SFAI Labs sells one thing: a 6-to-12-week idea-to-product engagement that takes one validated AI idea from a written PRD through a deployed product, billed at three named milestones, for a bundled fee around $150K total. This page is the price card. Three packages — Planning Week ($20–$30K), MVP Build ($80–$120K), Hardening + Handoff (~$30–$50K) — with named deliverables per milestone, explicit exclusions, a payment trigger schedule, and a documented founder-ownership matrix at handoff.
We publish the numbers so procurement-stage founders walk into a discovery call already knowing what the engagement costs and what they own at the end. If the shape fits, the next step is a 30-minute idea review. If not, we point to the adjacent category — agency, fractional CTO, or DIY — where your idea lives.
The three packages at a glance
| Package | Calendar window | Range | Payment trigger | Headline deliverable |
|---|---|---|---|---|
| Planning Week | Week 1–2 | $20K–$30K | On SOW signature | Eval-anchored PRD + capability-feasibility memo + 30–80 case eval set |
| MVP Build | Week 3–8 | $80K–$120K | On Planning Week acceptance | Deployed product with payment flow + 5–15 invited paying users |
| Hardening + Handoff | Week 9–12 | $30K–$50K | On MVP Build acceptance | Auth, observability, regression suite, runbook, 30-day on-call SLA |
| Bundled | Week 1–12 | $150K typical, $130K floor, $200K ceiling | Three milestone invoices | All of the above, sequenced |
The bundled rate is the common purchase shape — roughly 4 in 5 engagements take all three packages in sequence. The Planning Week is the only package sold standalone; we do not start an MVP Build without a Planning Week output. These ranges sit inside the broader 2026 AI MVP market range we documented in the AI MVP economics playbook and triangulate against publicly posted ranges from Designli, Apexon, and Markovate.
Package 1 — Planning Week
Range: $20K–$30K. Typical $25K. Calendar: Weeks 1–2 (10 working days). Payment trigger: On SOW signature.
The most under-priced phase in every competing service category, deliberately the most expensive labor in ours. Two senior people — a founding engineer and a product lead — spend two full weeks on one idea.
Deliverables:
- A 12-to-25-page PRD with a named user persona, a single primary task, an acceptance rubric, and three priority cuts (what ships, what defers, what is out of scope).
- A capability-feasibility memo that scores the chosen frontier model — Claude Opus 4.8, GPT-5, or Gemini 2.5 Pro — against the rubric on representative inputs, with cost-per-query estimate.
- A 30-to-80 case eval set, hand-graded by two graders independently, with inter-rater agreement above 80% on the headline rubric metric.
- A buy-vs-build memo for any third-party dependency surfaced during scoping (vector DB, observability stack, auth, payments).
- A signed-off MVP Build SOW with a fenced fee inside the Package 2 range.
Acceptance criterion: the founder reads the PRD unaided and answers three questions correctly — who is the user, what is the one task, what is the eval threshold for ship. If not, we extend the Planning Week inside the original fee.
Price drivers: regulated data (HIPAA-touching, financial), more than one frontier-vendor candidate in the capability probe, or a custom eval rubric requiring a domain expert push toward the ceiling. A founder who arrives with a draft PRD and 15–30 written example cases keeps it at the floor.
The Planning Week is also the cleanest off-ramp. If the feasibility memo concludes the idea is not buildable to rubric on a current frontier model, the founder walks away $25K poorer with a decision artifact and zero $150K commitment.
Package 2 — MVP Build
Range: $80K–$120K. Typical $95K. Calendar: Weeks 3–8 (6 working weeks). Payment trigger: On Planning Week acceptance (SOW signature for Package 2).
Six weeks from PRD to a deployed product on its own domain, with a payment flow and 5–15 invited paying users completing the core task. Not a demo. A product the founder can hand to a customer.
Deliverables:
- Application code in a repo transferred to the founder’s GitHub organization at end of package, with full IP assignment in the SOW.
- A live deployed product on a founder-owned domain — TLS, basic auth, payment integration (Stripe default, alternatives by request).
- A versioned prompt library — every production prompt in source — with model-version pinning and a documented swap procedure.
- The Package 1 eval set wired into CI as a regression suite. Every commit triggers eval-set pass/fail.
- A model contract on the founder’s billing. The founder owns the Anthropic, OpenAI, or Google API key from week 3; cost-per-query lives on founder books from day one.
- A first cohort of 5–15 paying users invited and onboarded — the founder’s contacts, not ours.
Acceptance criterion: at end of week 8, one paying user completes the core task end-to-end with no engineer in the loop, and the eval suite passes at the Planning Week rubric threshold. Both, not either.
Price drivers: a second frontier-model dependency requiring routing logic ($15–25K), custom data ingestion beyond file or API upload ($10–20K), or a non-web surface — native iOS or a Slack/Teams integration ($20–35K). One model, web-first, and standard payments keep it at the floor.
For an alternative pricing-model comparison, see fixed-price vs hourly vs milestone AI MVP pricing. The milestone shape we use is what makes the price a fence, not a floor.
Package 3 — Hardening + Handoff
Range: $30K–$50K. Typical $40K. Calendar: Weeks 9–12 (4 working weeks). Payment trigger: On MVP Build acceptance (SOW signature for Package 3).
The four weeks between “works in demo” and “runs at 3 a.m. without paging anyone.” Where the engagement earns the second half of its fee. Most procurement objections to AI MVPs trace to founders who skipped this phase.
Deliverables:
- Production auth (session management, password reset, basic RBAC) and rate-limiting on every model-touching endpoint.
- An observability stack — structured logging, latency and cost-per-query dashboards, model-output capture — sized to founder operations.
- A runbook a non-engineer founder can execute: read the dashboard, triage a model regression, swap the model version, roll back a deploy.
- An eval-regression CI gate that blocks any commit dropping the rubric metric below threshold.
- A 30-day post-handoff on-call SLA — Slack channel, business-hour response, one model-version-swap support session.
- A handoff session in which the founder, with the team observing but not driving, swaps the frontier model from one version to a newer release and re-runs the eval suite.
Acceptance criterion: the product runs unattended for 7 consecutive days with no manual incident, the runbook is exercised once by the founder while we observe, and the founder-driven model-version swap completes inside one hour. All three.
Price drivers: SOC 2 starter posture ($15–25K added, with the audit fee separate), regulated-data observability (PHI scrubbing, audit logging), or multi-region deployment push toward the ceiling. Single-region, no audit, and standard observability keep it at the floor.
Bundled rate and payment milestone schedule
The bundled engagement runs $130K floor, $150K typical, $200K ceiling for the full 12-week shape. Three milestone invoices, sequenced.
| Invoice | Trigger | Amount (typical $150K example) | Net terms |
|---|---|---|---|
| Invoice 1 | SOW signature for Planning Week | $25K | Net 7 |
| Invoice 2 | Planning Week acceptance (week 2) | $95K | Net 14 |
| Invoice 3 | MVP Build acceptance (week 8) | $40K | Net 14 |
Acceptance is documented in writing — a signed Acceptance Memo per milestone. If acceptance does not land by the calendar deadline, the next invoice does not issue; we resolve the gap inside the existing fee, or re-scope before invoicing if the gap is buyer-side scope drift.
Three things this schedule will not do: bill on calendar dates without an acceptance memo; stack invoices on a fixed schedule regardless of progress; or take a 50% deposit on the full $150K up front — that pattern transfers risk to the founder and signals an agency model misaligned with the milestone bill.
The economics behind this shape are in our case for milestone-based AI MVP billing.
What’s included across all packages
Across the three packages, the engagement bundle includes:
- Two senior people for 12 weeks — a founding engineer and a product lead, each at 70–90% allocation.
- Frontier-model API access during Planning Week (we operate against our keys for the capability probe; the founder pays Anthropic / OpenAI / Google directly from week 3).
- All design, prompting, and architecture decisions — no white-label sub-contracting.
- One round of week-9 user-research synthesis — five user interviews, written up as a Phase 2 priority memo.
- A founder-runnable three-slide product summary for the investor deck.
- Office hours during the 30-day on-call window — one hour per week with the founding engineer.
What’s excluded
The fence is the product. Anything below is excluded by default and gets quoted as a Phase 2 SOW after week 12.
- A 24-month roadmap. The engagement scopes to one product, one persona, one task. Multi-persona platforms get the wrong delivery shape.
- Frontier-model inference at scale. From week 3, the founder owns the Anthropic / OpenAI / Google bill. Typical MVP-cohort cost-per-query runs $0.01–$0.15 — we model this in the Planning Week memo.
- A marketing site beyond the product surface. Landing pages, blog, SEO, email marketing are out.
- A second frontier-model integration not flagged in Planning Week. If we re-scope, we re-scope before adding.
- A SOC 2 Type II or HIPAA audit inside the window. Starter posture is a $15–25K Hardening addendum; the audit itself is separately contracted with the assessor of choice.
- Enterprise sales support, contracts review, hiring search. Out of scope.
- Native mobile. Web-first by default; iOS / Android as a Phase 2 engagement.
- Long-tail post-launch feature work. The 30-day on-call covers incidents and model upgrades, not new features.
A defensible price is a fenced price. If a procurement counsel asks where the fence sits, point at this list.
Founder ownership at handoff
| Layer | Owner at week 12 | Notes |
|---|---|---|
| Application code | Founder | Repo transferred to founder’s GitHub org; full IP assignment in the SOW |
| Prompts and prompt templates | Founder | Versioned prompt library in the repo; no held-back proprietary library |
| Eval set + rubric | Founder | The procurement-anchored deliverable; carries through every future model upgrade |
| Runbook + on-call docs | Founder | Written to be operable by a non-engineer founder |
| Model contracts (Anthropic, OpenAI, Google) | Founder | Founder’s billing from week 3 |
| Customer relationships + data | Founder | Direct contracts; no intermediary |
| Model weights | Frontier vendor | Structural reality, not negotiable |
| SFAI Labs-internal templates | SFAI Labs | No transfer value to the founder; used internally to deliver |
The cleanest test of a handoff: at week 12, the founder swaps the underlying frontier model from one version to a newer release without us driving. If that test fails, we are still inside the engagement. For the broader category framing, see idea-to-product as a service: what you get, what it costs, what you ship.
How to start
Four steps, two weeks if everything lands.
- 30-minute idea review (free). Bring a one-page idea brief, a named user persona, and a back-of-envelope budget. We say whether the idea-to-product shape fits, and if not, which adjacent category to procure. Book the review.
- Paid scoping call (optional, $2K). Two hours, two senior people, deep on the rubric and the frontier-model feasibility question. Output is a written Planning Week SOW.
- Planning Week SOW signature. Invoice 1 issues ($20–30K Net 7). Week 1 starts the following Monday.
- Three-package SOW. Issued at end of Planning Week. Signature triggers MVP Build kickoff the following Monday.
Two pages worth reading before the discovery call: the AI MVP economics playbook and the master idea-to-product manifesto. For year-1 TCO, see the 7 cost lines most CFOs miss in AI projects.
Book the 30-minute idea review.
FAQ
How much does an SFAI Labs idea-to-product engagement cost?
$130K floor, $150K typical, $200K ceiling for the full 12-week bundle: Planning Week ($20–30K), MVP Build ($80–120K), Hardening + Handoff ($30–50K). The Planning Week is the only standalone package. Every fee is fenced at SOW signature — no T&M overruns.
Can I buy only the Planning Week?
Yes. The Planning Week is the only standalone package. Roughly 1 in 5 engagements buy it alone and contract MVP Build and Hardening as a separate SOW two weeks later. Roughly 1 in 12 Planning Weeks conclude the idea is not yet buildable to rubric on a current frontier model — the founder walks away $25K poorer with a decision artifact instead of a $150K build commitment.
What is the payment schedule?
Three milestone invoices. Invoice 1 ($20–30K Net 7) on Planning Week SOW signature. Invoice 2 (MVP Build fee, Net 14) on Planning Week acceptance at week 2. Invoice 3 (Hardening fee, Net 14) on MVP Build acceptance at week 8. No 50% deposit on the full bundle. No invoice issues without a signed Acceptance Memo for the prior milestone.
What happens if my scope changes mid-engagement?
The Planning Week PRD is the contract. After week 2, any change either fits original scope, fits with a defined cut elsewhere, or becomes a Phase 2 SOW after week 12. If the idea is still moving in week 4, we pause MVP Build and re-run part of Planning Week inside the existing fee.
Who owns the code, the prompts, and the eval set?
The founder owns application code (full IP assignment, transferred to the founder’s GitHub org at week 12), the prompt library, the eval set, the runbook, and the model billing relationship. Nobody owns model weights — that is the frontier vendor’s structural reality, not a contract negotiation point.
What’s NOT included in the bundled fee?
A 24-month roadmap, frontier-model inference at scale beyond the Planning Week probe (founder owns the API bill from week 3), a marketing site, a second frontier-model integration not flagged in Planning Week, a SOC 2 or HIPAA audit, enterprise sales support, native mobile, and post-handoff feature work beyond the 30-day on-call.
How does SFAI Labs pricing compare to AI agencies?
Posted AI agency MVP ranges (Designli, Apexon, Markovate) run $50K–$500K with concentration around $150K–$300K. The structural difference is the milestone bill: an agency at $400K over 9 months has $400K of optionality to expand scope; our 12-week engagement has zero. The fence is the product, not the calendar.
How does it compare to hiring a fractional CTO?
A fractional CTO at $5–15K per month buys advice, not a shipped product. Across 12 weeks the spend can look comparable on the low end ($15–45K), but the deliverable is different — a shipped product with paying users vs. a strategy document. See idea-to-product vs hiring a CTO.
Do you offer discounts or equity-for-fee swaps?
No discounts on the package floors. Equity-for-fee swaps are case-by-case and only on Hardening + Handoff — we do not take equity for Planning Week or MVP Build, because the milestone-billed discipline depends on the fee mattering on both sides.
What happens after week 12?
Two clean post-engagement options. A Phase 2 SOW scoped fresh against the M2 product reality, available 4–8 weeks after handoff. Or a low-touch on-call retainer at $5–15K per month for incident response and model-upgrade support. Extending the existing M4 engagement does not work — the milestone discipline collapses into agency T&M.
Arthur Wandzel
