Loading…
Pricing · Canarlo, Leeds
Four tiers, zero to a hundred grand. Written for the technical founder who has been quoted a six-figure discovery phase and walked. Honest prices, real comparisons, no day-rate fog.
Framing
"How much does AI engineering cost?" is the question every buyer types. It is the wrong one. The real question is what it costs not to ship this in the next quarter — while your competitor does.
AI engineering is a product investment, not a consulting engagement. You are buying a system that earns. Not a deck. Not a discovery phase. Not a slide that says "rebuild roadmap."
Scope-hour pricing rewards the agency that drags things out. Outcome pricing rewards the studio that ships. We publish fixed tiers because the work is repeatable — canarlo-core does the boilerplate, your money buys the part that is yours.
Cost drivers
Driver 1
A single agent with three tools sits at the bottom of the £25k tier. Six agents, retrieval, evals, and a human-in-the-loop dashboard pushes you to £60k. Scope creep is the silent killer — we fix it in writing.
Driver 2
Clean Postgres with documented columns? Quick. A SharePoint folder, a dead Notion, and three CSV exports labelled final_v4? Add four weeks. Most of the cost gap between tiers is data work, not model work.
Driver 3
One Stripe webhook and a Slack post — trivial. Salesforce, HubSpot, an internal SOAP endpoint from 2014, and a partner API with no sandbox — that is the £60k–£100k territory. Every integration is a failure mode.
Driver 4
B2C SaaS with no PII bar the email field — light touch. NHS data, FCA-regulated decisions, or SOC 2 in the next quarter — audit trails, residency, RBAC. That is production-tier work. No shortcuts.
Canarlo's transparent tiers
AI Readiness Assessment
£0
Free
Structured questionnaire, automated report, 30-min review call.
Right for the founder who has an idea but no engineering shape yet. Wrong if you already have a live LLM feature — go straight to the audit. You leave with a written brief. Implementation is not in scope.
AI System Audit
£8,000
Fixed fee · 2 weeks
Eval suite scaffolded against your live system, failure-mode report, prioritised fix list.
Right for teams with paying customers, an AI feature live for three months, and a technical founder who suspects the eval coverage is thin. Wrong pre-prototype. You leave with the harness, the report, and the fix list. Remediation is a separate engagement.
MVP build
£25,000–60,000
8–12 weeks
Production-ready system on your own cloud. AI engineering, web apps, or automations.
Right for founders who need a real system in customers' hands this quarter. Wrong for a six-month discovery. You leave with code, infra, docs, and a hand-over call. Multi-environment governance and on-call sit in the production tier.
Production build
£60,000–100,000+
16+ weeks
MVP plus observability, governance, multi-environment, on-call playbook.
Right for regulated workloads, board-visible launches, or anything where downtime costs revenue per minute. Wrong as a first engagement — start with the MVP. You leave with staging, prod, logs, traces, on-call runbook, optional retainer.
Deliverables
| Tier | Deliverable bundle |
|---|---|
| £0 AI Readiness Assessment | Structured questionnaire → automated PDF report → 30-min review call. Identifies high-value AI opportunities and biggest risks. |
| £8k AI System Audit | 2 weeks. Eval suite scaffolded against your live system. Failure-mode report. Prioritised fix list. One walk-through call. You keep all artifacts. |
| £25–60k MVP build | 12–16 weeks. Production-ready system on your own cloud (Vercel + Supabase). CI evals (AI projects) or full test suite (web apps / automations). Docs. Deployment. You own all code and infra. Hand-over walk-through. |
| £60–100k Production build | 16+ weeks. Everything in MVP plus: observability (logs/traces/metrics), governance policies, multi-environment (staging + prod), on-call playbook, optional retainer transition. |
Savings vs traditional agency
Equivalent scope, equivalent quality. The price gap is structural, not a margin trick. Big 4 teams ship from a blank repo every time — three sprints rebuilding the foundation, then the project starts. We ship from canarlo-core. The boilerplate is solved before kickoff.
| Dimension | Big 4 / large agency | Canarlo |
|---|---|---|
| Time | 6–12 months from engagement start | 12–16 weeks |
| Cost (equivalent scope) | £150k–£500k | £25k–£100k |
| Day rate (senior) | £1,500–£2,500 | £800–£1,200 blended (no junior offshore) |
| Lock-in | Vendor-controlled platform builds; multi-year licences | You own all code + infra; nothing is rented |
| Seniority | Layered teams; juniors do most of the work | Senior engineering on every line; canarlo-core does the boilerplate |
| Knowledge transfer | Often poor; "managed service" lock-in | Full docs + walk-through + optional Forge Care retainer |
| We're not the right fit for every engagement — if you need a global team of 200 with formal change-control, choose a Big 4. | ||
Gains vs in-house build
Long-term, in-house wins on unit economics. We will tell you that on the call. The trade is time. Hiring a senior AI engineer in 2026 takes six months, ramping takes three more. Canarlo buys the quarters back while your team gets built properly.
| Dimension | In-house build | Canarlo |
|---|---|---|
| Time to first production system | 9–18 months (hire → ramp → ship) | 12–16 weeks |
| Cost in first year | ≈£200k loaded comp for one senior AI engineer + ≈£100k for a second + opportunity cost | £25k–£100k one-time + optional Forge Care from £500/mo |
| Architecture risk | Greenfield; one engineer's opinions become law; failure modes unknown | Built on canarlo-core's proven foundation; failure modes pre-thought |
| Knowledge concentration | High — single engineer becomes a key-person risk | Docs + walk-through + retainer; team can read the code from day one |
| Hiring risk | 6+ months to fill senior AI engineering roles in 2026 | Zero hiring — engagement starts in 1–2 weeks |
| Once you cross ~£300k/yr in AI engineering spend, in-house starts to win on unit economics. Forge Care exists for the transition. | ||
Beyond the build
The build does not have to be the end of the relationship. Forge Care keeps your codebase in lock-step with canarlo-core. Quarterly or monthly regens. Security patches the day they ship. Your codebase becomes a subscription to current best practice, not a snapshot frozen at handover.
| Tier | Monthly | What's included |
|---|---|---|
| Forge Care Lite | £500/mo | Security patches; dependency upgrades; quarterly regen against latest canarlo-core; email support; uptime monitoring |
| Forge Care Standard | £1,500/mo | Lite + managed hosting (Vercel + Supabase); monthly regen; on-call within 1 business day; monthly maintenance call |
| Forge Care Pro | £3,000+/mo | Standard + dedicated engineer hours; same-day SLA; custom feature work absorbed against the retainer |
No other agency can offer Forge regens — your codebase stays in lock-step with our generator forever.
Know the number before you book the call.
Market rates
Big 4 — Deloitte, EY, PwC, KPMG — charge £1,500–£2,500 per engineer per day. The partner pitches, the senior architect scopes, the offshore junior writes the code. A £200k engagement gets you a slide deck and six months of weekly status calls.
Boutique London agencies start at £50k for a discovery phase that quietly becomes the product. Day rates £800–£1,500. The work is often good. The mark-up is paying for the Shoreditch office.
Offshore shops quote £15k for a six-month build and ship nothing you can run. No tests, no docs, no handover. The £15k saved costs £80k to rebuild.
We are not the cheapest. We are the studio that ships in twelve weeks because the foundation is already written.
Hidden costs
Hidden cost 1
Budget £200–£2,000 per month at MVP, £2k–£20k at production scale. Claude Sonnet beats GPT-4o on cost-per-task for most workloads. Cache aggressively. The first surprise bill is always retrieval calls firing on every keystroke.
Hidden cost 2
Logs, traces, metrics, eval dashboards. £100–£500/mo for Vercel + Supabase tiers, £1k+/mo if you add Datadog or LangSmith. We default to the cheapest stack that still tells you what broke. Resist the urge to platform-shop.
Hidden cost 3
Golden sets go stale in three months. Production sampling needs a weekly review. Budget two engineer-days a month or fold it into a Forge Care retainer. The teams that skip this learn from a customer first.
Self-assessment
If yes, start with the £8k AI System Audit. The findings size the next engagement properly. If no, the £0 Readiness Assessment shapes the brief before you spend a penny on engineering.
If yes, MVP tier is enough — your team picks up the docs and runs. If no, plan for the production tier plus Forge Care Standard, otherwise the codebase rots inside six months.
Twelve to sixteen weeks fits the £25–60k MVP. Six weeks means cut scope — pick one workflow, ship it, expand later. Quarters of patience usually means you should hire, not buy.
Generation-only evals fit the MVP budget. Retrieval evals plus calibration plus production sampling is production tier — the eval surface alone is two to three weeks of work.
Greenfield is cleaner and lands inside the MVP tier. Replacements add a migration phase, data backfill, and a parallel-run window — usually £15–25k on top, sometimes the production tier outright.
Not sure where you sit? Start with the free AI Readiness Assessment.
Why we publish prices
Most UK AI agencies hide their prices because their prices are reactive. They quote your perceived budget, not the work. A startup is told £40k. The same scope quoted to a bank becomes £400k. The deck is identical.
Our tiers are fixed because the work is repeatable. Canarlo-core ships the boilerplate. The bespoke part — your domain, your customers, your business logic — is what your money buys.
Publishing prices is also a qualification filter. If the number on this page is not the right number for you, neither of us wastes a discovery call finding that out.
See also: AI Engineering in the UK.
Start here
Twenty minutes, no slide deck — we tell you which tier fits and what the next step costs.