Insights / Finance operations

Why your finance team isn't using the AI tools you bought

AI finance tools stall when nobody owns the workflow, the old process keeps running underneath, and no one has agreed what the AI may do alone. How to restart one.

· Winnit Team

Document shapes detouring around an idle software window, reconnected through it by a single teal line

The renewal notice for your AI invoice tool arrives. You open the usage report before you sign. There are a handful of logins, most of them from the week of the vendor's onboarding session. Your AP lead is still keying supplier invoices into the ledger from PDFs, and the tool's own inbox holds a queue of extracted invoices that nobody has opened in a month.

The usual diagnosis is skills: the team needs more training, or the tool needs a champion. Training rarely moves the number, because the tool has usually stalled for operating reasons that no training session touches. The software can read the invoices. The work around it was never redesigned.

Three reasons AI tools stall in finance

Nobody owns the workflow

The tool was chosen by one person, usually the owner or the finance lead, configured during the vendor's onboarding, and then handed to a team whose targets did not change. When the extraction gets a field wrong, fixing the rule is nobody's job. So people work around it, and after a few weeks the workaround is the process.

The old process is still running underneath

Invoices still arrive in the shared AP mailbox. Someone forwards them to the tool, waits for the extraction, then checks every field against the PDF before keying the entry into the ledger as before. The tool has added a step instead of removing three. Faced with double work at month end, people drop the extra step, and the tool quietly goes unused.

Nobody agreed what the AI may do on its own

Without a written rule for which steps need a person's approval, every output gets reviewed line by line, and the pilot never leaves suggest-only mode. We covered where those gates belong in a separate piece on the human approval step. The short version: let the AI read, match and draft, and put a named approver in front of anything that moves money, changes supplier details or speaks to a customer.

Reading the symptoms

You can usually tell which of the three stalled your tool from what the team does instead of using it.

Common symptoms of an unused AI finance tool, and where to start
What you seeLikely causeFirst fix
Logins spike after training, then fadeNo ownerName one person and put the workflow in their objectives
The team re-checks every output against the sourceNo agreed approval rulesWrite down which steps need approval and which do not
Output is exported to a spreadsheet and reworkedThe tool is not connected to where the work finishesSend the output straight into the ledger or the approval queue
Documents still arrive in the old mailbox firstOld process still runningRoute documents to the tool at the point they arrive
Two people use the tool in two different waysNo written procedureA one-page runbook for the workflow

Restart one workflow, not the whole tool

Trying to relaunch the tool across everything it can do at once is often how the first rollout failed. Pick one workflow and get it running properly before touching the next.

  1. Pick the workflow with the most volume and the clearest rules. For most small finance teams that is supplier invoices.
  2. Map it as it runs today: every step, who does it, and which system it happens in. Include the workarounds, because they show where the tool failed.
  3. Redesign so the tool removes steps. Change where documents arrive, and decide which checks the tool now does instead of a person.
  4. Name the owner, who fixes rules when the output is wrong, and the approver, who signs off the actions. They can be the same person in a small team.
  5. Write the approval rules down and put them where your auditor can find them.
  6. Run the new path alongside the old one for one close cycle, then switch the old path off.

The last step is the one most teams skip. While the old path stays open, anyone under deadline pressure will use it, and the tool never builds the track record that earns it more trust.

Measure work moved, not logins

A login count tells you people opened the tool. It does not tell you the work moved. Track three things for the workflow you restarted: the share of documents that went through the tool without being reworked outside it, how often the approver edits the AI's proposal before approving it, and the time from a document arriving to its approval. Look at them at every month end.

The edit rate matters most. Falling edits on a category of work are your evidence for loosening the approval on it. Edits that stay high point to a rule the owner needs to fix, or to a job the tool cannot do.

When the tool really is the problem

Sometimes the software is the wrong fit. It cannot connect to your accounting system, it cannot pause a workflow and wait for a person, or its extraction fails on the formats your suppliers actually send. Test this before you blame the team: run the redesigned workflow on a sample of last month's real documents. If it fails there, replace the tool. If it passes, the problem was never the tool, and buying a second one without fixing ownership will repeat the first rollout.

How we can help

At Winnit we rebuild finance, marketing and support operations so the AI runs inside the systems you already have, with a person on your team approving anything that goes out. We map the workflow as it really runs, redesign it, set the gates, and hand over the code, documentation and training at the end. We will not make an unsuitable tool work, and we do not replace your accountant. If you are paying for AI tools your team works around, that is a good place to start the conversation.

Frequently asked questions

Why do finance teams stop using AI tools after the pilot?

Usually because nobody owns the workflow, the old manual process is still running alongside the tool, and nobody has agreed which steps the AI may complete without approval. Each of these turns the tool into extra work.

Will more training fix low adoption?

Rarely on its own. Training helps once the workflow has an owner, a redesigned process and written approval rules. Without those, people go back to the old path under deadline pressure.

Which finance workflow should we restart first?

The one with the highest volume and the clearest rules. For most small finance teams that is supplier invoices, because the documents are regular and the checks are well understood.

How do we know if the tool itself is the problem?

Run the redesigned workflow on a sample of last month's real documents. If the tool cannot connect to your accounting system, cannot wait for an approval, or misreads your suppliers' formats, replace it. If it handles them, fix the operating model first.

What should we measure instead of logins?

The share of documents processed through the tool without rework, how often the approver edits the AI's proposal, and the time from a document arriving to its approval.

Winnit Team

AI operations

The team behind Winnit. We rebuild finance, marketing and support operations with AI, starting with our own.