Evaluating AI agent providers
The market is full of AI providers making big promises. Here is what to look for and what to avoid when choosing one for your business.
Questions to ask any provider
- Can you show me a working example? Not a demo, not a slide deck. A real agent doing real work. If they cannot show you one, they have not built one.
- What happens when the agent gets something wrong? Look for human-in-the-loop controls, especially for outbound communications. If the agent can email your clients without approval, that is a risk.
- What does the pricing include? Hosting, monitoring, maintenance, bug fixes, and updates should all be included in the monthly fee. Watch for hidden per-message or per-token charges.
- What is the contract term? Monthly rolling is best. Avoid annual commitments until you have proved the value.
- Who owns the data? You should own your data. The provider should not use your data to train models or share it with third parties.
- What happens if I leave? There should be a clear exit process. Your data, your workflows, your configurations should be exportable.
Red flags
- Promises of "10x productivity" or similar vague claims
- No working examples or customer references
- Annual contracts with large upfront fees
- Per-message pricing that makes costs unpredictable
- No human approval controls for outbound messages
- Vague answers about data handling and security
Green flags
- Transparent, predictable monthly pricing
- Human-in-the-loop for external communications
- Monthly rolling contracts
- Willing to start small (one agent, one job) and scale up
- Can explain exactly what the agent does in plain language
- Based in the UK and understands UK business operations
Start small
The best way to evaluate a provider is to start with their smallest offering. An operations audit (£200 to £500) tells you whether they understand your business. A single agent (£300 to £500/mo) proves they can deliver. Scale up only after you have seen real results.