Vendors Now Embed AI Engineers in Finance Teams. Where Does That Leave a Mid-Size Team?

Two finance software vendors now send engineers into finance teams to build AI workflows. Ramp offers it to select enterprise customers. Datarails offers its existing customers 25 hours a quarter with a finance engineer. The model works because the context lives inside the team. A 6-person team can get it too, without tying the work to one platform.

What Are Ramp and Datarails Actually Offering?

Engineers who sit with the finance team and build. Both launched in 2026.

Ramp says its engineers "design, build, test, and deploy agents that operate inside existing finance workflows" (Ramp press release). Its approach starts with structured discovery of operations, systems and pain points, then builds on top of the Ramp platform (Ramp). Among the controls, the release names approval paths, audit trails and human review for high-risk decisions.

Datarails runs its package in four phases: Discover, Build, Deploy, Evolve. The engineer builds custom AI workflows directly in the customer's Datarails FinanceOS environment (Datarails release).

Why Are Vendors Sending People Instead of Features?

Our read: because features alone didn't land. Gartner surveyed 183 CFOs in June 2025: 84% of finance organizations have implemented or plan to implement AI. Only 7% report a high or very high impact (Gartner).

The Datarails release cites FERF research: only 15% of organizations consider themselves well or fully prepared to support advanced analytics and AI. Datarails' CEO puts the reason bluntly: "You cannot parachute a generalist engineer into finance and expect trustworthy output."

We agree with him. The value isn't the model. It's knowing which vendor bills in arrears, which accrual is judgment and which variance the board will ask about. That knowledge sits in your team. Someone has to go and get it.

What Does a Mid-Size Finance Team Get Instead?

Unless you're an enterprise account or already on a platform that offers this, often much less. In our view, the usual package is a login, an onboarding call and a help center. Sometimes an implementation partner for go-live, who moves on when the project closes. Maybe a ChatGPT or Copilot seat.

Then the controller becomes the integrator, at night, after the close.

Hiring the skill in is hard too. In Personiv's 2026 survey of 203 finance leaders, 43% named senior accountants the hardest role to fill (Thomson Reuters). A senior accountant who can also build reliable AI workflows is rarer still.

Your Options, Compared

The table is our view of each option, not vendor data. Two questions sort it. Will you still be on that platform in three years? And who runs the workflow on day three of every close, after the build is done?

What 25 Hours a Quarter Buys

Figures are illustrative.

Twenty-five hours a quarter is about two hours a week. Here's what one workflow, a three-entity bank reconciliation, might take to stand up. The split is ours, not Datarails'.

That's one workflow a quarter, which is a fair pace. But someone still has to run it every month, review the exceptions and update the rules when a new bank account opens. The build is the smaller half of the work.

How to Get the Embedded Model Without an Enterprise Contract

1. Pick one workflow. The one with the most hours or the most rework. Not "AI in finance."
2. Map it with the person who does it. Write down the thresholds and workarounds, not only the tasks.
3. Put the numbers in code. Spreadsheet formulas or scripts, tied to the trial balance. Never ask a chat model to do the arithmetic.
4. Use AI only to read and draft. Memos, emails, variance explanations. Every figure it cites comes from step 3.
5. Run it on a closed period first. Compare line by line with what your team booked.
6. Name who runs it monthly. A workflow with no owner decays quickly.
7. Keep a correction log. Every fix becomes a rule, so the next month is cleaner.

What to Automate vs Keep With a Person

Automate (calculate or draft):

• Matching, tie-outs and schedule roll-forwards
• Applying written thresholds and timing rules
• First drafts of commentary and vendor emails, citing the data
• Flagging what doesn't fit a rule

Keep with a person:

• Choosing which workflow goes first
• Judgment calls: estimates, disputes, write-offs
• Approving rule changes
• Sign-off on every number that leaves finance

How Flux Runs This

We work as an embedded close team for finance teams of 3 to 25. We aren't tied to a platform, and we don't stop at the build.

We embed and learn how your month-end actually runs, including the steps nobody wrote down. Then we start with one workstream, usually the one with the most hours in it.

Numbers are calculated in code and tied to your ledger before any words are written. AI is used only where judgment or drafting is needed, never to do the math. Anything we can't see in the data becomes a flagged question with an owner, not a guess.

Your team reviews and signs off. Every correction is kept, so the next month is faster. Then we run that workstream every close period, and add the next piece when the first one is steady.

We work from files you choose to export from the tools you already have. There's no system access, and a mutual NDA is signed before any files are shared. The work gets cheaper for us to run each month. You see that as a shorter close.

FAQ

What is a forward deployed engineer in finance? An engineer who works inside the finance team to build AI workflows on real data, instead of shipping a generic feature.

Can a mid-size company get Ramp's Applied AI Solutions? Ramp says it is available to select enterprise customers. It doesn't publish a size threshold, so ask Ramp directly.

Is 25 hours a quarter enough to implement AI in finance? It's enough to build about one workflow a quarter. Someone still has to run and maintain it every month.

Should we hire an AI-skilled finance analyst instead? If you can find one. The risk is the same as any single hire: the workflow lives with one person.