AI startup playbook.

London AI-native startups balancing product, infra cost and pipeline. The lever is treating operations like another model — instrumented, evaluated, automated — and proving unit economics under heavy GPU spend.

Claim a founding seat or book a demo →

Overview

A London AI-native startup looks like a SaaS company financially, but inference cost makes gross margin volatile week-to-week. Founders who do not instrument cost-per-feature and eval coverage typically discover both problems only after burning a tranche. This playbook covers operationalising a pre-seed to Series A AI company in London — entity setup, observability + eval stack, GPU/API procurement and a credible 12-month plan.

62% target

GM after inference

14h / week

Founder time saved

85% of prod paths

Eval coverage

№ 01 / What you'll need

Equipment list with real 2026 UK prices.

Sized for a 2–4 chair shop. Buy mid-range on chairs and clippers; cheap kit fails inside 12 months and walks away with your barbers.

Team + workstations (4–10 FTE)

Workstation per FTE (M3/M4 MacBook Pro or Linux)

£1,800–£3,800 each

Local dev GPU (optional, for inference R&D)

£1,800–£6,500 per workstation

Co-working membership

£280–£700 / desk / month

Model + inference stack

Frontier API spend (OpenAI / Anthropic / Google)

£800–£40,000 / month — scales with users

Open-source inference (Together / Replicate / Modal)

£300–£18,000 / month

Vector DB (Pinecone / Weaviate / pgvector self-host)

£0–£900 / month

Fine-tune + training credits

£500–£12,000 per training run

Eval + observability (LangSmith / Braintrust / Helicone)

£140–£900 / month

Product + infra

Cloud hosting (AWS / GCP / Vercel + DB)

£400–£8,000 / month

Auth + billing (Clerk / WorkOS + Stripe)

£90–£600 / month

Product analytics + session replay (PostHog / Mixpanel)

£0–£900 / month

Error + uptime (Sentry + Better Stack)

£70–£420 / month

SOC 2 + AI governance (Vanta + custom model card)

£6,000–£18,000 / yr

GTM + finance

CRM (HubSpot / Attio)

£18–£140 / user / month

Outbound (Clay + Apollo / Smartlead)

£140–£600 / month

Accounting + cap table (Xero + Vestd / SeedLegals)

£40 + £120–£480 / month

Payroll + HRIS (Pento / Onfolk / HiBob)

£8–£28 / employee / month

№ 02 / Where to set up

Location is half the business.

Format

Hybrid co-working + flexible all-hands

Most engineering is remote-leaning; sales + investor work needs London presence.

Tube proximity

Zone 1–2, ≤ 8 min walk

AI investor concentration sits in Mayfair / Soho / Kings Cross corridor.

Hub credibility

AI-cluster postcode (Kings Cross / Soho / Mayfair / Shoreditch)

Google DeepMind, Anthropic, Stability + most AI funds cluster here.

Power + cooling (if hosting local GPU)

Dedicated power circuit if running multi-GPU rig

Standard co-working circuits trip under sustained training load.

Soundproof booths

≥ 3 per 10 staff

Recorded demo + AI voice work needs quiet, callable space.

Area tier

Rent / month

Notes

Kings Cross / Granary Square (Zone 1)

£550–£900 / desk

Premium AI cluster (DeepMind, Anthropic London, Conjecture nearby). Best fundraising signal.

Soho / Mayfair (Zone 1)

£500–£850 / desk

AI VC corridor (Air Street, Hoxton, Index, LocalGlobe in walking distance).

Shoreditch / Old Street (Zone 1–2)

£380–£600 / desk

Sweet spot for product-led AI startups; lots of senior eng talent.

London Bridge / Borough (Zone 1–2)

£340–£550 / desk

Strong for AI in fintech / healthtech verticals.

№ 03 / Licences & registrations

Get the paperwork done before you sign a lease.

UK figures for 2026. Lead times assume you submit complete applications — councils will pause the clock if you miss documents.

Requirement

Who issues

Cost

Lead time

ReqIncorporate at Companies House

Companies House

£50

24 hours

ReqRegister Corporation Tax + PAYE + VAT (if applicable)

HMRC

Free

2–30 days

ReqICO data protection registration

ICO

£60–£2,900 / yr

Same day

OptSEIS / EIS Advance Assurance

HMRC

Free

6–12 weeks

OptEMI option scheme + share valuation

HMRC + accountant

£1,500–£4,500

6–10 weeks

ReqFounders + EMI agreements + articles

SeedLegals / specialist

£900–£4,500

2–6 weeks

OptAI Code of Practice / Bletchley Declaration alignment doc

Internal — DSIT signposted

Free

Internal review

ReqModel cards + sub-processor list (UK GDPR Art. 30)

Internal + DPO contact

Time only

Ongoing

OptSOC 2 Type I / Cyber Essentials Plus

Vanta / IASME certifier

£3,500–£18,000 / yr

8–16 weeks

ReqProfessional indemnity + cyber + AI liability rider

Hiscox / Superscript / Beazley

£900–£6,500 / yr

1–3 weeks

ReqEmployer's liability + public liability

Same broker

£280–£900 / yr

1 week

№ 04 / Build-out timeline

From decision to open in 12 weeks.

Phase 1 · Incorporate + capital

Month −6 to −3

Incorporate + open bank.
Issue founder shares, EMI scheme, apply SEIS AA.
Close pre-seed (typically £250k–£1.5m for London AI in 2026).

Phase 2 · Stack + design partners

Month −3 to 0

Set up product + eval + observability stack.
Recruit 2 engineers + 1 GTM + 1 designer.
Sign 3–5 design partners with monthly token budget caps.

Phase 3 · GA + unit economics

Month 0–6

Launch with metered Stripe billing.
Daily cost-per-active-user dashboard live in NAVIZIX.
Eval coverage to 70%+ of production paths.
First 30 paid workspaces.

Phase 4 · Margin + Series A prep

Month 6–18

Push GM after inference to 55–65% via prompt + cache + open-source mix.
Land 2–3 lighthouse enterprise pilots.
Weekly investor update auto-drafted.
£80k–£200k MRR + clear margin trajectory before A round.

№ 05 / Budget

Realistic startup and monthly cost ranges.

One-off · Startup

Setting up the shop

Incorporation + legal (founders / EMI / SEIS)

£900 → £4,500

Brand + website + design

£4,500 → £32,000

Team hardware + local GPU (4–6 staff)

£9,000 → £32,000

Year 1 product + eval + observability stack

£14,000 → £60,000

Year 1 frontier model spend

£12,000 → £280,000

Co-working year 1

£8,500 → £32,000

Insurance + Cyber Essentials + SOC 2 readiness

£3,200 → £24,000

Working capital / payroll buffer (6 months)

£120,000 → £600,000

Typical total

£35k → £85k

Recurring · Monthly

Keeping the lights on

Payroll (4–10 FTE incl. employer NIC + pension)

£28,000 → £110,000

Co-working

£800 → £4,500

Frontier API spend

£1,200 → £40,000

Open-source / GPU compute

£400 → £18,000

Cloud + DB + observability + eval

£600 → £8,000

GTM + finance stack

£400 → £2,400

Marketing (content + paid + events)

£800 → £12,000

Insurance + compliance (monthly equiv.)

£400 → £2,400

Typical total

£5.8k → £23k

№ 06 / Numbers that compound

How a ai startup actually looks on a chart.

Interactive projections rebuilt from real UK operating data — toggle the views to see ramp, mix and weekly load.

12-month revenue ramp

Low vs high scenario (£ monthly)

High Low

Source · NAVIZIX cohort · UK Seed/Series A AI-native startups · 2025–2026

№ 07 / First 90 days

Dates that actually matter after opening.

01Day 0

Stripe + token usage + analytics wired into NAVIZIX. Daily cost-per-active-user live.

02Day 14

First DNA scan — flags margin-after-inference and eval coverage.

03Day 30

First abuse / runaway-prompt alert tuned; free-tier caps in place.

04Day 60

First peer benchmark on GM after inference + cost-per-MAU.

05Day 90

Weekly investor update + model card snapshot ready for any reviewer.

№ 08 / What's inside

DNA, route, automations, integrations, benchmarks, advisors — tuned for a ai startup.

01Move

DNA dimensions

Active workspaces, token-spend per active user, gross margin after inference, eval pass-rate, model-cost drift, runway and review velocity.

02Move

Suggested route

Instrument token cost per feature, cap free-tier abuse, tighten eval coverage, then ship a customer-facing usage + value dashboard to lift retention.

03Move

Automations

Daily inference-cost vs revenue digest, abuse / runaway-prompt alert, eval-regression alert, weekly investor update and pipeline-to-burn projection.

04Move

Integrations

Stripe, OpenRouter / Anthropic / OpenAI / Replicate dashboards, LangSmith / Braintrust / Helicone, PostHog, HubSpot, GitHub, Xero.

05Move

Benchmarks

Gross margin after inference, token-cost per active user and eval pass-rate vs UK AI startups in the same vertical, refreshed weekly.

06Move

Advisors

Encore advisors with applied-AI deployment experience plus NAVIZIX AI for daily margin + eval nudges to the founders.

№ 09 / Stack

Modules used in this playbook.

01DNA Engine02Route Map03Launch Twin04Automation05Verified06Encore

Related playbooks

Owners of a ai startup often run — or cross-sell with — these adjacent operations.

Start with this playbook

14 days free. No credit card.

Full workspace, every module. The ai startup playbook loads on day one. Cancel anytime.

Private preview · Waitlist

3,751 operators in the queue

← All playbooks Talk to the founders →