// Case study
BookKeeper AI
An internal assistant for an accounting firm that answers bookkeepers only from published, validated SOPs — and shows the source behind every answer.
Bookkeepers work across many clients, each with its own close process, platforms and exceptions. The answers live in SOPs, videos and walkthroughs of varying freshness. A generic chatbot would guess — in accounting a confident wrong answer costs more than no answer. So the product is built around trust: scope, sources, status and a human review loop.
Design process
From how bookkeepers look for answers today to a clickable prototype with three roles.
Where answers live today, who owns them and what goes wrong when they’re outdated.
Source lifecycle and certification statuses, client scoping, what the AI may and may not say.
Ask → clarify → answer with sources, plus knowledge gaps and the SME / QC review loop.
Chat, knowledge library, upload and review dashboards in one component set.
A clickable prototype with role switching and realistic data, ready for handoff to engineering.
Problems
The knowledge exists, but it’s hard to trust: scattered, aging and tied to specific clients.
SOPs go stale. Platforms and policies change, and nobody knows which guide is still valid.
Every client has its own workflow. Mixing up two clients’ SOPs is a real risk.
When there’s no documented answer, people improvise — and the gap stays invisible.
Subject-matter experts are a bottleneck: validation requests pile up with no clear priorities.
Solutions
Make every answer traceable, scoped to the right client and backed by people who own the knowledge.
Answers come only from published sources, with citations, provenance and the source’s lifecycle and certification status.
Before answering, the assistant clarifies scope: general guidance or a specific client. Another client’s SOPs can’t appear in the answer.
Compliance-sensitive answers backed by a source that needs revalidation are shown only after an explicit acknowledgment.
Unanswered questions are logged and grouped into knowledge gaps; SMEs and QC get a prioritized overview of what needs attention.
User flow
The core path from a bookkeeper’s question to a trusted answer — and what happens when trust can’t be established.
A question in plain language
General or client-specific
Published sources only
Citations · source status
Helpful · report an issue
Design system
A calm, neutral interface where color carries meaning: blue for action, status colors for the state of knowledge.
UI design
A verified answer with sources, a knowledge overview for experts, a library with source status, honest “no answer” states, upload and AI insights.
Results that matter.
Answers bookkeepers can trust: every reply traced to a validated SOP, scoped to the right client, with gaps routed to the people who own the knowledge.
Let’s talk.
Tell me about the role or project — I reply within a day.