One workflow, three kinds of operator: how Onepilot experts work inside Forest

Regulated fintechs running compliance workflows on Forest keep hitting the same question: what happens the moment a case exceeds what an AI agent is allowed to decide, and no internal expert is free to catch it? Here is the answer we built with Onepilot. Three kinds of operator, an internal employee, an AI agent, and a Onepilot expert, running the same workflow under one permission model and one audit trail, so the escalation is never where governance stops.

When a compliance workflow escalates to a human, the rules must not change

A fintech runs its onboarding on Forest. AI agents pull the documents, verify identity, screen against the lists, and close the clean cases on their own. Then a case arrives that does not fit the mold: a convoluted ownership structure, a document in a rare language, an inconsistency the machine can flag but not resolve. The workflow does exactly what it should. It escalates.

The usual story stops there, as if "it goes to human review" solved the problem. It does not, and it raises a question most automation glosses over: when a person picks up that case, do they operate under the same rules as everything else in the flow, or do they step outside the system to get it done? Because if the human works off to the side, in a separate tool, on a copy of the data, with no trace, then the exception is exactly where your governed process quietly stops being governed.

Forest is built so that never happens. In a Forest workflow, an internal employee, an AI agent, and an outside expert are three kinds of operator running the same process, under one permission model and one audit trail. And when the expert who picks up the hard case comes from Onepilot, that is a partnership doing exactly what the model intends.

Forest, the agentic platform where every operator runs under one governance model

Start with what Forest automates. Forest is the operational layer where regulated workflows run on top of the providers a company already uses for KYC, screening, and monitoring. Agents call those workflows through Forest's MCP server; they triage routine cases, take the permitted actions, and escalate the rest. Every step, human or agent, is recorded at the record level, with the reasoning trace attached when an agent acts.

The point of that design is that automation is not a separate track from human work. It is the same track, with different operators on it. An agent closing a clean onboarding and an analyst handling an edge case are running the same workflow, hitting the same approval gates, writing to the same log. AI governance in Forest is not a policy bolted onto the agents; it is the same roles and permissions model that governs every human, applied identically to every agent. An agent gets a role, sees what that role can see, runs what that role can run, and clears the same two-person review. No exceptions for being a robot.

Which leaves one question: when a case exceeds what any agent is allowed to decide, and no internal expert is available to catch it, who does?

Onepilot: elastic human expertise, from customer support to KYC

Onepilot builds a new generation of customer support and operations outsourcing. Founded in 2020 in Paris, the company has raised close to $19M and works with more than 250 brands, such as Decathlon, Alan or Qonto. Its model: vetted human experts, augmented by proprietary AI, available around the clock across 35+ languages and every channel.

What sets Onepilot apart from a traditional call center is how a case gets assigned: not to whoever's free on the floor, but to an expert sourced and matched to that specific mission, based on the skills it actually requires. And they do not stop at e-commerce support. Onepilot already operates on regulated processes, including KYC, where their experts take on the steps that call for genuine human verification. Their conviction meets Forest's: the future is neither full automation nor all-human, but a loop where the machine handles volume and the expert steps in where judgment counts.

Inside a Forest workflow, a Onepilot expert works under the same permissions and audit trail

Here is where the two products meet, and why the fit is precise rather than convenient.

When a Forest workflow escalates a case beyond what an agent can decide, that escalation does not have to land in an anonymous queue and wait for whoever is in on Monday. It can be routed to a Onepilot expert, who receives the case with its full context already assembled by the workflow, makes the decision, and returns it into the flow. The agent did the volume. The expert takes the exception. The workflow carries both.

The part that matters is what governs that expert while they work. Because the Onepilot expert acts inside Forest, they operate under the same frame as any other operator: a scoped role, the permissions that role allows, a record-level entry for every action they take. The audit trail does not distinguish between a decision made by the fintech's own employee, an AI agent, or a Onepilot expert. All three are traced the same way, defensible before a regulator the same way. Permissions are enforced at the code and connection level, not written into a prompt or a service agreement, so an external expert cannot see or do more than their role permits, by construction.

Augmented operations only work when human and machine share the same rules

An augmented team works only when the human link holds under real conditions: available when the workflow summons them, and returning a decision under the same rigor as the rest. When it fails, the escalation, the one moment a case genuinely needed a person, is where the audit trail goes dark.

Forest and Onepilot close that gap from both sides. Forest provides the platform and the governance: the workflows, the permission model, the single audit trail that treats every operator alike. Onepilot provides elastic, vetted human expertise that plugs into that frame as one more governed operator, available when the workflow calls, wherever it calls from. As Onepilot's co-founder Adrien Hugon puts it, the tighter the integration between the two, the better off everyone is.

If your automated workflows stall the moment a case falls outside the frame, or if the humans who rescue them work off to the side where nothing is traced, that is exactly the loop Forest and Onepilot close. Come show us yours.