ProjekinAja. Founder's Playbook Stage 2 · Build The Right Thing Lesson 2.8
● Stage 2 · Lesson 2.8

From MVP to Product

The MVP is not the prototype and it is not the product — it is the minimum set of functionality that lets a real customer do a real job in their real workflow.

⏱ ~22 min read ⚙ Feature voter widget 🏭 Enterprise-specific

🎯 By the end you'll understand

The structural difference between an MVP-for-learning and an MVP-for-production, and why the transition is not just adding features. The enterprise-specific requirements that must be present before any manufacturing customer can deploy your product in a real workflow: data security, audit logging, uptime SLA, change management support, and integration requirements. How to write a one-page product brief that specifies outcome without over-constraining implementation. And why the most dangerous move in B2B SaaS is announcing general availability before you are actually production-ready — one failed deployment at a reference customer can close the market for 12 months.

KEY INSIGHTThe hardest mile is between "it works" and "it ships"

The prototype proved the idea. The learning MVP proved the demand. The production MVP proves you can operate — and operating in an enterprise environment is 80% things that have nothing to do with your core feature.

You have a working edge vision sensor. Your pilot customer at a Wisconsin stamping plant ran it for 60 days and saw defect detection improve by 37% over baseline. The dashboard loads. The alerts fire. The model is solid. You are six weeks from calling this "generally available." And you are probably about to make the most dangerous mistake in the Stage 2 arc: announcing readiness before you have earned it.

🔑 The transition nobody teaches

Learning MVPs are designed to be disposable — you build the minimum to generate a verdict, then rebuild with what you learned. Production MVPs are designed to be operated — they must survive real infrastructure, real compliance requirements, real IT organizations, and real user behavior at scale. The features that make a learning MVP a great experiment are often orthogonal to the features that make a production MVP deployable. The gap between the two is not a roadmap item. It is a different category of work.

CORE CONCEPTMVP-for-learning vs. MVP-for-production — not a spectrum, a phase change

The distinction matters because the two artifacts have different primary success criteria. Getting this wrong is how technically excellent products fail in enterprise markets: the team treats deployment as a feature rollout rather than a production readiness milestone.

Dimension MVP-for-learning MVP-for-production
Primary question Does this create value for at least one real customer? Can this be operated reliably by a customer team without heroics from us?
Success metric Validated learning: a behavioral signal that confirms or falsifies the hypothesis Customer successfully completes their core job without founder intervention in <30 days
Acceptable failures Downtime, rough edges, missing features — if the learning is captured, the failure is fine Zero downtime at critical job steps; any failure during core job is a production incident
Who operates it You (the founder), possibly on-site, debugging in real time The customer's operations team, without you in the room
Integration requirement Standalone acceptable; MES/ERP hooks optional Must integrate with at least the core data flow the customer already uses
Compliance bar Pilot agreement covers data handling informally Security review, data processing agreement, audit logging, uptime SLA must be in place

The operational implication: when you decide to transition from learning MVP to production MVP, you are not adding a feature. You are changing the operating model. The question is no longer "does this create value?" but "can this be trusted to create value reliably, without us, in a real production environment?" That is a materially different bar, and it requires a deliberate gate.

Prototype Proves the idea works technically Learning MVP Proves demand & generates a verdict GATE Production Readiness Review checklist Production MVP Operates reliably without founders present Q: Does it work? Q: Does anyone want it? Q: Can they use it themselves? don't skip me
The three phases and the gate between them. Most teams skip the production readiness gate. The consequence is deploying a learning MVP as a product and losing the reference customer when it fails under real operating conditions.

HOW IT WORKSEnterprise production requirements — the five categories you cannot skip

Manufacturing customers operate in regulated, auditable, safety-conscious environments. These are not features they will accept later — they are preconditions for signing a production contract. Each of the five categories represents a class of failure mode that has killed B2B hardware/software products at the deployment stage.

1

Data security and data sovereignty

Manufacturing vision data is treated as highly sensitive intellectual property — the defect patterns, production volumes, and product geometries your sensor captures are competitive intelligence. Enterprise customers will ask: Where is the data stored? Who can access it? Can it be stored on-premises or in a private cloud? What encryption standards are used in transit and at rest? What happens to the data if the contract ends? These questions are not paranoia — they are standard IT procurement. If you cannot answer them with a documented data processing agreement, you will not pass the security review. Forge's architecture decision (edge compute + encrypted telemetry to customer-controlled cloud or on-prem server) must be documented before the first production contract is signed, not after.

2

Audit logging and traceability

In manufacturing, the inspection record is a legal document. When a customer ships a part that later fails, they need to prove what their inspection system said about that part on the day it was inspected. This means your system must log every detection event with a timestamp, a result, a confidence score, and the model version that produced it — immutably, in a format the customer's QMS can export. This is not a "nice to have" — it is a contractual and regulatory requirement in any ISO 9001/IATF 16949 environment. A learning MVP that produces accurate detections but does not log them is not deployable in production.

3

Uptime SLA and failure graceful degradation

A production line cannot stop because your software crashed. The SLA question is: what happens when the edge device goes offline, loses its network connection, or encounters an anomaly outside its training distribution? The acceptable answer is almost never "the line halts" or "the alert fires every 3 seconds." The acceptable answer is usually: the device fails to a defined safe mode (pass-through with a flag, or halt with a clear error code), the ops team is notified, and the SLA specifies maximum acceptable downtime with a remediation process. For a first production deployment, 99.5% uptime with a 4-hour response SLA is a reasonable starting point. Make it explicit and contractual before you ship.

4

Change management support

The most underestimated production requirement. Your technology is not just competing with the previous inspection method — it is asking the plant to change every workflow that touches quality inspection. Line workers need training. Supervisors need new escalation protocols. The QA manager needs a new reporting structure. The IT team needs documentation for the network and firewall configuration. If you ship a technically perfect product without change management materials, your champion at the plant will spend three weeks answering the same questions from their colleagues instead of running the production deployment. Provide: an onboarding runbook, a training deck for line operators, an integration checklist, and a support escalation path. These are not marketing materials — they are what makes deployment succeed.

5

Integration with existing tooling

Manufacturing plants run MES (Manufacturing Execution Systems), ERP systems, and QMS platforms. Your defect detection outputs become useful in production only when they flow into the systems that already govern the plant's workflow — work orders, non-conformance reports, production dashboards. Integration does not necessarily mean full API connectivity on day one. But it does mean having a defined data export format (usually CSV, XML, or a specific MES-compatible schema), a pathway for the customer to pull your data into their existing systems, and a clear roadmap for deeper integration. "We'll figure out integration later" is what learning MVPs say. Production MVPs have an integration story before the first contract.

⚠ The reference customer failure mode

In manufacturing, procurement decisions are heavily influenced by peer references. If your first production deployment fails — the device logs correctly but the data format does not import into the customer's MES, or the device goes offline during an audit and the QA manager cannot produce the inspection record — the reference customer becomes an anti-reference. In a concentrated vertical (Tier-1 auto-parts manufacturers in the Midwest, for example), the buyer networks are tight. One bad story travels fast. The 12-month market-closure risk from a failed first deployment is not hyperbole — it is the documented pattern in industrial software sales.

IN PRACTICEThe one-page product brief — specifying outcome without over-constraining implementation

The product brief is the artifact that transitions from "what we learned from the MVP" to "what we are committing to build for production." It is one page because more than one page usually means you are specifying implementation rather than outcome — the most common failure mode for technical founders writing requirements documents.

A production-ready product brief has five components and nothing else:

📋 The five-component product brief

1. Job to be done (one sentence): The specific, measurable job the customer is hiring this product to do. Not "detect defects" — "enable a line quality engineer to identify and route failing stampings before they reach the assembly station, without stopping the line."

2. Success criteria (3–5 measurable outcomes): Stated as customer outcomes, not feature descriptions. Example: "Customer completes first shift review in under 15 minutes using only the dashboard. Detection rate exceeds current baseline by ≥20% at ≤0.5% false-positive rate. Zero production line stops attributable to device failure in 30 days."

3. Constraints (hard and soft): Hard constraints are non-negotiable (e.g., "must not require internet connectivity during production run"). Soft constraints are strong preferences (e.g., "should integrate with existing ERP export format within 90 days of deployment").

4. Out of scope (explicit): What you are deliberately NOT building for this version. The out-of-scope list is as important as the scope list — it is what prevents the scope from expanding during build. Example: "Predictive maintenance scoring is out of scope for v1.0. Multi-site centralized dashboard is out of scope for v1.0."

5. Definition of done: The specific, binary condition that means this version is complete and ready to deploy at the next customer. Not "when the engineering team feels confident" — "when the reference customer's QA manager signs off on the first 30-day inspection log report."

The constraint on length is doing real work. When you cannot fit the brief on one page, the excess is almost always: (a) implementation details that belong in the engineering spec, not the product brief; (b) aspirational features that belong on the roadmap, not the v1.0 definition; or (c) hedge language that should be replaced with an explicit tradeoff decision. The brief should be readable by your engineering lead, your first paying customer, and a new investor in under 5 minutes — if any of those people would get lost, trim it.

SEE ITMVP Feature Voter — rank and scope Forge's first production-ready version

⚙ Interactive · Feature prioritization for production readiness

Each row is a candidate feature for Forge's first production MVP. For each feature, set: Criticality to core job (1 = peripheral, 5 = the product fails without this) and Build effort (1 = one sprint, 5 = requires major infrastructure). The widget ranks features by their value-to-effort ratio, flags features where effort exceeds criticality by more than one tier (a build trap), and maps each shortlisted feature back to the specific job steps your Stage 1 discovery interviews uncovered. Adjust until your shortlist covers every mandatory job step without any build traps in the critical path.

Ranked shortlist & job-step coverage
—

Notice the diagnostic the widget provides: features in the "build trap" zone (effort > criticality by more than one tier) are the technical founder's instinctive builds — the things that feel important to build well but are not what the customer needs to complete their core job. The widget also reveals coverage gaps: if any mandatory job step has no shortlisted feature covering it, the MVP is incomplete regardless of how polished the rest of it is. A common mistake is building the feature set that makes the demo impressive while leaving the feature set that makes the deployment functional. The widget makes that tradeoff visible.

FORGEFirst shippable version — from pilot to production contract

Forge has completed its learning MVP phase: three paid pilot contracts at $8,000 each, deployed at one manufacturing site, with 30-day activation data showing measurable defect detection improvement. The question now is what the production MVP looks like — the version that justifies a full ACV contract at $30K–$50K per site and can be deployed by the customer team without Forge founders on-site.

🔬 Forge's production readiness checklist — what must be true before the next contract

Data security: Customer data processed on-device; only aggregate telemetry (uptime, model version, alert counts) sent to Forge cloud. Customer signs a DPA. Edge device is customer-owned hardware after 90 days.

Audit logging: Every inspection event logged with timestamp, part ID (from line scanner), result (pass/flag/fail), confidence score, and model version. Logs exportable as CSV or JSON. Retention configurable (default: 13 months, matching ISO 9001 record-keeping requirements).

Uptime SLA: 99.5% uptime commitment during production hours. Failsafe mode: if device loses inference capability, line passes through with a "no-read" flag rather than a false pass. Forge's maximum response time for a production incident: 4 hours during business hours, 8 hours off-hours.

Change management: Three-page operator quick-start guide. 90-minute onboarding session for QA lead (delivered remotely). Integration checklist for IT team. Support ticketing system with response SLA.

Integration: CSV export with configurable field mapping for MES import. JSON webhook available for customers with API-capable MES. Full ERP integration (SAP QM module) on 90-day roadmap as a paid implementation.

The milestone this stage: three paid pilot contracts signed at $8K each; edge device plus dashboard MVP deployed at one manufacturing site; first activation data collected; evidence that customers who complete onboarding see measurable defect detection improvement within 30 days. The production MVP version shifts the contract from pilot to recurring: a $30K ACV with a 12-month term, renewable. That price is justified not by the sensor's accuracy (which the pilot proved) but by the operational reliability, the compliance infrastructure, and the customer team's ability to operate it without Forge.

0% 40% 80% 100% Data Security 30% Audit Logging 10% Uptime SLA 60% Change Mgmt 20% Integration Story 40% Pilot status Required: production
Forge's production readiness gap at the end of the pilot phase. The shaded bars are where Forge is today; the dashed outlines are where it needs to be for a production contract. The audit logging and change management gaps are the largest — and the most commonly underfunded in technical-founder-led companies.

WEAKNESSThe GA announcement trap — why announcing before you are ready closes markets

General availability announcements in B2B enterprise markets carry a specific risk that B2C or developer-tool markets do not: the reference customer network. In consumer software, a bad launch can be recovered with a patch and a public apology. In manufacturing software, a bad launch at your one reference customer propagates through a tightly connected buyer network before you have time to fix anything.

Here is the failure sequence: Forge announces production availability at a manufacturing conference. A plant manager at a second auto-parts customer (who was interested but had not yet committed) reaches out for a reference. The reference customer — whose deployment had two integration failures in the first two weeks — gives an honest but tepid response. The second prospect defers. Two other prospects who would have inquired after the conference announcement hear from the second prospect that Forge's deployment "had issues." The market for Forge in Tier-1 auto stamping contracts in that region goes cold for 12 months, which is exactly the length of the annual budgeting and procurement cycle for capital equipment and software in manufacturing.

⚠ The 12-month window

Manufacturing procurement runs on annual budget cycles. Capital and software line items are approved in Q4 for the following year. If your failed GA announcement poisons the reference well in October, the next natural procurement window is October of the following year — after you have spent 12 months either rebuilding the product or trying to resell prospects who have already heard the bad story. The asymmetry is severe: 12 weeks of premature announcement pressure versus 12 months of market closure. The right call is almost always to delay the GA announcement until the reference customer can unambiguously say "this worked."

The discipline is simple to state and hard to hold under investor and competitive pressure: do not announce general availability until at least one customer has completed the full deployment lifecycle without founder intervention. "The full deployment lifecycle" means: device deployed, integration complete, operators trained, first 30-day inspection report generated, QA manager has reviewed and is using the output in their actual QA workflow. Not "the demo went well." Not "they said they were happy." The full cycle, run independently.

WEAKNESSWhere this bites

✗ Treating "the product works" as the same as "the product is deployable"

Technical founders make this conflation constantly: the model is accurate, the dashboard is usable, the API is clean — therefore the product is ready. Production readiness is not a product quality assessment. It is an operational capability assessment. The questions are: Can the customer's IT team deploy it? Can the customer's ops team operate it during a shift without calling you? Can the customer's QA manager pull a compliant audit report? Can the system fail without taking the production line with it? None of these questions have anything to do with whether the model is accurate, and all of them must be answered before calling something production-ready.

✗ The one-page brief becoming a ten-page spec

The product brief is not a requirements document. It is a decision artifact — it specifies the outcome the product must achieve and the constraints it must operate under, without specifying how. Every time a technical founder writes "the system shall" in a product brief, they are writing an implementation constraint, not an outcome. The consequence: engineers optimize for compliance with the spec rather than for the customer outcome, and every deviation from spec becomes a negotiation rather than a judgment call. Keep the brief short. The engineering spec can be long. The brief is for alignment, not implementation.

✗ Skipping the explicit out-of-scope list

Scope creep in the transition from MVP to production is not caused by adding too many features — it is caused by never explicitly deciding which features are out of scope. Without a documented out-of-scope list, every request that comes in during the build phase is an implicit in-scope item until someone explicitly says no. In the Forge context: predictive maintenance, multi-site dashboards, ML model retraining in customer environments, and MES deep integration are all features that customers will ask for and that are genuinely compelling. None of them belong in v1.0. They all need to be explicitly and documentedly out of scope — not because they are bad ideas, but because including them delays the production-ready deployment that is the actual revenue milestone.

KEY INSIGHTHow this connects

  • What MVP Actually Means (2.2) — this lesson extends the MVP definition from 2.2: you learned that an MVP is the minimum test of the riskiest assumption. A production MVP is a distinct artifact: the minimum set of functionality that a real customer can operate independently in their real workflow. The distinction between these two meanings is where most teams get lost.
  • Scoping the MVP — Ruthless Subtraction (2.3) — the subtraction discipline from 2.3 applies directly to the production MVP feature set. The one-page brief format and the explicit out-of-scope list are the production-phase instantiation of the same ruthless subtraction principle: every feature you include is a deployment you have to get right.
  • Testing the MVP — What Good Feedback Looks Like (2.6) — the feedback from pilot deployments (from 2.6) is what drives the production readiness gap analysis. If the learning MVP feedback identified integration and audit logging as the key friction points, those are the production MVP priorities — not the features that were easiest to demonstrate in the pilot.
  • Problem/Solution Fit — How You Know You Have It (2.9) — the production MVP is the artifact that produces the definitive problem/solution fit signal: a customer using the product independently to complete their core job. Lesson 2.9 defines what that signal looks like and how to measure it with activation data rather than satisfaction surveys.

The transition from MVP-for-learning to MVP-for-production is the moment Stage 2 delivers its milestone: not just evidence that the product creates value in controlled conditions, but evidence that it can be deployed, operated, and relied upon by a real customer team. That evidence — three paid pilots with documented activation data — is what makes the Stage 3 revenue architecture conversation possible. You are not pricing a product yet; you are pricing the first thing that has genuinely earned that name.

GLOSSARYKey terms introduced

Production MVP
The minimum set of functionality that a real customer can operate independently in their real workflow — without founder intervention. Distinct from a learning MVP, which is designed to generate a validated verdict; the production MVP is designed to be operated. The key success criterion is independent customer operation, not validated learning.
Production readiness gate
A deliberate checkpoint between the learning MVP and the production MVP phases, requiring sign-off on data security, audit logging, uptime SLA, change management, and integration readiness before deployment is declared complete. Skipping this gate is the primary cause of reference customer failures in enterprise B2B.
Audit logging (manufacturing context)
An immutable, timestamped record of every inspection event, including the result, confidence score, and model version that produced it. A legal and regulatory requirement in ISO 9001/IATF 16949 environments. Absent audit logging, a production deployment is not compliant regardless of inspection accuracy.
One-page product brief
A decision artifact that specifies: (1) the job to be done, (2) 3–5 measurable success criteria, (3) hard and soft constraints, (4) an explicit out-of-scope list, and (5) a binary definition of done. Specifies outcome without over-constraining implementation. Longer than one page usually means implementation details have been confused with outcome specification.
Change management support
The materials and process support that enable a customer team to adopt a new product into their existing workflow without requiring ongoing founder involvement. Includes operator training materials, an integration checklist for IT, a support escalation path, and a onboarding runbook. A prerequisite for independent operation, not an optional add-on.
GA announcement trap
The premature declaration of general availability before at least one customer has completed the full deployment lifecycle independently. In B2B enterprise markets with concentrated reference networks, a failed GA deployment can close the target market for 12 months — aligned with the annual procurement cycle.
Build trap (feature scoring)
A feature where build effort exceeds its criticality to the customer's core job by more than one tier. Build traps feel important to technical founders because they are technically interesting; they do not appear in customer job-to-be-done analysis because they are not what the customer needs to accomplish the outcome they are hiring the product to deliver.
Uptime SLA
A contractual commitment to minimum system availability during defined operating hours, with a specified response time for production incidents and a defined failsafe behavior when the system falls below the SLA threshold. For first-production manufacturing deployments, 99.5% uptime during production hours with a 4-hour response time is a reasonable starting point.

ACTIVE RECALLCheck your understanding

Answer in your own words first, then reveal.

1What is the structural difference between a learning MVP and a production MVP, and why is the transition not just a matter of adding more features?›

A learning MVP is designed to generate a validated verdict — a behavioral signal that confirms or falsifies the hypothesis that a real customer will use and pay for the proposed value. A production MVP is designed to be operated independently — by the customer's team, in their real workflow, without founder intervention. The transition is not additive because the criteria for success are categorically different: a learning MVP succeeds when it produces a clean signal; a production MVP succeeds when the customer can complete their core job without calling you. The features that generate clean learning signals (concierge support, manual workarounds, founder-on-site debugging) are actually antipatterns in a production deployment. The transition requires removing founder crutches, adding operational infrastructure (audit logging, uptime monitoring, change management materials), and documenting the product well enough that a customer IT team can deploy it from a spec sheet. That is not a feature — it is a different operating model.

2Why does the GA announcement trap create a 12-month market closure risk specifically in manufacturing, rather than the typical 2–4 week news cycle in consumer software?›

Three structural reasons: Procurement cycles. Manufacturing capital and software purchases are budgeted annually. A vendor that gets knocked off the approved list in Q4 must wait until the next budget cycle — typically 12 months — to re-enter consideration. Reference networks. Manufacturing procurement is heavily reference-driven. Plant managers talk to peer plant managers; quality directors know each other from industry groups. A bad reference from your first deployment propagates through the relevant buyer network before you have time to fix the underlying problem. Risk aversion. Manufacturing environments cannot afford production line disruptions. A vendor that failed in deployment at one plant is not given a second chance at another plant in the same network — the switching cost of a failed deployment (production downtime, QA record gaps, change management rework) is too high. In consumer software, a bad launch can be recovered with a patch note. In manufacturing enterprise software, a bad deployment is a multi-week production incident. The asymmetry in recovery time creates the asymmetry in market closure risk.

3A technical founder argues: "We don't need audit logging in v1.0 — our pilot customer didn't ask for it, and we can add it in v1.1." How do you respond?›

The argument confuses the absence of an explicit request with the absence of a requirement. Audit logging is not a feature that manufacturing customers ask for — it is a regulatory and legal baseline they assume. An ISO 9001 or IATF 16949 certified plant is required to maintain inspection records for every part they ship. If your system cannot produce those records, the QA manager cannot use your system as their inspection of record — which means it is not a production inspection system at all, just a parallel tool running alongside the real inspection process they are legally required to maintain. The pilot customer did not ask for audit logging because they assumed a production-grade inspection system would have it. The moment they discover it is missing, they will either (a) ask you to add it before expanding the contract or (b) realize the pilot was not actually a production deployment. Neither outcome is recoverable in v1.1 without re-doing the deployment. Build it before you call anything production-ready.

4You are writing a one-page product brief for Forge's production MVP. A colleague adds a paragraph describing the specific data schema for the MES export format. Should this stay in the brief? Why or why not?›

No — this belongs in the engineering spec, not the product brief. The product brief specifies outcomes, not implementations. The correct brief entry for MES integration is something like: "Customer's quality data from Forge is accessible to their existing MES within 30 days of deployment, using the customer's preferred export method." That specifies the outcome (accessible data, customer's preferred method, 30-day window) without constraining the implementation (the specific schema, the transport protocol, the API design). Once the specific schema is in the brief, the engineering team will build to the schema specification rather than to the outcome — and if the customer's MES uses a different schema than what was specified, you now have a spec compliance debate instead of a customer outcome problem. The brief disciplines the team to ask "did the customer get the outcome?" rather than "did we build the specified schema?" The schema lives in the engineering spec, where it is the right level of abstraction.

Next: measuring problem/solution fit with activation data, not satisfaction surveys →