The AI automation market is flooded with agencies promising to “transform your business” and “10x your productivity.”
Most of them are talking rubbish.
Here’s the uncomfortable truth: 95% of AI pilots never make it to production. That’s not a typo. Ninety-five percent. And a good chunk of that failure rate comes down to choosing the wrong partner in the first place.
I’ve spent 20 years in business, the last five deep in AI systems. I’ve seen brilliant implementations that saved companies thousands of hours. I’ve also seen six-figure disasters that delivered nothing but fancy PowerPoint decks and buyer’s remorse.
The difference? Knowing what to look for before you sign anything.
The AI Agency Landscape in 2026
Let’s be honest about what you’re walking into.
The UK AI consultancy market hit £23.9 billion in revenue last year — a 68% increase from 2023. Everyone and their dog now calls themselves an “AI agency.” Some are genuinely excellent. Many are marketing agencies who bolted ChatGPT onto their services last Tuesday.
You’ll encounter three types:
Strategy-first consultancies — They’ll audit your processes, identify opportunities, and design solutions. Some stop there. Others implement too.
Implementation specialists — They build the workflows, integrations, and automations. Less strategic thinking, more technical delivery.
Full-service partners — Strategy through to production, ongoing support, the lot.
None of these is inherently better. The right choice depends entirely on what you actually need.
Seven Questions to Ask Before Signing
These aren’t soft questions. They’re designed to separate the genuine operators from the hype merchants.
1. “What percentage of your AI projects reach production?”
Industry average: 31% make it to production. That’s the benchmark.
If they can’t answer this question — or if they dodge it — walk away. Any agency worth their fees tracks this metric obsessively.
Good answer: “72% of our projects reach production within six months.”
Red flag: “Well, it depends on the client…“
2. “Can you show me three case studies with measurable outcomes?”
Not testimonials. Not “we worked with a retail client.” Actual numbers.
You want to hear things like:
- “Reduced invoice processing time from 4 hours to 12 minutes”
- “Saved 2,000 hours annually in back-office admin”
- “Cut data entry errors by 99%”
If the case studies are vague, the results probably were too.
3. “What happens when the pilot fails?”
This question catches people off guard. Because pilots do fail sometimes. The question isn’t whether — it’s how they handle it.
Good answer: “We have a structured iteration process. If the initial approach doesn’t work, we diagnose why, adjust, and re-test. We’ve rescued projects that looked dead.”
Red flag: “Our pilots don’t fail.” (They’re either lying or haven’t done enough of them.)
4. “Who actually does the work?”
Agencies love to wheel out their senior team for pitches, then hand the project to juniors.
Ask specifically:
- Who leads the discovery phase?
- Who builds the automations?
- Who do we contact when something breaks at 9pm on a Tuesday?
Get names. Check their LinkedIn. This matters.
5. “What’s your approach when we don’t need AI?”
The best agencies will sometimes tell you that AI isn’t the answer. That your problem is better solved with a simple Zapier workflow, or a process change, or hiring someone.
If every problem looks like an AI nail to them, they’re selling hammers, not solutions.
6. “What does post-project support look like?”
This is where most agencies fall over.
They build something brilliant, hand it over, and vanish. Six months later, the workflow breaks because an API changed, and you’re stuck.
Ask about:
- Ongoing support contracts
- Response times for issues
- Who maintains and updates the automations
- Training for your team
7. “Can we speak to a client who had problems with your work?”
Anyone can provide happy references. Ask for someone who had issues — and how those issues were resolved.
If they refuse, that tells you something. If they provide it, and the client says the agency handled problems well, that tells you even more.
What UK Agencies Actually Charge
Transparency on pricing is rare in this industry. Here’s what the market actually looks like in 2026:
Discovery and Strategy Phase
| Scope | Typical Cost | Timeline |
|---|---|---|
| Light audit | £7,000–£15,000 | 2–4 weeks |
| Full strategy | £15,000–£30,000 | 4–8 weeks |
| Enterprise discovery | £30,000–£60,000 | 6–12 weeks |
This phase should give you: process maps, opportunity identification, priority recommendations, and a clear implementation roadmap.
Pilot/Proof of Concept
| Complexity | Typical Cost | Timeline |
|---|---|---|
| Simple workflow | £25,000–£40,000 | 4–8 weeks |
| Complex integration | £40,000–£80,000 | 8–16 weeks |
| Multi-system automation | £80,000–£150,000 | 12–24 weeks |
Production Implementation
Expect 2–5x the pilot cost for full production rollout. A £40k pilot might become a £120k implementation.
Day Rates
| Experience Level | Day Rate |
|---|---|
| Junior consultant | £350–£600 |
| Senior specialist | £800–£1,200 |
| Principal/Director | £1,200–£2,500 |
| Board-level advisor | £2,500–£6,000+ |
London agencies typically charge 10–20% more than regional firms.
Red Flags That Should Stop You Signing
After two decades, I’ve learned to spot the warning signs:
“AI will solve all your problems” — No, it won’t. AI is a tool. Tools solve specific problems. If they’re promising transformation without understanding your actual challenges, they’re selling dreams.
No case studies or references — How can they prove they deliver if they can’t show you evidence?
Fixed-price quotes without discovery — Anyone quoting a fixed price before understanding your systems, data, and processes is either padding massively or about to underbid and cut corners.
They talk more about technology than outcomes — If the pitch is all about GPT-5 and vector databases and agentic workflows, but nothing about what actually changes for your business, they’re tech enthusiasts, not business partners.
The senior team disappears after the pitch — Get it in writing who works on your project.
No mention of change management — 95% of AI project failures are organisational, not technical. If they don’t talk about training, adoption, and change management, they’re only solving half the problem.
The Hybrid Model Most UK Businesses Use
Here’s what actually works for most SMEs:
73% of UK businesses now use a hybrid approach — strategy in-house (or with a consultant), implementation outsourced.
This looks like:
Hire a fractional Chief AI Officer or consultant — 1–2 days per month, £2,000–£8,000/month. They set direction, evaluate opportunities, oversee quality.
Use an implementation agency — For the actual building. They follow the strategy your advisor sets.
Build internal capability over time — Eventually, your team handles maintenance and smaller automations.
This model costs more upfront but delivers better results. The advisor keeps the agency honest. The agency focuses on what they’re good at. You build knowledge internally.
The Decision Framework
Before you commit, answer these honestly:
What problem are you actually solving? — Not “we need AI” but “we spend 20 hours a week on invoice processing and it’s killing us.”
What does success look like? — Specific numbers. Hours saved. Errors reduced. Revenue unlocked.
What’s your budget — really? — Discovery alone starts at £7k. A proper implementation might be £50k–£150k. If your budget is £10k, be realistic about scope.
What’s your timeline? — Rushing AI projects is how you join the 95% failure rate.
Who internally will own this? — Someone needs to champion the project, work with the agency, and drive adoption. If that person doesn’t exist, fix that first.
What to Do Next
If you’re serious about AI automation:
Audit your processes first — Map where time actually goes. Find the bottlenecks. You can do this yourself with a spreadsheet.
Talk to 3–5 agencies — Use the questions above. Compare approaches, not just prices.
Ask for a paid discovery phase — Any agency worth their fees will scope properly before quoting implementation. This protects both of you.
Start small — One workflow. One process. Prove it works. Then scale.
Plan for the long term — AI automation isn’t a one-off project. It’s a capability you’re building.
The agencies that deliver results aren’t the ones with the flashiest websites or the biggest promises. They’re the ones who ask hard questions, give honest answers, and care more about your outcomes than their fees.
Find those people. Work with them.
FAQ
How long does an AI automation project typically take?
Discovery: 2–8 weeks. Pilot: 4–16 weeks. Production implementation: 8–24 weeks. Expect 6–12 months from first conversation to full deployment for complex projects.
What’s the typical ROI on AI automation?
UK businesses report an average 240% ROI within the first year, with payback periods of 6–12 months. However, this varies massively by use case and implementation quality.
Do we need to prepare our data before engaging an agency?
Yes. “Dark data” — unstructured, inaccessible, or poor-quality data — is a leading cause of AI project failure. A good agency will audit your data readiness during discovery.
Can we start with a small budget?
Yes, but be realistic. A simple workflow automation might cost £25k–£40k. If your budget is under £20k, consider DIY tools like Zapier or Make, or a consultant to guide your internal team.
What if we’ve already had a failed AI project?
Common. 95% of pilots fail. The question is why it failed. A good agency will diagnose the root causes — usually organisational, not technical — before proposing a new approach.
Related articles:
