The AI automation agency market in the UK has exploded over the past 18 months. Every strategy consultancy now has an "AI practice." Every freelancer has rebranded as an "AI specialist." Procurement teams are drowning in proposals.
The problem: most of them are selling strategy. You need deployment.
Here's how to tell the difference — six questions that cut through the noise.
1. "Can you show me something you've shipped?"
This is the most important question you can ask. Not a roadmap. Not a methodology deck. An actual deployed automation — what it does, which tools it connects, what results it delivered.
Agencies that build will have specific, detailed examples. They'll talk about edge cases they had to handle, integrations that were trickier than expected, and how they measured the outcome.
Agencies that consult will show you case study slides with vague metrics and no technical detail.
At Squirrel AI, every discovery call starts with us walking through live systems we've built — the deal origination engine that processes 10,000+ opportunities in 24 hours, the customer support agent that drove a Trustpilot score from 4.2 to 4.8. We can show you the workflow, explain the architecture, and point you to the results.
Red flag: Any agency that can't point to deployed systems with measurable outcomes.
2. "Who actually builds the automations?"
AI agencies fall into two categories: those where the person you're talking to builds the automations, and those where the pitch is done by a senior consultant and the work is passed to a junior offshore team.
Neither model is inherently wrong — but you need to know which one you're buying.
If you're spending five figures on an automation project, understand who will be on-site, who will build, and whether that person has done it before. Ask to speak with the builder before you sign.
At Squirrel AI, we're a small team of founders and engineers. When we take on a project, the people who design it are the people who build it. We don't have an "account management" layer between you and the work.
Red flag: Vague answers about team structure, or promises of "dedicated project managers" without clear answers on who writes the code.
3. "What's your tech stack?"
An AI automation agency should have strong opinions on tooling. They should be able to explain why they use what they use — not just list it.
What to look for:
- A primary automation platform with genuine expertise (we use n8n — self-hosted, open-source, no per-execution costs, and native AI nodes)
- A clear approach to AI model integration (Claude, GPT-4.5, Gemini — and when to use which)
- Honest views on platform limitations ("we wouldn't use Zapier for this because...")
- Data privacy considerations, especially for regulated industries
Red flags: Agencies that claim to be platform-agnostic (usually means they don't have deep expertise in any platform) or only use SaaS automation tools (expensive at scale, can create data privacy issues).
4. "What does your discovery process look like?"
The best automation projects start with a tight discovery phase — not a six-week strategy engagement. You should be able to map the highest-ROI automation opportunities in your business in 1-2 days, not 6-12 weeks.
Ask the agency to walk you through how they would approach your specific situation. A good agency will ask you about:
- Your current tech stack and integration points
- Which manual processes take the most team time
- Your error rates and what happens when things go wrong
- Your growth trajectory and what that means for operational capacity
If the discovery process leads to a 50-page strategy document, you've hired a consultant. If it leads to a prioritised list of automations ranked by ROI and buildability, you've hired an agency.
5. "How do you handle maintenance and things going wrong?"
Production automations break. APIs change. Vendors update their data schemas. An automation that ran perfectly for six months can fail when a CRM updates its field structure.
Ask specifically:
- Do you build error handling and alerting into every automation?
- Who do we call when something breaks?
- What does your support model look like post-deployment?
- Are our automations documented so we can maintain them internally if we need to?
At Squirrel AI, every automation includes monitoring, error notifications, and documentation. We build on n8n's self-hosted model, which means our clients own their infrastructure and aren't dependent on us to keep it running. Ongoing support is available but optional.
Red flag: Agencies that don't discuss error handling upfront, or whose model requires you to stay on retainer to keep the lights on.
6. "How do you measure success?"
Every project should start with defined KPIs. Before any work begins, you should agree on what "success" looks like — and it should be specific and measurable.
Bad KPIs: "improved efficiency," "better customer experience"
Good KPIs: "time from lead submission to first contact reduced from 4 hours to under 60 seconds," "manual hours per week on invoice processing reduced from 20 to 2," "document turnaround time halved with error rate below 2%"
An agency that builds will push you to define these metrics upfront, because they're confident they can hit them. An agency that consults will avoid them, because vague deliverables are easier to defend.
The Bottom Line
The UK AI automation market is full of agencies that will take your money and deliver a framework. A much smaller number will take your money and deliver a working system.
The six questions above are designed to separate them quickly. The right agency will welcome them — and will give you detailed, specific, confident answers.
Nihaar Udathu is Co-Founder of Squirrel AI. Want to put us to the test? Book a discovery call — we'll walk you through live systems and tell you exactly what we'd build for your business.