Three generative-AI use cases in production: policy chatbot, natural-language ERP reporting, automated strategy documents
An international organisation with staff across many countries wanted generative AI to speed up the three things that ate the most senior time: answering policy questions, pulling reports out of the ERP, and drafting strategic documents — all with sources, all under governance.
The challenge
Policy questions from a distributed workforce landed on a small HR and legal team who answered the same questions repeatedly, each time re-reading the policy. Finance reporting meant analysts hand-building ERP extracts for leadership every month. Strategy documents were assembled from previous versions, slide decks and scattered data — a multi-week effort each cycle.
The organisation wanted one governed approach rather than three disconnected tools, and would only go live if answers could be traced to their sources.
The approach
- Policy chatbotgrounded in every policy
- ERP reportingquestions → governed queries
- Document generationstructured drafts with sources
- Governanceone model, three use cases
- Policy assistant. Retrieval over the full policy library with permission-aware search; answers cite the policy, section and version, and the assistant declines when policies conflict or are silent — escalating to a human.
- Natural-language ERP reporting. Questions such as "spend by programme this quarter versus plan" are translated into governed queries over a curated BigQuery layer fed from the ERP — never free-form SQL on the source system — with the generated query and the figures shown together for audit.
- Strategic document generation. Templates that pull approved data, prior documents and leadership inputs into a structured first draft, with every figure and quotation linked to its source for review.
- One governance model. Shared identity, logging, evaluation sets and a review workflow across all three, so compliance signed off once rather than three times.
What we built
- Policy chatbot grounded in the complete internal policy corpus, with citations and escalation
- Natural-language reporting layer over the ERP via a governed BigQuery model
- Automated drafting of strategic documents with source-linked figures
- Shared evaluation sets, audit logging and a human-review workflow
Results
- Three use cases live with one organisation — the first generative-AI tools in production there
- Repetitive policy questions answered instantly with a citation, freeing HR and legal for exceptions
- Monthly reporting that took analysts weeks now turns around in minutes, with the query shown for audit
- Strategy document first drafts produced in hours rather than weeks, reviewed rather than written from scratch
What made it work
Curate the data layer
Natural-language reporting only works on a governed, curated model — not on raw ERP tables.
Show your working
Displaying the generated query and sources next to the answer is what made finance trust it.
Reuse governance
One identity, logging and evaluation setup made the second and third use cases far cheaper than the first.
Government and enterprise programmes are described under confidentiality agreements; client identities, figures and architectures are generalised where required. Marketing results are reported as the clients described them.