Can fintechs use AI agents before staff are trained?

Yes, but speed is the trap. A fintech can put agents to work reading and drafting immediately: support replies, onboarding checks, dispute summaries. The pressure is to let the agent act alone once drafts look good. That should wait until the few people who will oversee it are trained, because in a fintech those few people are the whole control.

Why is a fintech different?

Fintechs often have thin operations teams relative to their customer numbers. That is partly the business model. It means that when an agent moves from drafting to acting, there may be only two or three people able to notice a problem, and they are also doing five other jobs.

Fintechs also tend to configure before they prove. A product team can switch a capability on in an afternoon. Under a governed setup, that does not work: depth is earned, not configured. The agent's role sets a ceiling, evidence from Attest sets the depth, and the lower one wins. Training is what turns that evidence into something people can act on.

Where can agents start?

What must wait?

• Closing alerts or approving customers without a person (depth 3).
• Refunds, chargebacks and fee reversals (depth 3 and above).
• Any payment or transfer (depth 4), which should stay held for a named approver.
• Account restrictions and exits, which carry conduct and complaint risk.

Some things never reach an agent at all. An action withheld at the platform floor, such as reversing a settled payment on the firm's own rail, is refused for every agent at every depth.

What does a sensible start look like?

Take a fintech with a card product, a few hundred thousand customers and an operations team of twelve. It starts an agent drafting replies to card dispute queries. In week two, the head of operations notices the agent tells customers a dispute "has been raised" when it has only been drafted. A reviewer had approved fifteen such replies without noticing. The wording is fixed in the agent's brief, the replies are corrected, and the reviewer is coached.

The episode decides the plan. Every operations team member completes foundational training within a month. The four who approve drafts work towards operator-level certification. The head of operations will only seek depth 3 for simple dispute status replies once those four are certified and the agent has passed certification on the firm's own past disputes. A full plan sits in how fintechs should train staff before deploying AI agents, with a ready format in the AI agent training plan template for fintechs.

Which controls fit a small team?

• The firm's own limits hold the agent at depth 2 until a named person signs off more, on evidence.
• A daily sample of approved drafts, small but regular.
• A rule that nobody approves held actions in an area they have not been trained for.
• Every verdict filed, so a regulator or partner bank can see how decisions were made.
• A plan for when an agent drops to depth 1: who absorbs the work.

Fintechs make some predictable errors, such as training the product team and not operations, covered in what fintechs get wrong about AI agent training. Wealth managers face a slower version of the same decision; see whether wealth managers can use agents before staff are trained. A small team can cover the basics through a free foundational course such as the one in Pryme Intelligence Academy. The guide to training teams for AI agents covers who to train first.