- A good agency starts with your process and numbers, not with a tool or a demo.
- Insist on owning the accounts, workflows, prompts and data from day one.
- Ask how they test, monitor and hand over, not just how they build.
- Narrow, measurable first projects succeed far more often than broad "AI transformation" plans.
- Red flags: guaranteed results, vague pricing, no human escalation and lock-in to their accounts.
AI automation is now one of the most crowded services online. Some agencies build reliable systems that pay for themselves; others resell a template and a slick demo. If you are about to hire one, the questions you ask before signing matter more than the tools they use. Here are the twelve we would ask in your position.
Why choosing well matters
The failure rate is real. An MIT study found most generative-AI pilots produced no measurable P&L return (MIT NANDA, The GenAI Divide, 2025), and Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027 because of cost, unclear value or weak risk controls (Gartner press release, 2025). The difference is rarely the model. It is scope, integration and follow-through, which is exactly what you are hiring an agency for.
Scope and fit
- What problem will this solve, and how will we measure it? Expect a specific answer, such as missed calls answered, response time, hours of data entry removed or bookings per week, not "efficiency".
- What would you not automate here? A good partner pushes back on features that add cost without value.
- Can we start with one narrow workflow? The best first projects are small enough to ship in weeks and prove value quickly.
Ownership and data
- Who owns the accounts, workflows, prompts and data? The answer should be you. Systems should live in your accounts (your n8n or Make workspace, your CRM, your AI provider keys), or move there at handover.
- Where does our data go, and is it used for training? They should name every provider in the chain and confirm business terms that exclude training on your data.
- What permissions will the AI have? Look for least-privilege access and human approval for anything high-impact. Our AI agent security checklist is a useful reference.
Build quality
- How do you test before launch? Ask for test scenarios using your real calls, chats, emails or documents, including the awkward edge cases.
- What happens when the AI is unsure or wrong? There should be a clear fallback to a person, and logs showing what it did.
- Can we see similar work? Live demos you can try, case studies with specifics, or a reference you can speak to. See our case studies for the kind of detail to expect.
Cost and support
- What exactly is in the price? Build fee, monthly support, and the third-party costs you will pay directly (AI usage, telephony, automation platform) should all be spelled out.
- What does support look like after launch? Monitoring, prompt and workflow updates, response times, and who you talk to.
- What does handover include? Documentation, a walkthrough, and enough access that another developer could take over if needed.
Green flags and red flags
| Green flags | Red flags |
|---|---|
| Starts with your process and numbers | Starts with a tool they always use |
| Builds in your accounts, you own everything | Everything lives in their accounts |
| Clear itemised pricing, including third-party costs | One vague monthly fee |
| Human escalation and logs built in | "The AI handles everything" |
| Realistic timelines with a narrow first phase | Guaranteed results or "fully autonomous" promises |
| Talks you out of unnecessary features | Upsells every add-on |
Agency, freelancer or in-house?
Freelancers can be great for a single well-defined workflow. An in-house hire makes sense once automation is core to how you operate and you have steady work for them. An agency fits in between: you get design, build, integration and support from one team, without hiring. Whichever you choose, the twelve questions above still apply.
At Voxil AI you work directly with the people building your system, and you own everything we build. If you would like a second opinion on a proposal, or want to scope a first project, start with our AI automation consulting or book a free call.
Sources
- MIT NANDA, The GenAI Divide, 2025
- Gartner press release, 2025
Figures are reported as published by each source. Forecasts are the source’s predictions, not guarantees. We review this article regularly; spot something outdated? Tell us.
Written by Abdul Moeez, AI Automation Expert
Abdul designs and builds the AI voice agents, chatbots and automation systems Voxil AI ships: from conversation design and integrations to testing on real calls.