Twenty people got faster. The company learned nothing.

There is a measurement that looks good in every board pack and means almost nothing. It is the one where you survey the team, find that people are saving four or five hours a week, multiply by headcount and loaded cost, and present the result as return.

We have seen that slide in a dozen companies. In most of them, revenue per employee had not moved, cycle times had not moved, and the number of things the company knew how to do had not changed at all. The hours were real. The compounding was not.

Where the leak is

When an individual uses a tool to draft faster, the output improves and the improvement stops there. Nothing about the episode is retained in a form the organisation can use. The prompt is gone. The judgment that made the draft good is gone. The reason the customer said no is gone, unless someone typed it into a field, and even then it is a word, not a reason.

So the next quarter starts from exactly the same place, with the same institutional ignorance, and the only durable change on the balance sheet is a subscription line. Twenty people each twenty percent faster is twenty better work products and an organisation that has learned nothing.

Productivity is a rate. Capability is a stock. You cannot accumulate a rate.

This is not an argument against individual tools. They are useful and people like them, and liking them matters. It is an argument that they are a different category of purchase from the one most boards think they are approving, and should be justified differently.

The CRM rehearsal

Most companies over a certain age have already run this experiment, with a CRM, and the result is instructive because the failure was identical in shape.

A CRM records activity faithfully. Calls logged, stages moved, notes typed. Four years later it can tell you exactly what happened and it still cannot tell you why deals die, because recording is not learning. At no point does the system read across the four thousand recorded episodes and form a view. It has perfect recall and no judgment, which is the same as a filing cabinet with a search box.

The people who were disappointed by that are right to be suspicious of what is being sold to them now, because a great deal of it has the same shape: better capture, faster drafting, nothing that accumulates.

What accumulation actually requires

Three properties, and they have to hold simultaneously. Any one alone produces something that demos beautifully and dies in year two.

The episode has to survive with its context

The unit of memory is not a document, it is a case: a bounded episode with its circumstances attached. An enquiry, a visit, a quote and its revisions, who asked, what was offered, who decided, what happened next. Most of that exists somewhere in your systems and the connective tissue does not, because the connective tissue lived in the head of whoever was there.

And a case is only coherent if it is anchored to a resolved entity. If the same customer exists four times under four spellings, you do not have one history, you have four fragments that cannot be read against each other. This is why entity resolution, the least interesting work in the building, is load-bearing for everything else.

The system has to form claims, not fetch documents

Retrieval gives you the nearest thing that was written down. Judgment is what emerges across hundreds of cases: that quotes to this customer type stall at week three, that this discount shape precedes a loss, that two site visits before a quote is worth eleven points in this territory.

No individual sees those patterns, not because they are unobservant but because a career is too short and a territory too narrow. A rep with nineteen years has seen perhaps four hundred deals. The pattern needs thousands, or it needs to be looking at the right hundred with more discipline than a human applies to anything.

Crucially, a formed claim has to be arguable. Ours are written as sentences with evidence attached and an owner who accepted them. A pattern nobody can dispute is a pattern nobody will act on.

The knowledge has to expire

This is the part that almost nobody builds, and it is the one that decides whether the thing is still trusted in year three.

Every claim carries a confidence that decays with age and collapses when contradicting evidence arrives. Prices move. A competitor exits a segment. A plant closes. A buying manager retires. A system that never forgets does not stay accurate, it becomes confidently wrong, and the first time it is confidently wrong in front of the CEO, the deployment is over regardless of how accurate it is everywhere else.

How to tell which one you are being sold

One question, asked of any vendor including us: at the end of the year, what does the company know that it did not know before, and where is that knowledge held?

If the honest answer is that some individuals got quicker, you are buying productivity. That is a legitimate purchase and it should be priced like one, against hours, with a short payback and no expectation of compounding.

If the answer is that a specific set of claims about your own business now exists, is written down, carries evidence, is owned by someone, and will still be there when the person who taught it to the system has left, then you are buying capability. That compounds, and it is worth an entirely different kind of money.

Most of the market is selling the first and pricing it as the second.