsolescence overwhelm contribution; response commitments operationally impossible; real project interruption exposure too low to fund the premium.

Same prize

messy private reality
  → deterministic evidence
  → compiled contextual world
  → bounded AI judgment
  → human disposition
  → evidence-backed state, decision or commitment
  → write-back

Market vocabulary changes: certainty pack versus continuity enrolment. The problem shape does not. That is how specialisation can be broad without becoming generic — Chapter 13.

What both specimens refuse

Both refuse chatbot-as-strategy. Both refuse unbounded outcome guarantees. Both refuse to treat AI as the subject of the customer conversation. The customer buys certainty or continuity; AI silently performs enough reading, reconciliation, monitoring and preparation to make those promises economical. Code preserves state and executes approved actions. Humans retain technical, commercial and safety authority.

Both also refuse the fantasy that the first offer completes the transformation. Specimen success — even production success later — is one car. The firm still needs the system that keeps the promise and the foundry that finds the next. Hierarchy is not academic; it is how you avoid declaring victory after a single SKU.

Comparison snapshot

Different industries. Same prize.

Both specimens can still fail. The next chapter is dedicated to that honesty: a candidate that passes constitution and dies on transfer.

Frameworks that only pass favourites are marketing. This chapter is deliberate falsification: a design candidate that would impress a use-case committee, pass early gates, and still fail as a successor offer. The failure mode is transfer. The firm is a mid-sized data consultancy of the kind described in Chapter 11. No production metrics are claimed; the walk is a qualification exercise with real weight.

The candidate

If your firm has a version of this candidate under another name, good. Run the gates on yours, not only on this fiction.

Gate 1 — Friction: PASS

Customers and the firm both burn. Clients invest attention in long sales dances and large documents they cannot validate. Seniors burn unpaid hours joining incomplete requests to firm capability and risk. Finance sees margin variance on “won” work that rediscovers scope. The friction is structural of the commercial model, not a single awkward account.

Evidence shape: recurring unbilled pre-sales; multi-senior proposal authorship; buyers who say the quote does not match the decision they needed. Outside observers can nominate the shape; internal measurement would still be required for “biggest,” but for qualification we have enough to pass the existence of structural friction.

Failure mode avoided: this is not a chatbot for the intranet. Gate 1 holds.

Gate 2 — Suppressed promise: PASS

The market does not currently sell full-estate evidence products at a bounded price under pure labour economics. What it sells is either free-ish discovery theatre, days of assessment that re-open during delivery, or large implementation SOWs that smuggle uncertainty into assumptions. The capability sentence is honest: a machine must carry breadth of systems, documents and history that seniors cannot re-join economically on every bid.

Failure mode avoided: this is not rebranding an already-priced assessment product that already worked without AI. Gate 2 holds.

Gate 3 — Constitution: PASS

Remove machine-scale cognition. Can the firm still offer fixed-envelope, full-coverage, typed-uncertainty intelligence across a messy estate? Not as a product. It can offer sampled consulting at day rates. It can offer a slow, expensive senior team that never productises. The constituted promise — comprehensive evidence under a configured envelope — collapses. Snap-back staffing does not restore the same commercial object; it restores the old unit.

AI-Constituted Services would classify this as a serious candidate for constitution; we do not re-run the full internal checklist here. For this book’s gates, constitution holds.

This is where most AI programmes would stop and celebrate. The remaining gates are why they should not.

Gate 4 — Commercial boundary: PASS (on paper)

The pack is named. Eligibility bands are written. Inputs are listed: systems inventory, sample workbooks, access paths, stakeholder map. Outputs are listed: finding lines, dispositions, typed unknowns, decision options including stop. Exclusions are listed: implementation, data clean-up labour, political facilitation beyond the evidence surface. Price is a fixed envelope by estate band, not an open timesheet.

A sceptical buyer can still attack the fine print. For gate purposes, the design has a real commercial boundary rather than “transformation in a box.” Gate 4 holds on paper.

Gate 5 — Delivery physics: PASS

The substrate is primarily cognitive and documentary. There is no national parts network to fail. Constraints that remain — access provisioning, security review latency, client-side data owners who do not respond — are written as eligibility and timeline assumptions, not as invisible magic. Gate 5 holds for this design class. (A distributor continuity product would be far harsher here.)

Gate 6 — Unit economics: CONDITIONAL PASS

Early contribution looks attractive on slides. Willingness to pay is real for buyers who have been burned by large SOWs. Machine and infrastructure costs appear manageable. The trap is scarce disposition: the first three paid packs are delivered by the founder and two principals who absorb exceptions as love of the product. Those hours are under-costed. Exception tails are not yet measured. Sales and onboarding costs are blended into “strategic investment.”

Gate 6 is marked conditional pass: the shape could work if disposition is metered and priced honestly, and if exception rates fall with learning. It is not a free pass. Many firms would still green-light scale here. That would be premature for a different reason.

Gate 7 — Transfer: FAIL

This is the load-bearing failure. Develop it carefully.

Engagement one — impressive for the wrong reasons

The pack ships. The buyer is satisfied. Findings are sharp. Typed uncertainty is real. A decision is made. Internally, the production system was:

• founder judgment on every hard join;
• two principals rewriting client narratives after the model drafted;
• pricing exceptions held in heads, not in rules;
• bespoke code and private prompting skill;
• undocumented exception handling;
• several late nights classified as “ mobilisation.”

The machine improved extraction, drafting and breadth of first pass. It did not become the production authority path. Humans did not merely dispose material findings; they re-performed the product.

Engagement two — ordinary staff, same density

A second account is sold on the strength of engagement one. An “ordinary capable” team is assigned: strong consultants, not the founding trio. Training is a slide deck and a walkthrough. Shared assets are a folder of examples and a prompt library.

What happens:

• Escalations to the same principals return to approximately the same density as engagement one for material findings.
• Exception classes are tribal — “ask the founder when the ERP is multi-entity” — not typed into the vessel.
• Tests are informal; quality is “the principal would have said the same.”
• Pricing rules bend case-by-case because only the principals know which estate band was fiction.
• Evidence structures exist as documents, not as gates that block incomplete work.
• When a principal is on leave, the pack slips or quietly becomes a day-rate recovery project.

Under Gate 7’s test — can ordinary capable staff deliver materially more of engagement two because engagement one improved shared infrastructure? — the answer is no. Engagement one proved capability. It did not prove a transferable product.

What transfer would have required (repair path)

Not more enthusiasm. Explicit boundaries that ordinary staff can enforce; playbooks that change their daily work rather than decorate it; tests that fail builds; evidence structures with consumers; pricing rules that do not need a principal present; escalation classes with known rates; training that transfers authority, not only information; support ownership; reusable machinery improved by write-back from engagement one. Until those exist, the correct governance move is repair or demotion — not installed-base launch.

Gate 8 — Strategic migration: INCOMPLETE

Even if sales of the pack grow, without transfer the old unit is not truly compressed firm-wide. Unpaid discovery and consultant-days still dominate wherever the heroes are not present. The new unit grows only in a narrow corridor of principal attention. Assets (judgment, patterns) migrate into private practice, not into firm machinery. Gate 8 cannot fully pass while Gate 7 fails. Migration becomes a story about logos, not a change in the firm’s economic architecture.

What honest governance does next

1. Do not scale the promise into the installed base on the strength of engagement one.
2. Instrument dispositions on the packs already sold — start the elasticity meter even if the ratio is currently poor.
3. Choose repair or demote with a date. Repair means funded transfer design, not weekend heroics. Demote means return the work to bespoke high-touch until infrastructure exists — and stop calling it the successor.
4. Keep the learning: constitution and boundary work is not wasted; transfer failure is information.
5. Refuse the narrative that “we just need more sales.” Sales without transfer is how zombie products are born.

A framework that can fail a charismatic candidate is usable. A framework that only passes favourites is a brochure.

Why this chapter is load-bearing for the book

The honesty banner in Chapter 1 promised falsifiability. This is that promise kept. The successor offer is not “anything that uses AI and has a fixed price.” It is a bounded commercial unit that survives ordered gates — including the gate heroes hate most.

If you take only one operating habit from this book, take this: design engagement two before you celebrate engagement one. Everything else in the qualification machinery supports that habit.

Contrast: what a transfer pass would have looked like

Suppose engagement one had left behind more than a folder of prompts. Explicit finding classes with disposition rights for ordinary consultants. Automated gates that refuse to compile a client pack without required evidence fields. Pricing-band rules that non-principals can apply. A short list of exception types with known escalation owners and expected rates. A write-back path that turns principal corrections into tests. Training that includes authority to stop work when evidence is incomplete — not only how to run the tool.

Engagement two would still be hard. Transfer is not magic. But the density of principal interruption would be a designed residual, not a surprise. Elasticity would have a chance to move. Migration would have a production system to land on. The candidate would still need economics and destruction honesty — but it would no longer fail the productisation test that separates a named hero service from a successor offer.

That contrast is the point of the negative case. Constitution got you into the game. Transfer decides whether you have a firm capability or a starring vehicle for three people.

How to use this specimen in your own firm

1. Pick your most celebrated AI “product.”
2. Run Gates 1–6 quickly in writing.
3. Force Gate 7 with names: who delivers engagement two, what infrastructure improved, what density of scarce disposition remains.
4. If you cannot answer without embarrassment, you have a Chapter 12 candidate.
5. Choose repair or demote with a date — then protect the date from sales enthusiasm.

The market will not do this for you. Buyers will buy heroics while they last. Entrants will productise the layer that remains when heroics run out.

Once the successor offer is defined and gated, three positioning consequences follow. Get them wrong and you will either niche into a dead end, build one product and call the firm transformed, or sell delivery method as if it were the product.

The market is knowledge-intensive service businesses

Not “software companies.” Not “anyone with a process.” Knowledge-intensive service businesses: firms whose value capture still runs through scarce expertise, customer-specific judgment, and labour-priced coordination under uncertainty.

A firm can own trucks, warehouses and national branches and still be one. The capital-equipment distributor is not a pure software story. It is a service story with physical substrate. The test is economic, not aesthetic: if AI-native entrants can compress the labour-priced unit while you optimise utilisation, you are in the market this book addresses.

Reusable specialisation is a problem shape, not an industry

“Now build one for CFOs, one for sales, and one for operations” is a useful way to imagine scale and a dangerous way to design products. Department labels invite solution-first thinking. Industry labels can do the same.

The reusable unit is the problem shape:

messy private reality
  → deterministic evidence extraction
  → compiled contextual world
  → bounded AI judgment
  → human disposition
  → evidence-backed state, decision or commitment
  → write-back

That shape can produce a Data Readiness Review, a Margin Integrity Review, an Account Growth Decision Pack, an Operational Control Readiness product, fleet continuity, project preparedness, regulatory-assurance products, engineering-estate reviews. Market-facing vocabulary changes. Underlying compiler and governance organs largely remain.

This is how you are broad without becoming generic: specialise in converting context-heavy, exception-rich, expertise-constrained service work into bounded, evidence-backed commercial offers — not “AI consulting for services businesses,” and not a single vertical forever.

The firm hierarchy: offer → system → foundry

1. The successor offer
   What the customer buys

2. The governed delivery system
   What enables the company to honour the offer repeatedly

3. The offer foundry
   What repeatedly discovers, builds, launches and improves offers

A successful continuity offer or readiness pack proves that one new commercial product can exist. It does not prove the company has become AI-native. The stronger destination is a repeatable organisational capability for discovering and constructing AI-native successor offers. The first offer is an output of that capability. The foundry is what keeps producing the next as technology, expectations and industry economics move.

The foundry as destination — one section only

Five Postures owns the foundry: six functions, sales membrane, lifecycle, separate-enough-connected-enough design, and Monday operating moves. This book will not re-teach that organ.

What it requires you to remember:

• The foundry’s production unit is the next offer, not an AI capability.
• The loop at altitude: sense → select → compose the commercial transaction → build the delivery machine → launch through the installed base → compound.
• Utilisation gravity will kill construction if metrics stay on the old pyramid.
• Isolation without installed-base access kills sensing and launch.
• The durable prize is becoming the company that can keep finding and building cars — not forever starring the first car.

The first offer is the car. The more valuable prize is becoming the company that can keep finding and building cars.

Vendor-side three-stage commercial model

If you help other firms do this work, your own proposition can stack in three stages:

1. Successor Offer Discovery — friction map; current and successor economic unit; buyer and demand evidence; commercial hypothesis; bounded promise; candidate pricing unit; kill conditions; construction recommendation.

2. Successor Offer Construction — specimen; commercial transaction; delivery operating map; governed human–AI–code placement; pricing envelope; first-client proof; operating and evidence package; engagement-two transfer design.

3. AI-Native Offer Foundry install — sensing cadence; nomination format; selection board; pricing membrane; offer lifecycle; field-pattern ledger; transfer gates; separate successor metrics; scale, repair and demotion discipline.

Offer-build engagement mechanics — how one live client problem becomes one sellable offer with a working system — live in One Problem One Offer territory; this book does not re-explain them.

FDE is the delivery method, not the thing sold

Forward-deployed engineering is how you keep contact with reality across discovery, construction and foundry install. It is outcome-focused, AI-backed for complexity and governance, and accountable to the business. It is not the SKU. Customers buy the successor offer — or the discovery/construction/foundry engagements that produce one. They do not buy “FDE” as the product name on the invoice any more than they buy “agile.”

Confusing method with product is how firms rebundle hours under a fashionable label. You already know that failure mode under other names.

You do not need a twelve-month AI strategy offsite to use this book. You need artefacts that force the firm to name the commercial object, run the gates, and refuse false successors. This chapter is the operating close.

1. Name your current unit of sale

One line. Consultant-days. Service-hours. Person-months. Reactive tickets. Bespoke SOWs. Parts-plus-labour. If partners cannot agree what the firm actually sells as its economic unit, they cannot agree what a successor would replace. Write the top two units by revenue and by senior consumption. They may differ. That difference is information.

2. Write one suppressed-promise hypothesis

We could promise [bounded customer promise]
at [stable commercial unit]
if a machine carried [breadth / frequency / specificity /
                      coordination / uncertainty].

One sentence. Not forty use cases. If you cannot write it, you are still in process-map thinking. Return to Chapter 4.

3. Run the eight gates in order

Friction → suppressed promise → constitution → commercial boundary → delivery physics → unit economics → transfer → strategic migration. For each, write pass, fail, or conditional — and the failure mode you fear most. Do not reorder to protect a favourite demo. Chapter 6 explained why order is load-bearing; Chapters 7–8 made the gates runnable.

4. Draft the destruction statement

• Old unit that shrinks if this succeeds.
• New unit that grows.
• Assets that migrate.

If partners refuse to name an old unit, you are funding a side product. Fund it as such or stop calling it the successor. Chapter 10 is the full protocol.

5. Define elasticity before celebrating demos

Numerator: what counts as one paid bounded unit. Denominator: what counts as a scarce expert disposition. How both will be logged. What “improve” means across the next three engagements. What demotion looks like if the ratio stagnates. Chapter 9.

6. Design engagement two before scaling engagement one stories

Write the transfer test into the construction plan: who delivers engagement two, what shared infrastructure must improve, which escalations should fall, which tests and pricing rules must exist without the founder. Chapter 12 showed a candidate that passed constitution and failed here. Do not scale the promise into the installed base on heroics.

7. Keep the two honesty corrections

Fixed price is a strong signal, not a requirement. Enforce a stable unit of value. Outsized profitability is a selection criterion, not an intrinsic property. Write the contribution equation even when magnitudes are still estimates of shape.

8. Locate ambition on the hierarchy

Are you funding one offer, the governed delivery system behind it, or the foundry that produces the next five? Different capital, different metrics, different organisational separation. Chapter 13. Do not measure foundry work with utilisation of the old pyramid. Do not confuse FDE method with the product sold.

9. Refuse the wrong backlog

Any AI portfolio list that cannot name the unit of sale it leaves intact — and the successor unit it proposes — is a use-case inventory. Tolerate horse optimisation as operations spend if useful. Do not let it consume strategic capital or strategic attention. The public pattern is already full of pilots that never change client delivery; you do not need to add another.²

10. Keep the evidence posture

This framework was built from two design examples that are not completed production runs at successor-offer level. Treat your own first offer the same way: instrument, transfer-test, falsify. Prefer a demoted candidate over a zombie product with a beautiful name.

A one-page board summary

If you must compress this book for a board pack, use six lines:

1. Our current unit of sale is ___.
2. The suppressed promise we will test is ___ at unit ___.
3. Gate statuses: ___ (attach the eight-row wall).
4. Destruction: shrinks ___; grows ___; migrates ___.
5. Elasticity definition and transfer plan for engagement two: ___.
6. Capital asked: offer / system / foundry — pick one primary altitude this cycle.

Any board pack that cannot fill those lines is not yet about the successor offer. It is still about use cases.

Close

Stop hunting AI use cases that decorate the horse. Hunt the suppressed promise the market will not make. Bound it. Name it. Price it against a stable unit of value. Run ordered gates that can fail charismatic candidates. Meter scarce-expert elasticity. Name what you destroy. Build the governed system that keeps the promise. Then install the capability that finds the next car — because the durable prize is not a single product launch.

Find the new thing the business can sell because AI can now carry the breadth — and make sure it is valuable enough to replace the hours AI is going to commoditise.

Scott Farrell builds AI-constituted services and practice operating systems for consultancies and operators. LeverageAI · leverageai.com.au