Every vendor in this market will tell you their system preserves your institutional knowledge. We have said it ourselves: when a seasoned customs manager leaves, most of what they knew walks out with them, and a tool that remembers is worth more than a tool that merely processes. That argument is here, and we still believe it.
It is an easy claim to make. It is a remarkably hard one to check.
So here is a better question for anyone buying compliance software — and we would put ourselves squarely inside it. Not “do you preserve our knowledge?” but “how much of it, and how would I know?”
The metric is starting to have a name
In April, Gartner ran a session titled SaaSpocalypse Now: This Is How Agentic AI Disrupts the $630 Billion SaaS Business Model (Gartner, April 2026). Among its arguments was that organisations should start architecting for Knowledge Retention Rate — treating the considerable cost of running these systems as something that ought to leave an appreciating asset behind, rather than evaporating each time a session ends.
We are not going to pretend an analyst validated our product. They did not, and they were not talking about customs. But the vocabulary is useful, and the underlying point is one we have been making to trade directors for a while: the question is not what the software does, it is what your organisation is left holding afterwards.
Knowledge lives in three places, and they are not equally durable
In a conversation. Someone asks a good question, gets a good answer, acts on it. Tomorrow the thread is gone. This is where most AI in this industry currently lives, and it retains nothing.
With a person. The answer becomes part of someone’s working knowledge. Better — it survives the week. But it leaves the building when they do, usually with three months’ notice and no handover of the reasoning.
With the organisation. The decision, and why it was made, is attached to something that outlives any individual: a product, a supplier, a procedure, a case. A new joiner meets it on their first day without having to know who to ask.
Most compliance teams have a great deal in the first two and much less in the third than they would guess. Retention is simply the rate at which knowledge moves up that list.
Covered — or covered because someone said so?
This is the distinction that does most of the work, and it costs nothing to start asking.
Take whatever you consider “covered” today. For each obligation, is there a document behind it — a procedure, a ruling, a written methodology someone can open? Or is it covered because someone experienced said so?
On a status report those two look identical. They are not remotely the same asset. One survives a resignation, an audit and a change of adviser. The other is a person’s memory with a tick next to it.
We have started calling the difference retention debt: the share of your compliance position that would become undefendable if one named person left. Most organisations have never counted it. When they do, two things tend to surprise them — how large it is, and how much of it sits with the same one or two people.
That is not a criticism of those people. It is usually a sign they are good at their job and have been doing it a long time. But “our compliance rests on Marc, and Marc retires in March” is a board-level fact, and almost nobody has it written down.
A number that can only go up is a sales number
If you do start measuring this, insist on one property: it must be able to fall.
Delete the procedure document that half a dozen assessments were resting on, and an honest measure of retention drops that day. A metric that only ever climbs is a marketing artefact, not an instrument. The dips are the part you can trust — and they are the part that tells you something actionable happened.
The same goes for the arithmetic. If someone shows you a single retention score, ask how it is calculated. If the answer takes more than a sentence, or blends things measured in different units, it is a number designed to be shown rather than used.
A measure is only useful if it tells you what to do next
Counting the debt is the easy half. The harder question, and the one every director asks about ten seconds later, is what do I do about it?
The answer is unglamorous and it works: name the gap, make it somebody’s piece of work, write down what is currently in one person’s head, and put it where the system can find it. What was a person’s word becomes a document with a passage you can point at. The position measurably improves, and it improves because something real happened — not because a number was rebased.
That loop is not exotic. It is what careful teams already do, in wikis and shared drives and handover notes. Two things usually go wrong. The discipline lapses under pressure, because it depends on somebody remembering. And even when it holds, nobody can compute anything from the result: a page in a wiki cannot tell you what share of your obligations it covers, nothing re-checks it when it is written, and nothing notices when it goes stale. The knowledge is captured but never counted.
The difference worth paying for is that the write-up is attached to the specific obligation it answers. Then it can be counted, it is re-checked the moment it lands, and it visibly weakens again if the document is removed.
Knowledge Retention, as a place you can open
Viictor’s first job has always been memory. What is newer is that it keeps the reasoning alongside the answer — and records where that reasoning came from: a document you uploaded and the specific passage in it, or something a colleague asserted in conversation. Both are legitimate. They are simply worth different amounts six months later, when an officer asks why you chose that code.
That record is what makes the counting possible, and we are building it into Viictor as a section of its own — because a number nobody can find changes nothing.
It starts with what we know — the part an adviser can see. Open it and you get your compliance position sorted by how well it would hold up: what is backed by a document you could hand an officer, what rests on a colleague’s assertion, and how that balance has moved since last quarter.
It extends as Viictor takes on more of the work. What we configured — the decisions that stopped being advice in a conversation and became settings in a live trade-lane. And what we observed — evidence from what actually happened, which is the only thing that turns “we say we do this” into “here is the proof we did”.
Those are the same question asked at three depths, and together they are the only honest version of a retention number we know how to build. We would rather show it to you than describe it, and it is not finished. But the record underneath it — what was decided, why, on whose authority, and from which document — is the part that has to exist first, and that is what we have been putting in place.
Four questions worth asking any vendor, including us
- What share of our obligations is covered by something you can show me, rather than something we once told you?
- When the person who told you that leaves, what happens to it?
- Can your retention number go down? What makes it go down?
- Six months from now, can you show me why a decision was made — and what it was based on?
Any supplier who answers all four well is worth taking seriously. Any who cannot answer the third is showing you a marketing number.
The compounding part
There is a commercial argument buried in this, and it is worth stating plainly. Software you pay for and get nothing back from is an expense. A colleague who remembers is an asset that grows — every question answered, every decision recorded with its reasoning, makes the next one cheaper and the organisation slightly harder to disrupt.
It is also the part you keep. Systems get replaced; what your organisation knows, and can prove it knows, should not have to be rebuilt each time.
The point was never to replace the person who knows. It is to make sure that what they know does not leave when they do.