What's the single biggest red flag when vetting an AI automation agency?

An agency that can't clearly explain what happens to your code and data if you stop working with them. If ownership and exit terms are vague or evasive, everything else about the engagement is probably vague too.

Should I hire an agency that also sells its own AI products?

It's a reasonable point in their favor, not against them — an agency running its own product in production (like we run Botnira) has to solve the reliability, cost, and support problems you're about to pay them to solve for you, which is a different level of accountability than one that only ever ships client work.

How much should a first AI automation project cost?

It depends heavily on scope, but a narrow, well-defined first project — one workflow, one channel — is usually the right size to test an agency before committing to something larger. Be wary of any agency that only wants to sell you a large, multi-month engagement out of the gate.

Guide

12 questions to ask before hiring an AI automation agency

Published September 1, 2026

Before hiring an AI automation agency, ask who owns the code, how they handle your data, what happens if the underlying AI model changes or gets deprecated, how pricing is structured, and what support looks like after launch — the answers separate an agency that can actually run the thing they built you from one that only knows how to ship it.

1. Who owns the code when the project ends?

Ask this first, in writing. Some agencies hand over full source and infrastructure access on delivery; others keep the build on their own platform and effectively rent it back to you month to month. Neither model is automatically wrong, but you need to know which one you're signing up for before the invoice, not after. If you're told "we'll figure that out later," that's your answer.

2. Does the agency run its own product, or only build for clients?

An agency that operates its own live AI product has to deal with the same problems you're about to pay them to solve — uptime, cost control, edge cases real users hit, support tickets at 11pm. That's a different level of operational maturity than a shop that has only ever delivered one-off client builds and moved on. Ask what they run themselves and for how long.

3. How is customer and business data handled?

Get specific: where is data stored, is it used to train any model, who has access, and what's the retention policy. If the agent will handle anything sensitive — health information, financial details, immigration case files — ask directly how compliance is handled for your jurisdiction and industry. A vague "we take security seriously" is not an answer.

4. What happens if the underlying AI model changes or gets deprecated?

Every AI build sits on top of a model provider's API. Providers deprecate models, change pricing, and shift behavior with new versions — it's happened repeatedly across the industry and will keep happening. Ask how the agency's architecture isolates you from that churn, and whether swapping a model provider is a configuration change or a rebuild.

5. Is pricing a flat build fee, a subscription, or usage-based?

Get the full pricing structure before you start, not just the headline number. A flat development fee might exclude hosting, model API costs, or per-message/per-minute charges that show up later. A subscription product should have a published price list you can check yourself — the way our own pricing page lays out Botnira's plans, from $29/mo up.

6. What does support look like after launch?

Ask what's included after go-live and for how long — bug fixes, model tuning as real conversations reveal gaps, changes when your business process changes. A launch is the start of the useful life of an agent, not the finish line. If ongoing support is a separate negotiation with no defined terms, expect friction later.

7. Can they show a live example, not just a demo script?

A rehearsed demo proves very little. Ask to see or call a live agent handling real, unscripted traffic — a client's actual phone line or chat widget — and ask what its failure modes are, honestly. An agency that's comfortable showing you where their agents struggle is more trustworthy than one that only shows the highlight reel.

8. How is escalation to a human handled?

No agent should try to handle everything. Ask what triggers a handoff to a person, how much context the human gets when it happens, and whether that logic is configurable by you or hardcoded by the agency. This detail tells you more about how the system will actually behave under real conditions than any list of features.

9. What integrations are actually supported, versus theoretically possible?

"We can integrate with anything" is a sales line, not a technical claim. Ask specifically whether they've built a working integration with your calendar system, CRM, or line-of-business software before, and ask to talk to a reference where they did. A first-time integration for your specific stack should be scoped and estimated honestly, not waved through.

10. How long does a first build actually take?

Timelines that sound too fast for the scope usually mean corners get cut on testing, or the "agent" is closer to a scripted decision tree than something that can handle real variation in how people talk or write. Ask for a realistic range and what happens if it slips.

11. What happens if you want to leave?

Ask about contract length, notice periods, and what you walk away with — configuration, conversation logs, code — if you switch providers later. An agency confident in its work has no problem answering this clearly; one that's counting on lock-in usually hedges.

12. Can you start small before committing to something big?

A single, well-scoped workflow — one channel, one clear job — is a reasonable way to evaluate an agency before a larger engagement. Be cautious of anyone who insists on a large, multi-month contract before you've seen a single deliverable from them.

The honest recommendation

Most of these questions aren't about catching an agency in a lie — they're about surfacing how they actually operate before you're committed. A good agency will answer all twelve without hedging. If you get vague answers on ownership, data, or what happens after launch, keep looking; the technology is rarely the risky part of one of these projects, the vendor relationship usually is.