The Asset That Is Worth More Every Year

Compounding business knowledge is what happens when every question a business answers, and every mistake it corrects, is kept where the next person and the next system can use it. Almost everything else a business pays for loses value from the day it arrives. Knowledge held this way runs in the opposite direction, and the reason is a plain property of knowledge rather than optimism about AI.

The claim itself is not new on this blog. What an intelligence system is sets out the choice between a memory that compounds and one that decays, and the cost of corporate amnesia measures the decaying version in lost hours. This post is about the mechanism underneath, the single condition it depends on, and what the difference actually looks like three years in.

Why most business software loses value over time

Start with everything that runs the other way. The Office for National Statistics has to decide how long each kind of asset stays useful to measure the economy, and in its capital stock estimates it explains that while dwellings depreciate slowly over a weighted life of 50 years, "investment in software will depreciate more rapidly, having a weighted life of five years". That is an assumption about all software investment across the UK economy rather than a verdict on any particular tool, and it matches what we see in the businesses we work with.

Software does not wear out in any physical sense, so nearly all of that decline is obsolescence. The business changes and the tool stays where it was bought. Configuration captures last year's process, integrations break as the tools around them update, and the licence renews whether or not anyone still uses the feature it was bought for.

Most documents behave the same way. A process map is accurate on the day it is drawn and a little less accurate every week after, because nobody redraws it when the process moves. That is the default, and it is why a business can spend steadily on systems for a decade and still find that its most important answers live in people rather than in anything it bought.

How does business knowledge compound?

Knowledge has one property that almost nothing else a business owns shares, which is that using it does not use it up. A van driven by one person cannot be driven by another at the same time, and an hour spent by one analyst is gone. An answer can serve everyone who needs it, at once, for as long as it stays true.

The guidance the government wrote for managing its own knowledge assets, known as the Rose Book, singles this out as what separates them from most physical assets, noting that they are "often non-rival and can be easily replicated or used by others simultaneously". The Rose Book raises it largely as a problem, since the same property lets competitors benefit from an innovation without having put in the effort of creating it. Inside a business it is the whole opportunity. The work of reaching an answer is done once, and every later use of it costs nobody anything.

That is what makes compounding possible. Each answered question adds to a stock that every future question draws on, and each correction improves every later decision that reads it rather than just the one in front of you. In year one the stock is small, most questions are new, and the whole thing feels like overhead. The stock grows anyway, and the share of questions that arrive already answered grows with it.

An answer only compounds if it is written back

Non-rivalry explains why knowledge can compound. It does not explain why most businesses never see it happen.

The reason is where answers go after they are produced. A pricing exception gets worked out for one client and explained in one email, and the reason a supplier was dropped gets settled on a call. A good answer to a new starter's question sits in a direct message that nobody else will ever read. Each of those answers took real time to reach, and each was then withdrawn, used once and left somewhere the next question cannot find it.

Compound interest only works when the interest stays in the account, and business knowledge follows the same rule. The answer that goes back into the system, with the case it came from and the date, has been reinvested, and the next person who needs it will find it there. The answer that goes into an email has been withdrawn, and somebody will spend the time working it out again.

This is a different problem from keeping what is already there current, which comes down to dates, sources and an owner, and is covered in how answers stay current inside a working system. It is also separate from the way a single correction changes what a system does next time, which systems that remember walks through. Write-back comes first. Keeping a system current protects what is already in it, and corrections refine it, but only write-back makes it bigger.

We see this in our own writing. The checklist this post was drafted against has grown one check at a time, and nearly every check exists because a particular earlier post got something wrong. Each fix was made once, and every post since has inherited it without anyone having to remember, which is the whole mechanism in miniature.

What an intelligence system looks like after three years

Year one is largely capture. Most questions are new, the system can answer only a minority of them, and it is easy to conclude it is not worth the effort, particularly in the first few months when it is useful but has not yet changed how anyone works. The comparison that matters is not month three against month zero. It is year three against year one.

What changes	Year one	Year three
Questions that arrive already answered	A minority, since most are new	Most of them, with the source attached
A new starter's first month	Rediscovering how things work from whoever has time	Starting from what the business has already worked out
What reaches senior people	Old questions in new wording	Questions that are genuinely new
What a correction changes	The case in front of you	Every later case that reads it
How the rules read	Simple, and overridden often	Conditional, with the exceptions written in

Two things stand out. The first is that the benefit arrives as speed rather than as a line item. The business starts each piece of work further along than it did the year before, because less of every task goes on reconstructing what was already known. The second is where that time comes back. The people whose hours are scarcest stop spending them on questions settled years ago, which is the shift the maturity ladder describes as systems moving from doing the work to remembering it.

Our client work shows the same shape inside a single workflow. When we built lead intelligence for a global biometrics leader, the knowledge of which rep should handle which enquiry, by geography and by product expertise, was written down once as routing rules the system applies to every lead. Before that, the triage was worked out again by hand for each enquiry, and the cost of scaling rose in step with lead volume. Afterwards it was flat. The case study puts the consequence plainly. "Adding the ninth rep or a new product line is a configuration change, not a rebuild." Every lead since has drawn on the same written-down judgement, and each new rep joins a system that already knows how the work divides.

Is an intelligence system an asset?

In the sense that matters to the people running the business, yes, and the definition the government uses for its own knowledge assets is broader than most people expect. Most discussion of intangible assets in a mid-market business stops at the brand, the client list and anything with a trademark. The Rose Book's annex on asset types goes further. "Know-how refers to practical knowledge about how to do something." It then extends the category to things entirely separate from intellectual property rights, "such as the competitive advantage inherent in having a more efficient business model than a competitor". Knowledge as an asset, in other words, is a working definition in government guidance rather than a consultant's metaphor.

In the accounting sense it is a stranger thing. What a buyer asks about what you know covers why none of it appears on a balance sheet, and that is only half the oddity. The other half is that you could not sell it on separately even if you wanted to. Jonathan Haskel, then an external member of the Bank of England's Monetary Policy Committee, drew this distinction in a 2020 lecture on the intangible economy. Comparing a firm that borrows to put up a building with one that borrows to develop software, he pointed out that "buildings are readily traded and can be sold on, software development is a sunk cost with little or no independent market".

An intelligence system shares that property, and the property explains what kind of asset it is. Its value exists only in use, inside the business that built it, and it grows every year the business keeps using it. The reasoning is the asset. The system is what makes that reasoning belong to the business rather than to whichever person currently holds it.

Why do some knowledge systems stop getting better?

The failure that catches businesses out is usually a system that stays accurate and stops growing, rather than one that decays.

It happens quietly. People consult the system but settle new questions in conversation. Corrections get made in a meeting rather than in the source, and exceptions are handled from memory by whoever has been there longest. The system stays current, because someone looks after what is already in it, and it answers exactly what it answered at launch and nothing more. It has turned into a reference library, accurate and static, which depreciates like everything else.

The sign to look for is the question that had to go to a person. Every one of those is an answer the system could have held and did not, and if nobody writes it back, the business has effectively chosen to work it out again when the next person asks.

There is a second way compounding stalls, and it is harder to see. If the corrections your team makes live inside a supplier's tool, in chat histories and settings you cannot export, the compounding still happens, but in someone else's account. The stock grows, and it is not yours.

Practical takeaways

1. Write back every answer a person had to give. Any question that reached a colleague is an answer the system should now hold, with the case and the date attached.
2. Make corrections in the source. A correction made in a meeting improves one decision. The same correction made in the system improves every decision after it.
3. Judge it at year three, not month three. Early on most questions are new and the system looks like overhead. The measure is how many questions arrive already answered, and that only grows if you keep feeding it.
4. Keep the stock where you can reach it. Corrections held in a supplier's tool compound for the supplier. Hold them in formats and accounts the business controls.
5. Measure it in hours. The capacity calculator gives an hours-back estimate for the work that currently goes on reconstructing what the business already knows.

Frequently asked questions

How does business knowledge compound?

Through reuse and write-back working together. Knowledge is not used up by being used, so an answer that took an afternoon to work out can serve every later person who needs it without costing anyone that afternoon again. That property only makes compounding possible. It actually happens when each new answer, exception and correction is written back somewhere the next question will find it, so the stock of answered questions grows and each correction improves every later use. Leave the answer in an email or on a call and it is used once and lost, which is how most businesses work by default.

Is an intelligence system an asset?

The knowledge in it is, and the guidance the government uses to manage its own knowledge assets counts know-how and business processes alongside inventions and data. The system is what makes that knowledge belong to the business rather than to whoever happens to hold it. It is not an asset in the sense of something you could sell on separately, because its value exists only in use inside the business that built it, and it does not appear on the balance sheet. That is also why it compounds for the business that holds it and for nobody else.

What does an intelligence system look like after three years?

If it has been fed properly, most of the questions it receives have been asked before, so the answer is already there with its source attached. New starters begin from what the business has already worked out instead of rediscovering it, and the questions that reach senior people are genuinely new ones rather than the fifth version of an old one. The rules it holds have become conditional, because three years of corrections have written the exceptions in. Very little of that is visible in the first few months, which is why it is so easy to judge too early.

Why does most business software lose value over time?

Because the business moves and the software stays where it was bought. In its capital stock estimates, the Office for National Statistics gives investment in software a weighted life of five years, against 50 for dwellings, and since software does not wear out in any physical sense, most of that decline is obsolescence. Configuration captures last year's process, integrations break as the tools around them change, and the licence renews whether or not the tool still fits. Knowledge that is written down and maintained runs the other way, because each year of use adds to it rather than wearing it down.

Why do some knowledge systems stop getting better?

Because answers are being taken out of them and nothing new is going back in. A system can be perfectly current and still stop improving, when people consult it but settle new questions in conversation, make corrections in meetings rather than in the source, and handle exceptions from memory. It ends up as an accurate reference library frozen at the size it was at launch. The fix is a habit rather than a project, in which any question that had to go to a person gets its answer written back so the next person finds it there.