Short answer. Count the tools that each do one job. When a tool covers one step that your team repeats every day, with data you already have, building your own can cost less and leaves you with something you own. But the seat price, which you can budget, then becomes a usage meter, which you can't. Put a ceiling on it before you start. And if the licence line is small, keep paying it.
The moment
The renewal email arrives. The price per user has gone up again, or a new tier appeared and your current one no longer includes what you use. This is the moment to look, because after the renewal the line is fixed for another year.
Which licences are candidates
Not every tool is worth replacing. The ones that are usually share three traits:
- One job. The tool does one thing: summarising calls, drafting replies, filling in a form, checking a document.
- Many people, every day. The seat price is multiplied by a headcount that keeps growing.
- Your own data. The input already lives in your mailbox, your CRM or your shared drive.
If a tool matches all three, building your own version in your own environment is worth a look.
Which licences you keep
- Your core systems: accounting, CRM, ERP. Those hold your records. Build on top of them, not instead of them.
- Tools a few people use now and then. The seat cost is small and the effort isn't.
- Tools you pay for because of certification, legal cover or support. You can't build those.
From a fixed price to a meter
A seat is fixed and easy to plan: price times people. What you build yourself is paid per use. Every time it reads a document or writes a draft, the AI model behind it charges for what it processes.
For most chat-style use, that bill is smaller than people expect. It changes when the work becomes automated in several steps, where the system reads, checks, looks something up and tries again. Each step sends everything gathered so far along with it. Same team, same company, a much larger bill. So the shape of the work matters more than the number of users.
Most finance teams have never had a technology line that behaves like this. That is a reason to plan for it, not a reason to avoid it.
Put a ceiling on it before you start
Four habits keep the meter under control:
- A monthly ceiling per application. When it's reached, the application stops or alerts someone. Never an open tap.
- The right model for the step. Simple steps go to a lighter, cheaper model. Only the hard ones go to the strongest model.
- Don't send the same thing twice. Instructions and reference texts that don't change can be stored and reused, which costs much less than sending them again each time.
- Read the log every month. One overview: which application, which model, what it cost.
Our own threshold
We only advise building when the yearly licence line is at least four times what it costs to build the replacement. Below that, the saving rarely pays for the effort and the extra thing to maintain. Most readers will end up with "keep the licence". That is a fine answer.
The cheapest licence is sometimes the one you keep.
Who owns it
What you build should run in your own environment, in your name. Ask whoever builds it: are the settings, the documentation and the access in my name? Can I continue with someone else later? If the answer is no, you have swapped one subscription for another.
If the answer is yes, you pay once to build it and then only for maintenance and use.
When not to do this
- The licence line is small.
- Nobody in your team has time to follow up a new application.
- The tool you would replace changes every month and you depend on those changes.
In those cases, renew and look again next year.
And yes, that includes us
What we build for you runs on the same meter, with the same ceilings. If your line doesn't clear the threshold, we'll tell you, and the advice is free: keep the licence, switch on the cost settings you already have, and look again in two quarters.