How to Choose an AI Automation Partner: 12 Questions That Reveal Everything

Maps workflows before quotingDesigned exception handlingLeadership-visible audit trailYou own the code outrightSystems in production 1yr+12QUESTIONSSPIGA CONSULTING

The AI services market is flooded with demo shops: teams that can wire a model to a UI in a weekend but have never kept a system alive through a quarter of real operations. The 12 questions below expose the difference in one meeting. They cover the five things that actually predict success — diagnosis discipline, exception handling, auditability, ownership, and operating history — and any partner worth hiring will enjoy answering them.

Questions 1–4: how they start

How a partner begins tells you how they'll finish.

  • 1. Do you map our workflows before quoting, or quote from a call? (Diagnosis before prescription — anything else is guessing with your money)
  • 2. What would make you tell us NOT to automate something? (A partner with no disqualification criteria is a vendor, not an advisor)
  • 3. Which workflow would you automate first, and why that one? (Listen for volume, pain, and measurability — not whatever's trendy)
  • 4. What do you need from our team, and how many hours? (Honest answers here predict honest timelines later)

Questions 5–8: how it runs

Production is where demos go to die. These four separate the survivors.

  • 5. What happens when the agent hits a case it can't handle? (You want a designed exception path with human review, not 'the model is very good')
  • 6. Can leadership see an audit trail of every action? (If not, it will fail its first executive question)
  • 7. How do you test before production? (Real historical data, edge cases, stress tests — not a happy-path demo)
  • 8. What breaks when our systems change, and who fixes it? (Systems live in changing environments; maintenance posture matters)

Questions 9–12: what you own

The difference between an asset and a hostage situation.

  • 9. Who owns the code and the prompts? (The only acceptable answer: you do, outright)
  • 10. Can we run it without you? (Yes should be immediate — with a handover plan)
  • 11. What's the model dependency? (The foundation model should be a swappable component, not a lock-in)
  • 12. Show me a system that's been in production over a year. (Operating history is the credential; demos age in weeks)

How we answer them

For transparency: we diagnose before we build (paid discovery that maps every system and handoff), we design exception paths and audit trails into every build, clients own 100% of the code with the model as a swappable component, and our systems run daily operations across finance, logistics, real estate, and SaaS — some for years. We've eliminated over 50,000 hours of manual work that way.

The fastest way to evaluate any partner — including us — is to bring a real workflow and watch how they think about it. Book a conversation, bring your messiest process, and ask all twelve.

Frequently asked questions

What should I look for in an AI automation company?

Five predictors: they diagnose before quoting, they design for exceptions (human-in-the-loop paths), every action is auditable, you own the code outright with no platform lock-in, and they can show systems that have run in production for a year or more.

What are red flags when hiring an AI vendor?

Quoting without mapping your workflows, no answer for exception handling, no audit trail, retained code ownership or proprietary-platform lock-in, and a portfolio of demos rather than operating systems. Aggressive certainty ('AI can automate anything') is itself a red flag.

Should we run a pilot before committing to a big program?

Yes — but a production pilot, not a sandbox demo. A tightly scoped first system on a real, high-volume workflow proves ROI and reveals how the partner operates. A demo on synthetic data proves nothing about production.

Keep reading