What this is for
A short procurement and due-diligence checklist for AI in contact centres: enquiry handling, appointment booking, summarisation, routing, agent assist, and customer-facing decision support.
The aim is simple. Separate useful AI from demos that look good in a controlled script and fail once they touch real customers, messy data, and regulated processes.
What it tests
- Customer impact: who is affected if the model is wrong, slow, unavailable, or over-confident.
- Human oversight: where staff can intervene, override, and record why.
- Evidence: what the vendor can prove about accuracy, bias, safety, monitoring, and incident handling.
- Operational fit: whether it works with existing channels, workforce processes, data quality, and service levels.
- Governance: how the tool feeds a register, risk tier, review gate, and board assurance view.
How to use it
Use it before procurement scoring, not after supplier selection. The useful answers are specific: evidence packs, test results, monitoring design, escalation routes, and named owners. Vague confidence is not evidence.
If you want an independent read on a vendor response or a contact-centre AI proposal, send the pack and the business problem. One paragraph is enough to start.