How to Choose an AI Automation Consultant
June 8, 2026 - by Themba Mahlangu - 5 min read
When you hire someone to automate your work with AI, the thing that decides whether it works is not the model they use or how good the demo looks. It is whether they can connect AI to your tools in a way that keeps running, and tell you the truth when something breaks. Most of the work, and most of the risk, sits there. So that is what to judge them on.
Most buyers judge the opposite things: which model the consultant uses, how fast they can show a working chatbot, how confident they sound in the room. None of that tells you whether the system still works in a few months.
What actually breaks
The failures in these projects are almost always boring. Usually a connection expires and nobody reconnects it, so the automation quietly stops. Sometimes a step that worked in the demo breaks once it is running against real accounts at real volume. The most damaging case is a wrong number, because the moment someone catches it, they stop trusting the whole system.
We saw all of this when we ran Hyper. The integration people reached for most, Meta's ad tools, was also one of the least reliable, because an upload step kept failing in the middle of campaigns. One customer left for a competitor over it, while the model itself had been working fine the whole time. The breakage was in the plumbing, and the model had nothing to do with it.
A consultant worth hiring has hit these failures before and builds for them from the start. That is what you are paying for, and it is invisible in a demo. It is also the pattern in the wider data: Gartner expects organisations to abandon 60% of AI projects through 2026 that are not built on connected, usable data (Gartner, February 2025).
The questions that predict whether it works
A few direct questions tell you more than any demo.
What would you connect first, and why? A consultant who knows what they are doing has a specific answer for your business, usually the one workflow you run every day, because that is where the value shows up and where you notice quickly if it breaks. Someone who wants to connect everything and see what sticks will leave you with several half-working integrations and no clear result.
Who keeps it running, and how do they know when it breaks? Connections expire and platforms change their rules without warning. Ask whether they monitor the connections or whether you will be the one to discover something stopped. Ask whether a failed step retries or just disappears. If maintenance is treated as an afterthought, the system will break and stay broken.
Am I buying a model, or an integration? If the pitch is mostly about how advanced their model is, slow down. You can swap the model out easily. The work that lasts is the integration around it, and that is what you are actually paying for.
A good consultant starts cheap
The clearest signal is whether they want to look before they build. Someone confident in their judgement starts with a small paid audit: a close look at your business that produces one specific thing to automate, with a clear view of what to connect and where it is likely to break. Someone who pushes straight to a large build, before understanding your business, needs the contract more than they trust their read on it.
A good audit makes you pick one problem to solve that is actually worth solving, and tells you plainly when AI is not the answer.
What it should cost
A consultant who will not give you a range is hiding something. An audit should start around $3,000, and a good one is credited toward the build if you go ahead. A focused integration, connecting your tools and getting one workflow running reliably, usually runs $5,000 to $15,000, depending on how many systems are involved and how messy the data is. Keeping it running is a separate, ongoing cost, usually a retainer of $2,000 to $6,000 a month. If a build is quoted with no plan for maintenance, it will break after launch and stay broken.
For a larger company, the question is return. Weigh the cost against the work being automated and against the failures you are avoiding. A connection that silently breaks for a week, or a wrong number that ends a client relationship, costs more than the maintenance that prevents it.
The red flags, in short
Keep looking if a consultant leads with the model and skips the integration, cannot name your first connection or defend the order, has no answer for who maintains the system after launch, only ever demos on clean test data, or pushes a big build before understanding your business. The work here is mostly invisible plumbing, and the people who are good at it will happily walk you through that plumbing in detail.
How we work
We are an audit-first AI partner for agencies and small businesses. We start with a paid audit because it is the honest way to find the one workflow worth automating and to tell you, before you spend on a build, where it is likely to break. Then we build the integration around those failure points and maintain it, so a dropped connection is our problem and not yours.
Sources
Gartner, "Lack of AI-Ready Data Puts AI Projects at Risk", 26 February 2025