The clearest AI consulting red flag is a firm that promises results before it has seen your process or your data. Close behind: no named people, no answer on where your data is processed, AI described as doing work that people actually do, and no plan for what happens when you stop. Each one can be checked in writing before you sign.
#Key takeaways
- A promised outcome before discovery is a sales line, not a plan.
- If the firm cannot name who does the work, assume someone you have not met will.
- Ask what is automated and what a person does. A vague answer is the warning.
- No hosting answer means no one has thought about your data rules yet.
- You should be able to stop and keep the accounts, the data and the off switch.
#Why do buyers need a red-flag list for AI firms?
Because a lot of what is sold as AI is relabelled. Gartner described "agent washing", vendors rebranding chatbots, robotic process automation and assistants as agentic AI, and estimated that only about 130 of the thousands of vendors claiming agentic AI are real (Gartner, june 2025). In the same release Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027.
The firms selling those projects are not all bad actors. Many simply start with a tool and look for a place to put it. The checks below catch both.
#Which red flags show up in the first meeting?
- results promised up front. a percentage saving quoted before anyone has mapped the process.
- no names. "our team" instead of the people who will run interviews and build.
- tool first. the conversation starts with a model, a platform or a demo, not your workflow.
- no hosting answer. a shrug when you ask where data is processed and stored.
- no human in the loop. nobody asks who on your side should approve the output.
#Which red flags show up in the proposal?
Read what the proposal says you will receive. If it lists activities (workshops, interviews, a roadmap) but no artifact you can act on, the engagement ends at the recommendation. Ask for a sample deliverable with the client removed.
Then read the exit terms. Who owns the accounts the AI runs in? Where does the data sit after the work ends? Can your team switch it off without calling the firm? If those answers are missing, you are renting something, not buying it.
#What does a good answer look like?
A good answer is specific and checkable. "the audit returns five ranked workflows, each with the cost it removes and how it is measured, plus one recommended build with its hosting region" is checkable. "we help you unlock value with AI" is not. Human oversight should be named too: the NIST AI risk management framework, january 2023 treats clear accountability for AI output as part of managing risk.
We are
Stennir says what is automated and what a person approves, names the people who do the work and writes the hosting decision into the scope.
We aren't
Stennir does not promise a saving before mapping the process, and does not describe work as automated when a person is doing it.
#What should a buyer do next?
Turn the lists above into written questions and send them to every firm you are considering. Compare the answers, not the presentations. There is a longer question list in how to choose an AI consulting company. To see Stennir's own answers, read AI consulting and how we work, or book a 30-minute call. The same checks apply to any department; AI for operations shows where most first workflows sit.
#Common questions
Is it a red flag if an AI firm has few public case studies?
Not on its own. Many AI engagements sit under nda, and newer firms have fewer stories to publish. The red flag is a firm that cannot describe its deliverable, name its people or show a sample artifact with the client removed. Ask for those instead of logos, and judge the specificity of the answers.
Is it a red flag if the firm recommends not using AI?
No, it is a good sign. Some processes are better fixed with a clearer rule, a form or a simple integration. A firm that only ever recommends AI is selling AI. A firm that sometimes says "not this one" is looking at your process. Stennir's audit can return workflows where the answer is a plain fix.
How do we check claims about what the AI does?
Ask to see it run on an example of your own input, with the person who built it in the room. Ask which steps are automated and which are done by a person, and how errors are caught. If the demo only works on prepared data, treat the claim as unproven until it runs on yours.
Where does our data stay if we work with Stennir?
Where your rules require. The hosting decision is written into the scope before anything is built: which country, which cloud and which model can run there. The accounts are in your name and the off switch sits with your team. Stennir works with operations teams in the UAE and Saudi Arabia, so this is a standard first question.
