Yr archif.

Every business has a Dave.

Dave knows which supplier actually delivers on time, rather than which one has the best numbers in the procurement system. He knows which customer needs calling rather than emailing. He knows why you don't change that apparently pointless field in the spreadsheet.

And he knows that when the machine on line three starts making that noise, you've probably got about twenty minutes. None of this is in Dave's job description.

Some of it might be written down somewhere.

Most of it isn't.

The business employs Dave. But it doesn't own what Dave knows. I've started calling the difference Language Capital.

We have always known this stuff is valuable

None of this is new.

Businesses have always worried about people leaving with knowledge. We call it experience, institutional memory, know-how, domain expertise or simply knowing how things work around here. We try to capture it.

We write procedures. We build intranets. We create knowledge bases. We hold handover meetings. We ask Dave to document what he does before he retires.

Then somebody creates a folder called Dave Handover Final. Six months later nobody can remember where it is. The problem wasn't that businesses didn't understand the value of what Dave knew. The problem was that we didn't have a particularly useful way of turning it into something the business could use.

We could store language.

We couldn't really operate with it.

That has changed.

Computers used to like numbers

For most of the computer age, if you wanted a machine to do something useful, you had to turn the world into data. Names went into fields. Sales became transactions.

Customers became records. Machines produced telemetry. Websites produced clicks.

Businesses spent decades collecting, cleaning and protecting this stuff because computers could calculate with it. And that made data valuable. Meanwhile, another asset was accumulating almost unnoticed.

Contracts -> Policies -> Emails -> Meeting notes -> Specifications -> Customer conversations -> Procedures -> Exceptions. The reasons why something is done one way rather than another.

And, most importantly, the judgement sitting in people's heads.

Most of it is language.

Until very recently, computers weren't particularly good at doing anything with it.

They could store it. Search it. Retrieve it if somebody knew what to look for.

But understanding it was largely Dave's problem.

AI changes that.

Language can now do something

Imagine Dave works in purchasing.

The procurement system knows that Jones Engineering is an approved supplier.
It knows their prices, previous orders, payment terms and probably their average delivery performance.

Dave knows something else.

We normally use Jones, but don't use them for a rush order in November because their lead times go to pieces before Christmas. That's not really data. It's judgement expressed as language. And it might be considerably more useful than the data when someone needs a component next Tuesday.

A customer service manager knows: That customer always says they need delivery Friday, but what they actually need is confirmation by Wednesday because that's when they allocate their engineers.

Someone handling returns knows: You can approve that refund, but if it's one of these three products, talk to Sarah first because there's probably a warranty claim behind it.

These sentences exist everywhere. Until now, capturing thousands of them would probably have resulted in a magnificent knowledge base that nobody read. That's why most businesses stopped trying. But an AI system can use them.

The supplier knowledge can influence an agent choosing where to place an order.

The customer knowledge can change how work is prioritised.

The warranty knowledge can interrupt a refund before the wrong thing happens.

The language has stopped being something we merely store. It can participate in the work. And that changes its value.

Information tells you what happened

There is an important distinction here.

Information tells a machine what happened.

Language Capital tells it how your business thinks.

Your sales database can tell me that you lost a customer. Dave might be able to tell me why.

Your ERP can tell me that one supplier costs 8% more than another.

Dave might know why you happily pay it. Your service system can tell me that a particular fault keeps occurring. The engineer who has been repairing it for ten years might know the three questions that tell you which component is actually failing.

That judgement is part of what makes one business better than another. Yet we have spent years treating it as something that happens to reside in the heads of the people we employ.

The balance sheet problem

If Dave leaves on Friday and the business is less capable on Monday, something of value has left the organisation. You won't see a line disappear from the balance sheet. No transaction will appear in the accounting system.

The building, machinery and customer database is still there. The software still works.

But the business knows less. I'm not suggesting accountants should start putting Dave's November supplier rule into the annual accounts. I am suggesting that our traditional idea of what constitutes a business asset is becoming incomplete.

For years we've talked about physical capital, financial capital, intellectual property and, increasingly, data. AI makes another category economically useful.

Language Capital is the accumulated knowledge, judgement, rules, relationships and ways of working expressed in the language of an organisation.

Some of it is already written down.

Much of it isn't.

The interesting question is what happens when we start treating it as something worth identifying, protecting and investing in.

This isn't about recording every conversation

There is an obvious terrible version of this idea. Record every meeting, transcribe every call, ingest every email. Throw it all into an AI system and announce that you've captured the company's knowledge.

News Flash: You haven't.

You've created an enormous pile of words. Language Capital isn't valuable because there is lots of it. It is valuable because some of those words encode judgement.

The useful sentence isn't: We discussed Jones Engineering at Tuesday's procurement meeting.

It's: Don't use Jones for urgent orders in November.

Better still: Don't use Jones for urgent orders in November because their lead times become unreliable before Christmas. Now we have a decision, a constraint and the reason behind it.

That can be challenged. It can be updated. And increasingly, it can be used by a machine.

We are asking the wrong AI question

A lot of businesses are currently asking which AI products /services they should buy. It's understandable. There are plenty of people eager to sell them one. But I think there's a more interesting question to ask first.

What does this business know that we've never bothered turning into an asset?

For a manufacturer, it might be twenty years of engineering judgement. For a solicitor, it might be how experienced partners interpret particular circumstances before giving advice.

For a farm, it might be knowing which field holds water after three days of rain, where livestock naturally gather, or which gate becomes unusable in January.

For a care company, it might be the hundreds of tiny decisions experienced coordinators make when a rota starts falling apart at seven in the morning. Those businesses don't necessarily need more information. They need to recognise the value of what they already know.

Because AI has changed something fundamental. For the first time, machines can work with much of the language that businesses previously relied on people to remember. That doesn't make Dave less valuable.

Quite the opposite. It means we've finally got a way of making what Dave knows compound.

Before next Friday

Find your Dave.

It might be Dave.

It might be Sarah in accounts, Mohammed in engineering, Helen on reception or the person who has been driving the same delivery route for fifteen years.

Don't ask them to document everything they know.

That's how you end up with another folder nobody reads.

Ask them something simpler.

What do you know about how this place really works that the computer doesn't?

Then listen carefully to the answer.

Because some of the most valuable assets in your business might currently be going home at five o'clock.