AI agency red flags: 11 patterns to walk away from

AI agency red flags fall into three buckets: pricing tells, team-structure tells, and contract tells. I see the same 11 patterns reliably when SMEs forward me proposals. The agency looks credible on the surface. The website has case studies. The deck has a section called "our proven process." But buried in the small print are the signals that tell you whether the agency ships systems or sells decks. For the hiring playbook once you know what to avoid, see how to pick an AI agency and AI agency pricing.

Red flag 1: pricing is only available after a discovery call

Any AI agency that refuses to publish pricing is adjusting its quote based on how large your budget looks. The discovery call is a qualification call. By the time you get to a number, you have already spent 45 minutes warming up, and anchoring starts from there. Agencies with real pricing confidence publish it. If a proposal comes back with "pricing to be discussed following a needs assessment," the number will be whatever they think you can bear.

Red flag 2: case studies name impressive logos but show no delivered systems

There is a significant gap between "we worked with Company X" and "we shipped a system for Company X that did Y." Agency case studies routinely fill that gap with vague process language: "partnered with," "supported the team at," "collaborated on an AI roadmap for." A delivered system has a measurable outcome. Ask: what did you ship, what was the metric before, what was the metric after. If the answer is not specific, the case study is a relationship reference, not evidence of a delivered system. See the related flags in AI consultant red flags. The pattern is identical.

Red flag 3: the team page shows 30 people but delivery runs through a partner network

Headcount on a website says nothing about who does your work. I have seen agencies with 40 listed staff where actual build capacity was two freelancers in a Slack channel. The tell is the word "partners" in the team page description or in the delivery section of the proposal. "We work with a vetted network of delivery partners" means your project is being staffed after you sign, not before. Ask before signing: who specifically will work on this engagement, what is their employment status with the agency, and can I speak to one of them.

Red flag 4: the retainer has no monthly deliverable defined

A retainer without deliverables is a subscription to access. You are paying a monthly fee to have the agency on the call list, not to receive a defined output. Every legitimate engagement has at minimum a list of what you will receive each month: a number of systems shipped, a report with specific data, a working integration test. If the proposal says "ongoing support and strategic guidance" under the monthly scope, there is no scope. The agency will invoice you for 12 months and point to the strategy calls as evidence of delivery.

Red flag 5: they use the word "transformation" more than "implementation"

Transformation is the consultant vocabulary for a process that never ends. Implementation is the operator vocabulary for a specific thing that ships by a date. Count both words in the proposal. If "transformation" wins, the agency has priced for a long cycle of managed ambiguity. The transformation framing typically precedes a 12-month engagement at £8,000 to £15,000 per month. The implementation framing produces a working system in four to eight weeks.

Red flag 6: the proposal is a deck, not a scoped statement of work

A pitch deck is a sales tool. A statement of work is a legal document with obligations. An agency that submits a PDF deck as a "proposal" is sending you a marketing artefact that cannot be held to. A real proposal specifies what will be delivered, by whom, in what timeframe, and what happens if those terms are not met. If you receive a 25-slide deck with a "next steps" slide at the end, the statement of work will appear only after you commit verbally. Do not commit verbally.

Red flag 7: they quote a savings number without showing the baseline

"Our system will save you £120,000 per year in operational costs." That number is fiction until you know the current cost baseline it is measured against. I see this regularly in proposals forwarded by SMEs who are genuinely excited about a number that was generated by an agency, not an audit. Ask two questions: what is the current verified cost of the process being replaced, and how do you arrive at that figure from that baseline. If they cannot answer both in the proposal, the number was chosen to close the sale.

Red flag 8: the technical lead in the meeting is different from who does the work

The person presenting the proposal is the senior team member. The person doing your project is a junior contractor hired the week after you sign. This is the staffing model for a significant portion of the AI agency market. The tell is asking directly in the meeting: "will the person presenting this also be the one building it." If the answer is anything other than yes, ask who will, and ask to meet them before signing. An agency that cannot introduce you to your delivery team before contract is protecting a staffing model it does not want you to see.

What does a good AI agency proposal actually look like?

A good proposal names the person doing the work, specifies what they will deliver in the first four weeks, states the price without qualification, includes a stop-loss clause, and defines success in a metric you can verify independently. You should be able to read it without a follow-up call. If the proposal raises more questions than it answers, it was written to generate a call, not to inform a decision.

Red flag 9: there is a setup fee with no specific output tied to it

Setup fees exist to cover agency overhead on new clients: onboarding admin, access provisioning, internal project management setup. None of that is work you should pay for. A setup fee that appears as a line item with no defined deliverable attached to it is a cost transfer from the agency to you. The only legitimate setup fee is one tied to a specific output: environment configured and tested, API access established and verified, initial data pipeline built and running. If the fee is not tied to an output, it is overhead markup.

Red flag 10: they will not put a stop-loss clause in the contract

A stop-loss clause lets you exit the engagement after a defined number of months if specific milestones are not met. Any agency confident in its delivery should accept one. The agencies that refuse to include stop-loss language are implicitly telling you they expect the value to arrive late, or not at all, and that they intend to hold you to the full contract term regardless. The correct response when an agency refuses stop-loss is to walk away. For more on what a fair contract looks like, see AI agency pricing.

Red flag 11: their existing clients are all in the same vertical

An agency with a portfolio concentrated in one sector has one playbook. It works inside that sector because the workflows, the tools, and the data models are similar enough that the same system transfers. That specialisation stops being an advantage the moment your business sits outside it. The tell is a case study section with five clients in hospitality, or six clients in property, or a team page that lists industry-specific certifications but no cross-vertical examples. Ask: what is the most different client you have worked with, and what did you deliver for them. If the answer is vague, the agency cannot adapt.

What does good look like? Five green flags to check for

The inverse of these 11 patterns is a short list of observable signals that an agency will actually ship.

Pricing is public before you get on a call. The proposal is a statement of work with named deliverables, specific dates, and a named delivery person. The agency has clients in at least two different verticals. The contract includes a stop-loss clause. The senior person in the pitch is the person doing the work, or they introduce you to that person before you sign.

For what this looks like in practice, see how to pick an AI agency and the full AI agency guide. For where the red flags overlap with consultant-side patterns, see AI consultant red flags and AI strategy consultant for the comparison.

Does the size of the agency actually matter?

No. Agency size and delivery quality are not correlated in the SME market. Larger agencies carry more overhead and more distance between the person you speak to and the person building. The signal to look for is not headcount but specificity: can the agency name who will build your system, what will be delivered in month one, and what metric it will move.

FAQ

What is the most common AI agency red flag SMEs miss?
The mismatch between the person presenting the proposal and the person doing the work. It is invisible in the sales process because agencies present senior staff and deliver with juniors. Asking to meet the delivery team before signing closes that gap.

Should I ask for references before signing an AI agency contract?
References are useful but not sufficient. An agency controls which references it provides. Ask for one outside their primary vertical, and ask that reference one specific question: what did not go well, and how did the agency respond. That surfaces delivery quality far more reliably than a general testimonial.

How do I know if an AI agency is white-labelling tools?
Ask directly: which tools are built by your team, and which are third-party platforms you are reselling. A straightforward answer either way is fine. An evasive reframe to "our platform stack" is a tell. White-labelling is not automatically a problem, but you should know what you are paying for.

What should a stop-loss clause look like in an AI agency contract?
It should specify the milestone that triggers the exit option, the notice period, and whether fees are refunded or prorated. A typical structure: if deliverable X is not met by month 2, the client may exit with 30 days notice and receive a prorated refund of month 3 fees. Any agency that balks at that structure is not confident in their month-2 output.

Want a sanity-check on a proposal you have been sent? Book a call.