
Guide
Fintech
Build or buy your fintech ops back-office? Why fintechs choose Forest over building in-house
Guillaume Rigal
What a fintech ops back-office really costs to build, when building in-house still wins, and a five-step plan to run Forest on your own backend.
AI coding tools made the interface of a fintech back-office cheap. The cost sits in what surrounds it: permissions on every write action, audit trails, human validation, escalation routing, vendor integrations and agent access. Forest (forest.app) provides that layer on top of a fintech’s own backend, and with Forest Everywhere, teams keep building their own interfaces on it.
Isn’t building internal tools cheap now that AI writes the code?
Yes, for the interface. A back-end team using Claude Code or Cursor generates an admin screen in an afternoon, and ops leads vibe-code review queues in Lovable. Everything a regulated team needs around that screen does not get cheaper, and maintaining it falls to the engineers who own the product roadmap.
Ops teams describe the result the same way:
Every change goes from ops to a product spec to a developer.
Workflows get built once and never improved, because touching them costs a sprint.
Each new screen recodes its own permissions and logs.
What do you actually have to build and maintain around the core?
Component | What it takes in-house | What happens when it’s missing | How Forest covers it |
|---|---|---|---|
Permissions on write actions | Rules per role, per customer segment and per amount | An extra zero turns a €60k direct debit into €600k | Action-level permissions and approval thresholds |
Audit trail | Event logging across every service | Compliance cannot show who did what, or when | Activity logs per record, filterable by author, exportable |
Human validation | Review steps and queues for risky actions | Agents either cannot write, or write unchecked | Validation steps inside workflows |
Escalation routing | Queues to finance, compliance and billing, with SLAs | Out-of-category tickets wait about 2 hours; a payout sent to the wrong recipient on Friday night waits until Monday | Workflow routing with assignees and SLAs |
Vendor integrations | One roadmap slot per KYC, fraud or payment provider | Each new vendor waits for next year’s roadmap | Connectors, MCP library and n8n node |
Every new tool and agent | An API, permissions and logs rebuilt for each one | Each new screen or agent opens a gap in the audit trail | One Gateway behind the CLI, MCP, Gateway API and Zendesk app, with the same roles and logs |
Ops autonomy | A dev ticket for every process change | Processes freeze at version one | Ops edit workflows and screens; developers review changes |
Each row is a project in its own right, and each needs an owner long after go-live.
What does that layer look like in production?
We looked at ten fintechs that run their workflows and automation on Forest in production (October 2026). Three patterns stand out:
Writes go through rules. 91% of data changes run through a Forest action, with its permissions and log, rather than a direct record edit. The permission layer carries almost all the traffic.
Four-eyes control is the default. Nine in ten approval requests forbid self-approval, and one decision in four comes from an outsourced team working in the same tool, under the same rules.
People stay in every process. Even where AI prepares or runs steps, a person handles at least a third of the steps in each process. Queues, routing and validation stay necessary long after automation arrives.
Our team already built an MCP. Why would we need more?
An MCP that reads data covers half the job: ops query customer and transaction data in natural language, and lookups work well. The moment an agent needs to block a card, cancel a payout or update a KYB status, the team needs three more things:
write permissions per role, and per agent;
a validation step for high-risk actions;
an audit log compliance can export, with each agent under its own name.
The Forest MCP Server covers that write side. People connect their AI assistant as themselves and see only what their role allows; autonomous agents run under a service account with the role you assign. Every write lands in the same activity log as an operator’s.
When does building in-house still make sense?
Building stays the right call when all three conditions hold:
Compliance is light, and no regulator will ask who changed a record.
Fewer than 20–30 people use the back-office, so hand-coded guardrails still hold.
Workflows are stable and rarely change.
Outside those conditions, the layer around the core grows with every operator, agent, vendor and tool you add.
How do fintechs roll Forest out alongside their own stack?
Step | What happens | Who |
|---|---|---|
1. Install the backend |
| One back-end developer, about 5 days for steps 1–2 |
2. Expose core actions | Start with 4–5 Forest actions, such as create account, block card or refund | Back-end developer |
3. Set roles from the CLI | Export every role’s permissions to one file, edit it, review it in a pull request, apply | Admin, alone or with Claude Code |
4. Build the first workflow | One use case end to end, such as KYB review or payment exceptions, with approvals and audit | Ops lead, with Forest |
5. Open it to each team’s tool | Support works from Zendesk, analysts from their AI assistant, agents through MCP, custom queues on the Gateway API | Ops team; developers review changes |
FAQ
How much does Forest cost compared with building in-house?
Pricing is on the Forest pricing page. To compare, estimate the engineering time to build and maintain each row of the table above, plus the cost of one incident without guardrails.
Can we keep building our own interfaces?
Yes. The Gateway API gives the tools you build in Lovable, Cursor or Claude Code your records and Forest actions. Each user signs in with their Forest account, and every write is checked against their role and logged.
Who maintains Forest after go-live?
Forest maintains the platform: permissions, audit trail, workflows, Gateway and integrations. Your team maintains its own backend and the actions it exposes.
Does Forest work with our existing compliance tools?
Yes. Forest connects to the KYC, fraud, screening and payment providers you already use, through connectors, its MCP library and the n8n node.
Read next: What is agentic ops? · Why fintechs switch to Forest from Retool, Salesforce and AI copilots
