The shape of the problem
Somebody is supposed to read all the contracts, all the correspondence, all the regulatory updates, all the reviews. In practice they read some, and things are missed.
A classification system can read everything and flag the small proportion that matters, which is exactly the shape AI is good at.
What we build these for
- Contracts and correspondence for clauses, dates and obligations
- Regulatory feeds for changes affecting your sector
- Customer messages for churn signals, complaints and escalation risk
- Reviews and social mentions for issues clustering around one product
- Internal reports for the same defect described in different words
False alarms decide whether it survives
An alert that fires weekly and is usually nothing gets ignored within a month, at which point the real one is missed too. We tune for very few false alarms even at the cost of missing marginal cases.
Start wide, review monthly — how many fired, how many mattered, what was done — and tighten on evidence.
Every alert needs an owner and an action
- Who receives it
- What they check first
- What they do if it is real
- Where the outcome is recorded, so the tuning improves
Alerts without an action attached are noise with a notification.
Cost and scale
Monitoring one source type with classification and alerting typically £15,000–£35,000. Running cost is low because classification uses small, cheap models — this is one of the more economical AI applications to operate.