The playbook
- 1
Mistake 1 — Picking the wrong use case
Building for a low-volume, judgement-heavy task. Fix: only automate processes with clear rules and 5+ hours/week of usage.
- 2
Mistake 2 — Skipping the baseline
Deploying without measuring current performance. Fix: capture baseline metrics in week 1 — accuracy, time, cost.
- 3
Mistake 3 — No human-in-loop period
Going fully autonomous on day one. Fix: run shadow mode for 1–2 weeks, document edge cases.
- 4
Mistake 4 — Over-engineering the first build
Trying to handle every edge case at launch. Fix: solve the 80% case first, layer on edge cases later.
- 5
Mistake 5 — No named owner
Diffuse ownership = drift. Fix: one named exec accountable for outcomes, monthly review.
- 6
Mistake 6 — Hiding it from the team
Surprise launches kill trust. Fix: involve affected staff from day 1, frame as augmentation.
- 7
Mistake 7 — Buying a SaaS chatbot
Generic tools don't beat custom for SME use cases. Fix: build for your data, your tone, your workflows.
- 8
Mistake 8 — No post-launch governance
AI drifts without monitoring. Fix: monthly accuracy review, quarterly retraining cadence.
What you walk away with
Frequently asked
What's the most expensive mistake?+
Picking the wrong use case. Easily £30k+ wasted on a build that never delivers ROI.