The single highest-impact decision in a 6-week AI MVP is not what to build — it is what to defer. McKinsey’s 2025 “State of AI” survey reports that 80–85% of enterprise AI pilots never reach production, and the dominant failure mode is not model quality. It is scope inflation: founders include the admin panel, the second feature, multi-tenant, polished UI, analytics, billing, SSO, and a public API in the first build — and the calendar slides from 6 weeks to 14, the budget from $80K to $180K. The cheaper path is not a cheaper engagement; it is a different scoping decision. This piece names 8 features to defer, the structural reason each defers, the trigger for adding it post-MVP, and a side-by-side worked trade.
The framework builds on the AI MVP economics playbook. It pairs with the 6-week scope and milestones piece and the case for the smallest possible AI feature. The broader program frame is the idea-to-product manifesto.
Why a deferral framework matters
Founders over-scope for a predictable reason: every feature on the cut list feels like it earns its keep. The admin panel “saves engineering time later.” The second feature “differentiates us.” Multi-tenant “we will need it for customer #2 anyway.” Each addition is locally rational; the compounded result is a 14-week build, a $180K invoice, no production traction, and an eval surface a two-person team cannot keep current.
A defer-or-include framework reverses the default. Instead of “what should we cut?”, it asks “what proves itself worth including inside 6 weeks?” The threshold is high on purpose. AI builds carry a multiplier non-AI builds do not: every feature adds prompt versions, eval cases, fallback paths, and observability surface. A second feature is not 2× the work; it is closer to 2.6×.
The 8 items below are the features founders most often want to include and that most often turn a 6-week build into a 14-week one. None are wrong forever. All are wrong now.
The 8 features to defer
Each item follows the same pattern: founder’s instinct for urgency, the structural reason it defers, the named trigger that flips deferral into action, and a defensible cost-of-inclusion range. Ranges assume US senior-engineer rates at $200–250/hour and a single-feature scope.
1. Admin panel
Defer. Use a database client and a runbook.
- Why it feels urgent. Founders want a UI to manage users, refund payments, force re-runs, inspect failures — operational pain felt before it arrives.
- Why it isn’t yet. Inside 6 weeks, the engineer who would build the admin panel is also the engineer who would use it. Direct database access plus a runbook covers the role at a fraction of the cost.
- When to add. When a non-engineer (CS, support, ops) is hired and needs routine corrections without engineering involvement.
- Cost of including in 6 weeks. $12K–18K and one calendar week. A stripped-down dashboard with user list, search, manual overrides, audit log. RBAC adds $6K–10K.
2. The second feature
Defer. Ship one capability, observe, then decide.
- Why it feels urgent. Founders fear the first feature is too thin to justify a paid SKU and want a second to “round out the offering.”
- Why it isn’t yet. A second feature multiplies eval surface, prompt-management overhead, failure-mode catalog. Work is closer to 2.6× because evals double, fallbacks branch, dashboards disaggregate. The deeper cost: the founder cannot learn which feature is the actual wedge.
- When to add. After 4–8 weeks of post-launch data when conversion or retention signal points to a specific unmet need.
- Cost of including in 6 weeks. $25K–40K and 2–3 calendar weeks.
3. Multi-tenant architecture
Defer. Run single-tenant until tenant #2 signs.
- Why it feels urgent. Multi-tenant reads as “scalable architecture,” and founders worry single-tenant becomes a costly migration later.
- Why it isn’t yet. Multi-tenant adds 2–3 weeks of platform work (tenant isolation, per-tenant config, data partitioning, auth scoping) for an audience of one customer. The eventual migration cost is bounded (1–2 weeks) and is paid against revenue, not runway.
- When to add. When the second customer signs a contract. Pipeline confidence is not a billable event.
- Cost of including in 6 weeks. $20K–30K and 2–3 calendar weeks. Tenant-aware data layer, scoped auth, per-tenant config, tenant-aware observability.
4. Polished UI
Defer. Ship a defensible-but-plain UI. Polish after usage signal.
- Why it feels urgent. Founders want to demo well, want a premium first impression, and worry that “ugly” UI sandbags adoption.
- Why it isn’t yet. Polished UI is a commitment to a UX hypothesis the founder has not earned. Real users reshape interfaces in ways no one predicts in week 2. Polish before usage signal is polish wasted.
- When to add. After 50–200 real users and a conversion or activation funnel surfacing specific UX friction.
- Cost of including in 6 weeks. $15K–25K and one calendar week. Design system, illustrative assets, micro-interactions, accessibility passes.
5. Analytics
Defer. Ship event logging plus a SQL workbook. Skip the dashboard.
- Why it feels urgent. Founders want to “be data-driven” and assume a 6-week MVP without a dashboard is operating blind.
- Why it isn’t yet. Event logging to a warehouse table plus a shared SQL workbook gives the same answers — DAU, activation, retention — at a fraction of the cost. A vendor product (PostHog, Mixpanel, Amplitude) instrumented in an hour gives more.
- When to add. When a non-technical team member needs self-serve access — typically a marketer or PM hired post-launch — or when SQL queries take more than 30 minutes a week.
- Cost of including in 6 weeks. $10K–18K. Chart components, time-range pickers, cohort logic, metrics layer.
6. Billing
Defer. Invoice manually until customer #10.
- Why it feels urgent. Founders associate “real product” with “self-serve checkout” and want to feel commercially serious.
- Why it isn’t yet. Stripe Checkout is fast to wire but slow to fully implement — subscription state, dunning, proration, invoicing, tax, refunds, edge-case admin tools. A Stripe payment link plus a Google Sheet plus a manual invoice on the 1st is faster, cheaper, and yields direct revenue-conversation data the self-serve flow obscures.
- When to add. When the founder is spending more than 4 hours a week on manual billing, or when customer #10 signs.
- Cost of including in 6 weeks. $15K–25K and 1.5 calendar weeks. Subscription state, plan switching, dunning, invoicing, webhooks, basic admin tooling.
7. SSO (Single Sign-On)
Defer. Use email-plus-password until the first enterprise buyer demands SSO.
- Why it feels urgent. Founders selling to companies assume SSO is table-stakes and worry that lack of SSO blocks enterprise deals.
- Why it isn’t yet. SSO is a buyer-driven feature, not a builder-driven one. SAML and OIDC done properly take 1.5–2.5 weeks and are rarely the deciding factor in a first close.
- When to add. When a specific enterprise buyer makes SSO a contractual condition. Build it then, often against pre-paid revenue.
- Cost of including in 6 weeks. $18K–28K and 1.5–2.5 calendar weeks. IdP integrations (Okta, Azure AD, Google Workspace), JIT provisioning, per-buyer config tooling.
8. API (public or external)
Defer. Ship the product surface first. Externalize when a partner asks.
- Why it feels urgent. Founders dream of platform reach — partners integrating, marketplaces forming. An API makes the product feel infrastructure-grade.
- Why it isn’t yet. A public API is a contract with the outside world: versioning, docs, rate limits, auth, error semantics, SDKs — 2–3 weeks competing with the product surface most early users actually touch. Pre-PMF, the API has zero callers.
- When to add. When a named partner has a concrete integration use case and is willing to pilot a stable interface.
- Cost of including in 6 weeks. $18K–28K and 2–3 calendar weeks. Versioning, auth keys, OpenAPI docs, rate limiting, SDK stubs.
Worked sample: $80K vs $180K
The same idea — a single AI capability for a B2B audience — built two ways. Both are real engagements at defensible 2026 rates. Numbers anchor to the per-item ranges above and to the $250K AI project budget decomposition.
| Line item | Deferred ($80K, 6 weeks) | All-included ($180K, 14–16 weeks) |
|---|---|---|
| Core feature (planning, eval, build, hardening) | $65K | $65K |
| Handoff and runbook | $10K | $10K |
| Inference and infra (6 wks) | $5K | $5K |
| Admin panel | — | $15K |
| Second feature | — | $32K |
| Multi-tenant architecture | — | $25K |
| Polished UI | — | $20K |
| Analytics dashboard | — | $14K |
| Billing flow | — | $20K |
| SSO | — | $22K |
| Public API | — | $22K |
| Subtotal | $80K | $250K |
| Less: integration efficiency | — | −$70K |
| Total | $80K | $180K |
| Calendar | 6 weeks | 14–16 weeks |
The integration-efficiency line reflects roughly 28% off the additive total for doing the extras in one engagement rather than two. The savings are real but do not change the strategic question.
What the cheaper path teaches that the expensive path obscures: the founder shipping the $80K version reaches first usage in 6 weeks and gets unfiltered signal on the one feature. The founder learns whether the admin panel is needed (often no — the customer base is 5 people who email support), whether multi-tenant is needed (often no — tenant #2 has not materialized), whether SSO is needed (often no — the first 10 customers use email-and-password).
The founder shipping the $180K version reaches first usage in 14–16 weeks having already paid for admin, SSO, multi-tenant, billing, polish, and API — most serving zero users for the first 90 days. When users churn, the founder cannot tell whether the core AI feature missed the mark or whether the broader surface created friction. More product, less learning.
This is the trade. Not “spend less” — “learn first, then spend.” The deferred path keeps $100K of budget alive for what real signal tells the founder to build. For runaway-project root causes, see the 5 cost-side root causes; for eval-anchored budgeting, see budget AI projects in eval runs.
3 founder mistakes when applying this framework
Mistake 1: Treating the framework as a menu
Founders read the list and want to defer six items and include two. The framework is not a menu — it is a default. Each item is presumed deferred unless the founder can name the specific trigger condition that is already true today. “We will need SSO because enterprise” is not a trigger. “Acme Corp has SSO in the MSA we signed Friday” is. Deferral is the default; inclusion is the exception that must be defended in writing.
Mistake 2: Hidden inclusions via the back door
A founder defers the admin panel, then asks the engineer to “just add a quick user search page.” The user search page is the admin panel under a different name. Hidden inclusions are the single most common cause of 6-week slips. The countermeasure is a written exclusion list signed at week 1 — the 8 deferred items by name — and a written change-request process for any addition during the build.
Mistake 3: Confusing deferral with abandonment
Deferral is sequencing, not subtraction. The framing for investors and buyers is “MVP-1 ships the core feature; MVP-2 picks up admin, billing, and SSO based on the trigger conditions we have defined.” Named triggers are the credibility move — they signal structural thinking about sequencing rather than corner-cutting.
Download the AI MVP Scoping Worksheet
The 8-item deferral framework, named triggers, defensible cost ranges, and the worked $80K vs $180K trade are the core of the AI MVP Scoping Worksheet — a single PDF a founder takes into a vendor conversation to score scope decisions in real time.
FAQ
Is a 6-week AI MVP actually feasible in 2026?
Yes — for one narrow feature wrapped in five supporting layers (eval, fallback, observability, prompt versioning, handoff) and with all 8 items here deferred. The window collapses the moment a second feature, admin panel, or multi-tenant requirement enters scope. Any wider scope is 9–12 weeks at $120K–180K, documented in the AI MVP economics playbook.
Why is the second-feature multiplier 2.6× and not 2×?
A second AI feature doubles the eval set (50–200 labeled cases each), branches the fallback logic, disaggregates the observability dashboard, and complicates prompt versioning. Added overhead is roughly 60% of a single-feature build, yielding 2.6×. Non-AI builds do not carry this surface, so the CRUD-app multiplier is closer to 1.8×.
What if my enterprise buyer demands SSO during the 6 weeks?
That is the named trigger. Build it then, scoped against the contract value. The framework is not anti-SSO; it is anti-speculative-SSO. The clarifying question is whether SSO is a contractual condition with a signed buyer or a feature added in case a future buyer asks.
How do I tell the team that polished UI is deferred?
Show the team the worked sample. The $100K delta between the deferred and all-included paths funds 4–6 months of operating runway. The premise is not “no design” — it is “no decorative design until real users tell us what needs to change.”
What about security and compliance?
Security baseline (OWASP LLM Top 10, secret management, input validation, audit logging) is in scope inside 6 weeks. SOC 2 readiness is not — it is a 3–6 month workstream. If a buyer signs a contract requiring SOC 2 or HIPAA inside the MVP window, the engagement is 9–12 weeks, not 6.
Can the deferred features be added piecewise after launch?
Yes, and that is the point. Deferred features are not “never” — they are “after the trigger fires.” Most are 1–2 week additions when prioritized correctly. The piecewise post-MVP roadmap is documented in the AI MVP economics playbook.
How do I justify the deferral framework to investors?
Investors who understand AI builds know the 80–85% pilot-stall rate and read scope discipline as a credibility signal. The framing “we deferred 8 items to MVP-2 with named triggers so we could ship in 6 weeks and learn from real usage before allocating the next $100K” lands well.
What happens if I include 2–3 items anyway?
The calendar slips proportionally. One added item adds 1–2 weeks; two add 3–4; three make the engagement a 10–12 week build. Including one of the 8 does not break the framework — it breaks the calendar.
Where does this sit in the broader idea-to-product program?
A decision-support piece within the broader idea-to-product program. The foundational guide is the AI MVP economics playbook, which sits inside the idea-to-product manifesto. The in-scope companion is the 6-week AI MVP scope and milestones piece; the discipline upstream is the case for the smallest possible AI feature.
The deferral framework is not a cost-cutting exercise. It is the structural decision that decides whether the 6-week MVP becomes a learning instrument or a $180K hypothesis. The 8 items defer for the same reason: each feels urgent, each is locally rational, and each — added inside 6 weeks — converts an AI MVP into something the team cannot keep current and the founder cannot read. Defer the eight, build the one, learn from real usage, then decide which of the eight earns its way back in.
Arthur Wandzel
