AskDost: AI Agents for Finance Teams With Human Approval

What happened

Dost, a native-AI finance automation platform, has launched AskDost, an AI business intelligence assistant built for finance and operations teams. The tool connects to a company's ERP and other financial data sources, answers questions in natural language, and executes finance workflows only after a human approves them.

Each team can run its own AI agents on its own data. Finextra, which reported the launch, presented the product as a way to give finance teams their own agents rather than one shared tool for the whole organisation.

The reporting covers the launch itself. Pricing, deployment timelines and customer availability were not part of the coverage we reviewed.

Why it matters

The significant shift is from AI as a reporting layer to AI as an operational participant. A conversational interface over ERP data lets a finance leader or operations manager ask a question and get an answer without building a report or raising a ticket. Once the same assistant can trigger workflows, the interaction moves from insight to action, which changes how finance work is sequenced and who is accountable for each step.

The approval gate is the detail to watch. Requiring human sign-off before a workflow runs keeps people in the decision loop for consequential financial actions while still removing manual steps from routine work. For organisations modernising their finance operating model, the pattern of AI proposing and people approving is likely to shape how automation is governed as much as how it is built.

The Renascence take

Most readers will focus on the conversational interface. The more consequential design choice is the scoping: separate agents running on each team's own data. That structure reflects a sound behavioural principle. People adopt tools that behave predictably inside a domain they understand, and they resist systems that reach across everything at once. The approval step does similar work by making every consequential action visible before it happens.

A customer-obsessed operator should treat the approval gate as a service-design feature, not a compliance hurdle. Define in advance which actions need sign-off, and tell approvers what they are approving and why. Then measure whether they trust the recommendations enough to act faster, because adoption depends on that trust as much as on model capability. The real question for finance leaders is not whether AI can answer their questions, but whether their teams can see enough of the reasoning to rely on the answers.