AI Strategy · Knowledge Economics

You’ve Paid to Write That Data for Ten Years. You Never Paid to Read It.

Corporate systems were built for data input, not reuse. Reading was human-priced. AI just made reads free and continuous — which means the archive you already paid to write is no longer a graveyard. It is an asset. Here is the metric that proves it.

By Scott Farrell · LeverageAI

Ask most organisations what their knowledge systems are for, and you will hear a story about capture. Upload the proposal. File the SOW. Drop the post-mortem into the project site. Log the ticket. Fill the CRM field. The software is a series of well-lit inboxes. The process metrics celebrate how much went in.

Ask what happens to that material three years later, and the honest answer is quieter. Most of it is never meaningfully read again. Not because it was worthless when written — you paid real people real money to write it — but because reading was always the expensive half of the transaction. Finding the right document, reconciling versions, translating engineering language into a bid narrative, checking whether the answer still holds: all of that stayed human-priced. So the industry did the rational thing. It perfected the write path. It never built a read path.

That choice is now obsolete. And it is obsolete in a way a CFO can score.

You’ve paid to write this data for ten years. You never paid to read it. That’s the asset sitting idle.

SharePoint did not fail. It finished its job.

SharePoint — and every enterprise content platform that grew up beside it — is often blamed for “knowledge dying.” That is the wrong indictment. These systems did exactly what the economics asked of them. Writing was mandatory: audits, handovers, compliance, project close-out. Reading was optional: only when someone had a free afternoon and already knew roughly where to look. So product roadmaps optimised check-in, metadata, permissions, and storage. Success dashboards measured adoption of the input habit.

The result is a familiar shape, not a conspiracy. Industry benchmarks still show office workers struggling to find the current version of a document, and a large majority admitting they have recreated something that already existed somewhere in the estate.¹ Knowledge workers burn hours every week waiting on colleagues or rebuilding knowledge that is already on a drive somewhere.² Large US companies have been estimated to lose tens of millions a year to inefficient knowledge sharing.³ Classic knowledge-management programmes fail at rates that would embarrass almost any other capital programme — and when they do, people stop using the tool and go back to asking a person.^4,5

That last behaviour is the tell. When the system cannot retrieve, the organisation routes the query through a human calendar.

Capture is the write path. Compilation is the read path.

I have already argued, in Capture Was Never the Bottleneck, that disciplined capture can still fail if you never compile. Capture is the write path: append every document, every Q&A, every ticket comment. Compilation is the read path: conclude what is canonical, reconcile versions, order by demand so the next person does not start from a dump. Conflating the two is why “we already have a knowledge base” so often means “we have a very expensive attic.”

I will not re-derive that machinery here. The point for this article is narrower and more financial. For thirty years, corporate IT funded the write path because writing was process exhaust you could require. The read path stayed a human job: find cost plus comprehension cost. The natural ratio was thousands of writes per meaningful read. That was not a culture problem. It was a price problem.

AI does not magically make every archive true. It changes the price of a read. Continuous re-reading, summarising, joining, and re-expressing old material in the dialect of the person who needs it this morning is no longer a multi-day project. It is a conversational turn. When reads become free and continuous, the archive re-prices. What looked like a graveyard starts to look like inventory you never put on the shelf. That is the economic move I have called archive repricing elsewhere — keep the raw bronze when deletion was the tidy answer under the old cost curve; the curve moved underneath the decision.

Crucially, you do not have to change how people write. They can keep filing into SharePoint. The missing layer is a system that does the reading — continuously, on demand, with receipts back to the source — so the write habit finally has a yield path.

Experts have been the organisation’s RAG system

Watch an RFP assemble and you will see the old read path in pure form. A bid lead opens a request. Nobody holds the whole answer. So the org does retrieval by delegation:

• “Terry — can you write the change-management chapter?”
• “Simone — can you cover deployment?”
• “Who was the PM on the Telstra work?”

Those are not strategy requests. They are search queries addressed to humans. Experts become a slow, calendar-bound, lossy retrieval layer. The answers do not compose cleanly. You get a Frankenstein document nobody wants to reread, stitched from silo voices, after three days of chasing diaries. Winning language evaporates back into the next set of personal folders. The Intelligent RFP diagnosis names that evaporation for what it is: knowledge trapped in documents instead of reusable systems.

The sharper operational claim is this: the organisation has been using its experts as a RAG system. That is a terrible use of expensive judgment. Retrieval-by-delegation burns the calendar of the people you hired for novelty, risk calls, and client trust — not for remembering where last year’s methodology paragraph lives.

A working read path flips the split. One writer holds the narrative thread. The system pulls prior deployment language, standard change principles, and the right people when a person really is the answer. Experts get consulted for judgment and novelty. The three-day chase collapses toward one conversational turn of grounded retrieval, then a shorter human review. That is not “AI writes the bid.” That is AI stopping the bid from being a scavenger hunt.

The seventeen-year edge that no schema stored

Read-path economics sound abstract until you watch a dead system pay a dividend.

I have been building a personal work wiki over a multi-decade career archive — projects, email, the early dumping-ground databases that predated modern SharePoint. Software for some of those stores no longer exists in any usable form. The text outlived the application. That is not a metaphor. It is a systems fact, and it is exactly why cheap access to old containers matters as much as cheap comprehension.

One afternoon I asked for a full certification list, including old ones. The system found a Lotus Notes 5 certification in a projects dump, then related email, then an edge I had never modelled: the Notes 5 exam had been discounted fifty percent because I already held Notes 4. No database column related LN4 to LN5 pricing. The relationship lived latently across an email body and a cert record. Seventeen years later, inside the half-life of a follow-up thought, it surfaced.

That is what “data outliving the application that wrote it” looks like in practice. Hoarding that felt irrational under the old economics — keep a dump from software that is gone — becomes vindicated when comprehension and join cost collapse. The archive was not valuable because I became sentimental. It was valuable because a read finally became cheap enough to run.

Stop measuring footfall. Measure resurrection.

If the CFO sentence is the pitch, the dashboard is the proof. Conventional KM metrics measure footfall: queries, monthly active users, page views, documents uploaded this month. Those are write-path and door-count metrics. They tell you the library is open. They do not tell you the long tail is circulating.

Name the primary read-path metric carefully. Call it the resurrection rate.

Companion metrics (the rest of the family)

Notice the inversion. Traditional dashboards measure what people put in to prove adoption. This family measures what the system pulls back out to prove the archive was worth keeping. Write-path era and read-path era, one metric family apart.

And the wiki layer is not theatre for the model alone. The same compiled map is what makes the archive usable for humans under pressure — a point I have argued separately: you built the wiki thinking it was for the AI; the humans needed it first.

A worked economic example (illustrative boardroom sketch)

Hard ROI for a specific client requires their payroll, corpus size, and instrumented logs. What follows is a worked example — directional arithmetic you can re-run with your numbers, not a claim about any one company’s measured result.

Cost already paid to write. Suppose forty knowledge workers contribute documentation and process exhaust for roughly two hours a week, at a fully loaded AUD $120 per hour, for forty-eight weeks a year, for ten years:

40 × 2 × $120 × 48 × 10 = AUD $4.61 million

That is capital already spent creating the corpus. It does not appear on next year’s opex as a new line item. It is sunk into SharePoint, email, ticketing, and proposal archives whether you activate it or not.

Value of the first wave of resurrected reads. In the first year of a working read path, suppose two hundred documents that had been dormant for three-plus years are returned to real use. If each avoided re-creation or expert chase saves about four hours of senior time at AUD $180 fully loaded:

200 × 4 × $180 = AUD $144,000

That is only the first-order labour recovery on two hundred activations. It ignores bid-cycle compression, cross-department reuse, fewer wrong-version mistakes, and first-reads of material that would never have been rediscovered by free human recall. Even on this conservative sketch, the activation layer does not need to “pay for the archive” — the archive is already paid for. It only needs to recover a thin slice of the write-path capital that has been sitting idle.

If your numbers differ, keep the structure: (fully loaded write cost of the corpus over years) versus (hours avoided × rate × resurrected documents in period). The CFO conversation is about yield on capital already deployed, not about a new content-creation budget.

What to instrument in the next ninety days

1. Pick the eligible corpus. Proposals, SOWs, architecture decisions, major incidents, client history — not the entire drive.
2. Define dormancy N. Three years is a good default. Log last human open (or last non-author open) per document.
3. Log resurfacing events. When the read path cites or recommends a document, and when a human opens it from that path, write an event: document id, timestamp, requesting role/department, citation id.
4. Ship the four numbers. Resurrection rate (with raw count), first-read count, cross-department reads, double-click events. Review monthly with the same seriousness as pipeline.
5. Pair with one workflow. RFP assembly or “who worked on X” expert-finding is enough. Do not boil the ocean. Prove one read path before you rebrand the intranet.

Pitfalls: counting bot crawls as reads; treating every open as a resurrection without a dormancy filter; optimising for query volume instead of long-tail circulation; building another upload campaign when the write path was never the bottleneck.

The pitch you can take upstairs

You do not need a philosophy of knowledge management to start this conversation. You need one sentence and one dashboard.

The sentence: you’ve paid to write this data for ten years; you never paid to read it — that’s the asset sitting idle.

The dashboard: resurrection rate with a clear numerator, denominator, and window; first-reads that expose how much of the corpus never earned a second pair of eyes; cross-department reads that prove translation is happening; double-clicks that prove trust is being checked against bronze sources.

Thirty years of tooling explained the write path brilliantly and left the read path to human calendars. AI did not invent your intellectual property. It changed the cost of reading it. The organisations that treat that shift as a capital-allocation story — not a search-plugin story — will stop funding attics and start collecting rent on what they already wrote.

If this framing matches the idle corpus on your estate, the next useful step is not another capture initiative. It is a ninety-day instrumentation of the read path on one eligible collection — and a board-ready chart that shows what returned to circulation. I work with Australian mid-market leadership teams on exactly that class of capital-allocation problem; the free readiness assessment on leverageai.com.au is a low-friction place to start if you want a structured second look at where your AI budget is actually compounding.

Scott Farrell advises Australian mid-market boards and C-suites on AI capital allocation, governance, and architecture. He writes at leverageai.com.au.