· 1 min read
"We should do something with AI" is now a board-level sentence in most UK SMEs. The useful question is narrower: which workflows have enough volume, tolerance for review, and clear success criteria to justify an agent?
Four patterns that earn their keep
- Inbox triage and drafting. Classifying, routing and pre-drafting responses to structured enquiries — quotes, bookings, support tier-one. Human approves, agent does the typing.
- Document extraction. Invoices, delivery notes, compliance forms into structured data. Boring, high-volume, immediately measurable.
- Internal knowledge answers. A grounded assistant over your own policies, prices and product docs — with citations, so staff can verify rather than trust.
- Marketing operations. Generating on-brand variants, scheduling, reporting and campaign handover workflows for physical-commerce businesses.
The pattern behind the patterns
Notice what's missing: fully autonomous anything. Every deployment above keeps a human on the approval step where mistakes are expensive. The ROI comes from compressing hours of preparation into seconds of review.
Starting sensibly
Pick one workflow, measure its current cost in hours, and pilot for a month. If the numbers work, expand; if they don't, you've spent a fraction of a failed "AI transformation". That's the whole strategy.