Use cases
The problems we are usually brought in for
Recurring engagement shapes — each with the root problem, what we build, and how success is measured.
Internal knowledge01
Answers from manuals and contracts
- Problem
- People open hundreds of pages to find one clause, and two people answer differently.
- What we build
- Retrieval-augmented search that cites the source document for every answer, with role-based access.
- How it is measured
- Accuracy against an internal test set + average time to find an answer.
Customer support02
Triage and summarise incoming requests
- Problem
- Requests arrive by email, chat and forms, making priority hard to judge.
- What we build
- Classification, summarisation and draft replies that a human approves before sending.
- How it is measured
- Category accuracy + first-response time.
Planning & inventory01
Forecast demand ahead of time
- Problem
- Chronic overstock or stockouts because planning uses a flat historical average.
- What we build
- Weekly forecasting that accounts for seasonality, promotions and external factors.
- How it is measured
- Forecast error (MAPE) against the method currently in use.
Quality inspection03
Computer vision on the line
- Problem
- Manual inspection is slow and inconsistent between shifts.
- What we build
- A mobile app that captures an image, checks it automatically and queues unclear cases for a human.
- How it is measured
- Defect detection rate + false-alarm rate.
Finance & accounting01
Invoices and receipts into the ledger
- Problem
- The finance team keys document data by hand every month, and formats vary enough to cause mistakes.
- What we build
- Field extraction from PDFs and scans, cross-checked against purchase orders, with only the mismatches queued for a human.
- How it is measured
- Field-level accuracy + time per document + number of items requiring correction.
Sales02
Proposal drafts from past documents
- Problem
- Proposals take too long because material is scattered and written from scratch each time.
- What we build
- Retrieval across past cases and documents, then a draft with citations that the sales team edits.
- How it is measured
- Time to first draft + share of content the team has to rewrite.
Meetings & internal ops02
Meeting summaries that become tasks
- Problem
- Decisions made in meetings are never captured as work, so follow-through slips.
- What we build
- Key points summarised and action items extracted with owner and due date, pushed into the tools already in use.
- How it is measured
- Tasks closed on time + time spent writing summaries.
Help centre02
Semantic search on the website
- Problem
- Customers cannot find answers themselves, so every question becomes a ticket.
- What we build
- Semantic search that matches real questions, returning a short answer plus a link to the source document.
- How it is measured
- Self-service resolution rate + change in ticket volume.
These are example scopes of work, not results attributed to any client, and no outcome numbers are invented. Named case studies are published only with the client’s permission.