Optional analytics

With your permission, we use Google Analytics and Vercel Analytics to count visits and improve the website. We do not use this data for advertising or send form details. Cookie policy

Skip to content
Back to blog
Comparing3 min read

How to choose an AI automation agency and what to ask

Nine questions for the first call, the evidence to request and the operating details that matter after launch.

ByAsier Mugica· Founder of Ace Digital· Updated 2026-09-18

A polished demo shows that one path can work. It does not show how the system will handle your data, permissions, exceptions or later changes. Choose an agency by how clearly it defines the business process, evidence, failure modes and operating responsibility.

The nine questions

  • What exactly will exist at the end, written down? If the deliverable cannot be described specifically enough to disagree about, it cannot be delivered.
  • How will you bill, and who carries estimation and scope-change risk? Fixed price may fit stable scope; time-based billing may fit discovery.
  • Who owns and can access what you build? Ask to see the configuration during the project, not after.
  • What happens when an integration breaks? Not whether, when. APIs change. Ask who notices and how fast.
  • Which unit rates and volume assumptions are used to estimate ongoing cost?
  • Can my team run this if needed, and what documentation, accounts and access will we receive?
  • What have you built that is closest to this? Ask for the one that went badly too. The answer tells you more than the case study.
  • What would make you advise against this project or reduce its scope?
  • Who is actually doing the work? On small engagements, find out whether the person on the call is the person building it.

Answers that should worry you

  • "We'll figure out the scope as we go" without a discovery boundary, budget cap or change-control process.
  • A percentage of savings without an agreed baseline, attribution method, exclusions and audit rights.
  • Vague about ownership. If they cannot answer who has access, assume you do not.
  • Case studies with no timeframe, baseline, sample, exclusions or definition of success.
  • No mention of maintenance. Automations are not furniture. They need somebody watching them.

Agency, freelancer or build it yourself

OptionBest whenThe risk
Build it in-house with a no-code toolThe process is simple, internal, and low-stakes if it failsMonitoring and maintenance may lack a named owner
FreelancerOne well-defined piece, and you can specify it yourselfNo continuity, and the knowledge leaves with them
AgencyIt touches customers or revenue and has to keep workingYou pay for overhead you may not need on a small job

Ask what happens after launch

Ask who monitors failures, reviews quality, updates documentation and responds when an API or business rule changes. Put response expectations, included support and handover conditions in writing. This does not guarantee a good outcome, but it makes responsibility and first-year cost much easier to compare.