Every business runs on knowledge nobody has written down. Which customers can be pushed and which need a call from the founder. Who actually approves a payment plan when the CFO is travelling. Which vendor's invoice always arrives twice and which chronically late payer is fine. This knowledge usually lives in one person's head, and when that person leaves, the business gets worse at money for six months.
A finance agent that does its job well will capture this knowledge. It has to. You cannot send invoice reminders well without learning that Meridian pays twelve days late every quarter and does not appreciate being reminded. The question is not whether the agent learns. It is who owns what it learns.
The default answer is the wrong one
The default in software is that learned behavior is the vendor's asset. It is the moat. It is the reason you cannot leave. Every year, the product gets a little better at your business and leaving gets a little more expensive, and the sales deck calls this 'compounding value'. From the customer's side it is called lock in.
We think that is an especially bad deal in finance. The judgment an agent captures about your customers, vendors, and approvals is not a feature of the software. It is a description of your business. It existed before the agent and it should survive the agent.
Lock in is a confession that the product is not good enough to be kept.
What 'yours' actually means
It is easy to say the customer owns their data. Most terms of service say it. Here is what we mean by it, specifically, in the product:
- Every rule the agent learns is written in plain language, with the evidence that produced it. 'Dover & Co. routes approvals through the controller; the CFO must be cc'd on second notices. Based on 6 prior cycles.' That example is illustrative.
- You can read it, edit it, and delete it. If the agent has drawn the wrong conclusion, you correct it and the correction is what it uses from then on.
- You can export all of it, at any time, in a format another system could read. Not on cancellation. Any Tuesday.
- It is stored per company, never pooled, and never used to train anything for anyone else.
Why this is good for us too
There is a commercial argument here, not just a moral one. If a customer can leave with their knowledge intact, we have to be kept for the right reason: the agent does the job better than the alternative. That is a discipline. It keeps the roadmap honest and the pricing honest, because there is no captive customer to extract from.
It also makes the product better. An agent whose reasoning is visible gets corrected faster. A customer who could see the rule 'escalate on day 30' would tell you it should be day 45 for that account. A customer who could not see it would just be quietly annoyed at a reminder that went out too early, and would not know why.
The uncomfortable part
Making the judgment layer readable means exposing when the agent is wrong. Early on, it will be wrong fairly often. It might infer an approval route from three examples when one of them was an exception. It might decide a customer is 'always late' from one bad quarter. Showing that in plain text is less flattering than a confident dashboard.
We decided we would rather be visibly wrong and corrected than invisibly wrong and trusted. Finance is not a domain where you want a black box making judgment calls, even small ones. Especially small ones, because they compound.
If a customer could not rebuild their agent's behavior from the export alone, the export is not finished.
So: the brain is yours. Not as a slogan, but as a set of specific product decisions that cost us the easiest retention tactic in software. We think it is the only honest way to sell an agent that learns your business. We also think, over time, it is the way that wins.