Governed natural-language analytics for a finance team

A reference architecture for plain-language questions over finance data via a semantic layer, with visible SQL.

Illustrative scenario. Not a completed client engagement. It describes how we would approach the problem; no client, results or quotes are implied.

/ 01Problem

The situation

Budget holders send ad-hoc questions to a small finance analytics team, creating a backlog. Earlier experiments with text-to-SQL over raw tables produced inconsistent revenue figures.

/ 02Baseline

Measuring before building

A log of analyst requests over two months is categorised to find the most frequent question types and their turnaround time.

/ 03AI architecture

Architecture snapshot

Data

Curated finance marts
Metric definitions approved by finance
Row-level security by cost centre

Understanding

Map question to metrics, dimensions, time range
Clarifying question when ambiguous

Query

SQL generated against curated views only
Static validation and cost limits
Execution under user identity

Answer

Table and chart
Generated narrative marked as such
Visible SQL and definitions

Assurance

Golden question suite
Analyst review queue for flagged answers

Illustrative reference architecture.

/ 04Delivery

How it would be delivered

1
Metric modelling
Finance and data engineers agree definitions for the top metrics.
2
Copilot build
Question mapping and SQL generation tuned against golden questions.
3
Controlled rollout
Analysts use it first, then a group of budget holders.

/ 05Outcome measures

What would be measured

Accuracy on golden questions
Share of requests self-served
Analyst backlog and turnaround

Limitations

Questions outside the modelled metrics are refused rather than guessed.
Requires ongoing ownership of metric definitions.

No results are claimed for this scenario. Targets would be set from the measured baseline in a real engagement.

/ 06Client quote

Illustrative scenarios do not carry client quotes.

Want this architecture for your workflow?

We will adapt it to your systems and constraints and give you a phased, costed plan.

Get a proposal Book a consultation