The short version
- SMBs (10-500 people) have three options: SaaS, enterprise consultants, or a boutique AI agency. Most fall into the gap between options 1 and 2.
- Boutique agency = senior developer-led, fixed-price project, £4K-£150K, 6-10 weeks, no account managers, you own the code.
- SaaS hits a ceiling at $2M-$20M ARR for most SMBs — you outgrow Shopify apps, Gorgias, Vic.ai, Lofty, Avoca etc.
- Enterprise consultancies (Big-4, Accenture, etc.) are designed for £500K+ engagements. The economics don’t fit SMBs.
- The boutique middle ships fast, costs less, owns less of your business than either alternative, and pays back in 6-12 months.
If you run an SMB — somewhere between 10 and 500 employees, doing real revenue, a real team, real systems — and you’re trying to figure out how to actually use AI in your business beyond ChatGPT for everyone, you’ve probably noticed that nobody is selling to you.
The SaaS vendors are selling to everyone, generically, $99/month. The enterprise consultancies are selling to people three brackets above you. The hyperscalers (AWS, Azure, GCP) are selling to your IT team but the AI bit is buried under a 14-product platform pitch. Your inbox fills up; nothing fits.
This is the case for the boutique AI agency model. Not because it’s glamorous — it absolutely is not — but because it’s the right shape for what 10-500 person companies actually need.
The three options actually on the table
Option 1: SaaS
You sign up for an off-the-shelf product. ChatGPT Enterprise, Microsoft Copilot, Lindy, Chatbase, Gorgias, Vic.ai, Lofty, Avoca, Adit, Yuma, Spellbook, Harvey, Klaviyo, the list goes on. €20-€800/seat/mo. You configure within their bounds. It works fine for the average customer.
Strength: live in days. Cheap upfront. No project to run. No vendor to choose. No fight with IT.
Weakness: SaaS is built for the 80% case. SMBs are weird in their specific 20% — that’s how you stay in business. The moment your weird matters more than your average, SaaS hits a ceiling. You stack 5 SaaS tools to cover the gap, end up with $4K/mo of subscriptions and still nothing quite fits the workflow.
Option 2: Enterprise consultancy
You commission a Big-4, a Big-Tech-Reseller, or a regional “digital transformation” firm. £200K-£2M engagements. Six-month timelines. 12-page proposals. Multiple accounts on the agency side. PMs, BAs, strategists, junior devs.
Strength: capability and brand cover. Useful when the work is genuinely massive (multi-country rollout, regulated industry, multi-system) and your CFO needs to point at a recognisable name to defend the spend.
Weakness: the economics are designed for clients 5-10x your size. You pay for layers of people who aren’t doing the work. The first three weeks are kickoff theatre. The senior people you met in the pitch don’t do the build. The output is reasonable, the cost is unreasonable, the speed is glacial.
Option 3: Boutique AI agency
A small specialist team or solo senior developer. £4K-£150K. 6-10 weeks. No PMs. No account managers. The person you meet in the pitch is the person doing the work. Fixed price after a one-page spec.
Strength: economics that fit SMB scale. Speed that matches actual business cycles. Senior thinking applied to your actual problem rather than to the agency’s billing structure.
Weakness: bus factor. If your boutique disappears, you’re on your own. Mitigated by clean code, owned IP, no vendor lock-in — but you have to insist on those terms upfront.
Estimate your custom AI project in 30 seconds
Three questions, an instant cost range and timeline based on real shipped projects. After 30 minutes on a discovery call you have a written fixed-price quote.
When SaaS stops being enough
Almost every SMB starts with SaaS. That’s correct. The economics are unbeatable for the first 12-24 months of trying to figure out where AI helps. The wrong move is to commission custom for “summarise our emails” or “help us draft documents” — ChatGPT Enterprise does that better than custom.
SaaS stops being enough when one or more of these is true:
- Your weird is your differentiator. Your sales process, your service model, your customer language has 5 things nobody else does. SaaS rounds those off; custom keeps them.
- You’ve stacked 4+ tools to cover one workflow and the seams are starting to break. Tickets fall between Gorgias and Klaviyo. Returns get stuck between Yotpo and Loop. The duct tape is getting expensive.
- SaaS subscriptions exceed £3K-£10K/month for the workflow in question. At that spend, custom pays back in under 18 months and stops compounding.
- You need actions, not answers. SaaS AI deflects tickets. Custom AI issues refunds, generates labels, updates CRMs, sends payments. The cost of failure is real, the audit trail matters, the integration is deep.
- Vendor lock-in is starting to feel risky. Acquired? Going public? Fundraising? Buyers and investors don’t love finding out the company runs on a SaaS stack that can be price-hiked or sunset.
If 2 or more of these are true, you’re in custom-AI territory. The next question is who builds it.
When enterprise consultancies are wrong for you
If you’re reading this, the answer is probably: now.
Enterprise consultancies aren’t bad. They’re great at certain things — multi-country rollouts, regulated-industry transformations, executive-level political cover, brand reassurance for the board. If your project genuinely involves ten countries, three regulators and a £500K+ budget approved by a steering committee, hire one.
For everyone else, the maths is:
- Their cost structure is designed for project sizes 5-10× yours. A £200K project at a Big-4 is a small project — staffed by junior consultants supervised by a partner who flies in for the steering committee.
- The project ramp is half the budget. Six weeks of discovery workshops to scope what a senior developer can scope in two hours of screen-share. You pay for the kickoff theatre.
- Speed doesn’t scale down. Six-month timelines are not a sign of seriousness; they’re a sign of headcount overhead. SMB cycles don’t accommodate them.
- The handover is rough. Big consultancies hand you over to your in-house team at the end. If you don’t have an in-house team capable of receiving the handover, you’re stuck.
None of this is dishonest. It’s the operating model the firm exists to deliver. It’s just not aimed at you.
What “boutique AI agency” should actually look like in 2026
The term gets abused. Plenty of agencies that call themselves “boutique” are actually small versions of the enterprise consulting model — same theatre, smaller stage. Real-deal boutique looks like this:
- Senior developer-led. The person on the discovery call is the person writing the code. Not a sales engineer who hands you off. Not a PM who translates between you and an offshore team. A senior shipper.
- Fixed price, after one page of scope. The discovery call ends with a one-pager that defines what gets built. Quote becomes fixed price, signed in week 1. After that, scope changes are explicit and re-quoted. No surprise invoices.
- 6-10 weeks for most projects. If your project is 6 months, it’s either too big for a boutique (in which case go enterprise) or too small for them (in which case it’s overscoped, push back).
- Weekly Friday demos. Live software, every week, from week 2 onwards. Not status decks. If something is wrong, you see it in week 3 and have time to course-correct.
- You own the IP. Code is yours. Prompts are yours. Integrations are yours. The agency builds, hands over, optionally maintains. No license fees, no exit fees, no vendor lock-in. Walk-away clause day-one.
- Direct comms. WhatsApp, Slack, email — whatever’s natural. Same-day responses to urgent stuff. Async-first.
- Honest scope. If the project is wrong — SaaS would do it cheaper, in-house would do it better, the use case is half-baked — the agency tells you. The wrong agency takes the money and ships something that doesn’t move the business.
If your candidate boutique misses 3+ of these, they’re a small enterprise consultancy in disguise. Keep looking.
The economics, with numbers
Three real-world example projects to make the point. All anonymised, all shipped.
| Scenario | SaaS path | Big-4 path | Boutique path |
|---|---|---|---|
| D2C returns automation, $8M ARR | Loop + Yotpo + Yuma stack ≈ $2.4K/mo, gaps remain | £180K, 6 months, two-PMP'd engagement | £26K, 7 weeks, 78% of returns automated |
| HVAC company, 32 techs, missed-call problem | Avoca + Lacy ≈ $1.5K/mo + $2/call | Big-4 wouldn’t take the project | £14K one-off, 5 weeks, 58 captured calls/week |
| Mid-market AP automation, €40M revenue | Vic.ai ≈ €5K/mo + setup; partial coverage | €350K transformation engagement | £42K, 9 weeks, AP touches 4 → 0.8 |
None of these are SaaS-vs-boutique slam-dunks. SaaS is the right answer for plenty of cases. But where SaaS leaves a real gap and a Big-4 over-charges by 5-10× for the same outcome, the boutique route is the missing middle.
When the boutique route is wrong for you
Honesty matters. Cases where you should NOT pick a boutique agency:
- Your scope is genuinely massive. Multi-country, multi-regulator, hundreds of millions in budget exposure. You need an enterprise firm with the capacity and the political weight.
- You have a strong in-house team and 12+ months runway. Hire and build. Year 2 in-house economics beat any agency. You also keep the institutional knowledge.
- The business case requires a brand-name vendor for board reassurance. Sometimes the political cost of going boutique is higher than the financial cost of going enterprise. That’s a real consideration; respect it.
- Off-the-shelf actually works for you. If ChatGPT Enterprise + Copilot + a couple of niche SaaS tools genuinely cover 95% of what you need, don’t commission custom. Spend less, ship more.
How to spot the boutique that fits you
Buyer’s checklist. If your candidate agency can answer these directly, in writing, you’re probably in good hands:
- Who specifically writes the code? Will I work with them or with a sales engineer?
- Will you commit to a fixed-price quote after a one-page scope, or do we need 6 weeks of discovery first?
- How long until first production? (If > 12 weeks, the project is either too big or overscoped.)
- Do you do weekly Friday demos with live software, from week 2 onwards?
- Will I own the code, the prompts, the integrations? Will you sign over IP at the end?
- What happens if I want to take this in-house in 12 months? Will the handover work?
- Are you willing to talk me OUT of the project if it’s the wrong shape?
- Can I see one shipped project from a similar-sized client?
The full version of these questions is in this article — takes about 10 minutes to run through with a candidate.
If the boutique model fits you
Two things to do today:
- Try the cost estimator — three questions, a real number in 30 seconds. The numbers behind it come from 50+ shipped projects and they’re honest.
- Or skip the maths and book a 30-minute discovery call. Free, no pitch. By the end you’ll know whether your situation is SaaS-shaped, custom-shaped, or in-house-shaped — and I’ll tell you straight even if it costs me the project.