Skip to main content
XentraSol — Where Innovation Meets Excellence
Finance & Fintech

Hive Tax AI

An AI tax assistant that cut research time for accounting teams

70%

Less time spent on routine research

against a measured pre-launch baseline

3x

More client queries handled per day

100%

AI answers returned with source citations

01 — Context

Where the client was

A tax advisory firm serving mid-market clients, where senior practitioner hours are the product. Every hour spent hunting through legislation is an hour not billed, and answers given to clients carry professional liability.

02 — Problem

The problem, commercial and technical

Tax professionals were spending hours per client manually searching legislation, prior filings, and internal notes, with answers scattered across PDFs and email threads.

Generic AI assistants were unusable here: an answer without a verifiable source is a liability, not a shortcut. The system had to retrieve from the firm's own document estate, cite exactly what it used, and never present a confident answer it could not evidence.

03 — Approach

What we did

  • Mapped the research workflow with senior tax practitioners before writing code
  • Built a retrieval-augmented assistant grounded in the firm's own document library
  • Added document organisation, citations, and an approval step for every AI answer
  • Hardened storage and access controls for sensitive financial records

Key decisions and trade-offs

Retrieval over fine-tuning

Legislation changes constantly. Grounding answers in a live document index keeps the assistant current without retraining, and makes every answer traceable to a source document.

Refusal as a feature

When retrieval confidence is low the assistant says so rather than generating. Practitioners trusted the tool faster because it was visibly willing to say 'not found'.

Human approval in the loop

No AI answer reaches a client without a practitioner accepting it. That single gate made the tool viable under the firm's professional indemnity requirements.

05 — What we run today

After go-live

We maintain the retrieval pipeline and evaluation suite, re-run answer-quality evals on every model or index change, and ship improvements against a monthly budget.

"The assistant became the first place our team looks. It answers with our own documents, not generic internet advice."

Product lead, Hive Tax AI

Technology

ReactNode.jsPostgreSQLVector searchAWS

Industry

Finance & Fintech

Engagement

Delivered by a dedicated senior squad. See engagement models.

Book a technical scoping call

Tell us the problem. We'll come back with a scope, timeline, and price.

A 45-minute call with an engineer, not a sales deck. You'll get honest feedback on approach, effort, and fit within one business day.