Skip to content

Reality check

AI Adoption Failure-Mode Checklist

A practical checklist for spotting the hidden risks that can derail outreach, free audits, workflow sprints and AI-assisted implementations.

Cold outreach and reputation risk

A relationship-led island market punishes sloppy outreach quickly. Before contacting a prospect, make sure there is a role-relevant reason, a recorded source, a respectful follow-up limit and a clear route to opt out.

  • Do not use automated or recorded calling for marketing.
  • Check internal do-not-contact notes before calling or following up.
  • Be clear who is calling, why the conversation may be relevant and how to stop further contact.
  • Avoid unsolicited electronic marketing to individual subscribers unless consent or a valid soft opt-in route exists.
  • Prefer warm introductions, sector events and named workflow reasons over generic AI pitches.

Free audit boundary

A free workplace audit can create liability, unpaid consulting and false confidence if it is not bounded. Treat the free step as a workflow opportunity scan, not an assurance report.

  • Keep the free scan to one workflow, one owner, one review point and one value proxy.
  • Do not request client files, credentials, live exports, personal data or confidential records.
  • Do not present legal, compliance, cyber-security or regulatory assurance unless separately scoped.
  • Route complex findings to paid diagnostic, governance work or no-fit rather than stretching the free scan.
  • Present hypotheses and next-step options, not guaranteed findings.

Agentic workflow controls

Agentic language can sound like unsupervised automation. The safer default is an AI-assisted workflow with explicit human review, data boundaries, access limits and rollback rules.

  • Start with search, drafting, classification, summarisation or evidence preparation before production action.
  • No autonomous regulated, legal, financial, eligibility, employment, safety or client-impacting decision.
  • No production write access until IT/security, data and rollback routes are agreed.
  • Record prompt-injection, output reliability, supplier, retention and overreliance risks.
  • Use scale, improve, pause or stop as the closeout decision; a stop decision can still protect the buyer.
Back to all resources

Next step

Want to apply this to a real workflow?

Book a short fit check and we will pressure-test whether AI can genuinely help, what data is safe to use, and who needs to be involved.

Book a 20-minute Fit Check