
A Digital Product Strategy Should Tell You What to Kill
A digital product strategy is the set of decisions that survives contact with your CFO, your regulator, and your model bill. Seven of them: problem and evidence, value proposition, build versus buy, users and workflow, KPIs and kill criteria, regulatory posture, and unit economics. With a one-page template.
A digital product strategy is the documented set of decisions that determines what you will build, who it is for, what you will refuse to build, and what evidence will make you stop. Seven decisions, specifically: the problem and its evidence, your value proposition, what you build versus buy, the users and their workflow, your KPIs and kill criteria, your regulatory and risk posture, and the unit economics of serving the product, including what every AI feature costs to run. A roadmap is the output of those decisions. Most teams maintain the roadmap and skip the decisions.
We published the first version of this essay in 2019. The steps we recommended then, research through pricing, were right, and they have since become table stakes. Every competitor has personas now. Every competitor has a vision deck. What changed is where the hard decisions moved: into AI unit economics, build-versus-buy, kill criteria, and, for anyone shipping into healthcare, automotive, or industrial environments, regulatory posture. This is the 2026 rewrite, built from what we have deployed since.
What a Digital Product Strategy Is (and Isn't) in 2026
Strategy is not a roadmap. A roadmap is a delivery plan: what ships, in what order. Strategy is the decision system that tells you whether any of it should ship at all. That distinction matters more now than it did in 2019, because AI collapsed the cost of shipping. Marty Cagan's team at SVPG calls the result the AI Productivity Paradox: teams generate prototypes, specs, and code faster than ever, and outcomes have not improved. When production is cheap, judgment is the scarce asset. Knowing which of the hundred things you could now build is worth serving at scale.
Strategy is also not an AI feature list. Product coach Dean Peters describes the failure mode exactly: executives demanding "add AI to the product, PRONTO" with no use case attached is "just executive arson with a $500K burn rate." An AI mandate with no unit economics, no evidence requirements, and no stopping rule is a budget leak with a demo attached.
| Question | Roadmap answers it? | Strategy answers it? |
|---|---|---|
| What ships next quarter? | Yes | No, deliberately |
| What problem, and what evidence says it's real? | No | Yes |
| Which AI capabilities do we build vs. buy? | No | Yes |
| What does this feature cost per heavy user? | No | Yes |
| What result kills the initiative, and by when? | No | Yes |
| What will the regulator ask for, and when do we start producing it? | No | Yes |
Why the 2019 Playbook Stopped Working
The evidence that vision-deck strategy fails at deployment is now hard to argue with. MIT's Project NANDA reported in its 2025 GenAI Divide study that 95% of enterprise GenAI pilots deliver no measurable P&L impact, and that externally procured tools reach production roughly twice as often as internal builds. McKinsey's State of AI survey (November 2025, n=1,993) found 88% of organizations using AI in at least one function, nearly two-thirds still stuck in experiment-or-pilot mode, and only 39% reporting any EBIT impact at all.
Meanwhile the money got serious. In a16z's survey of 100 enterprise CIOs, AI budgets grew roughly 75% year over year ("what I spent in 2023 I now spend in a week," one CIO told them), and the share of LLM spend drawn from experimental innovation budgets collapsed from 25% to 7%. AI is a permanent line item now. Permanent line items get CFO scrutiny, and a persona deck does not survive CFO scrutiny.
None of those are technology failures. Pilots die because nobody wrote down what evidence would justify scaling, what the feature costs to serve, or what result would kill it. Those are strategy omissions. The 2019 toolkit did not fail so much as commoditize, and most strategy documents never followed the hard decisions to where they moved.
Buyers already have a name for the resulting condition: pilot purgatory. The demo worked, the steering committee applauded, and eighteen months later the initiative is still "in evaluation" because nobody can say what number would end the evaluation in either direction. A deployment-first strategy makes purgatory impossible by giving every initiative its own exit conditions, up or out, from the day it is funded.
The Deployment-First Product Strategy: A Seven-Step Framework
We call our framework the Deployment-First Product Strategy because it is written backwards from the day the product runs in production, at real usage volumes, at real token costs, in front of real auditors, rather than forwards from a vision deck. The seven steps mirror our 2019 sequence deliberately. The sequence was never the problem. What a decade of deployments and three years of AI economics rewrote is the content of each step.
1. Problem and Evidence
Discovery still starts with qualitative research: talking to the people who feel the problem, watching the workflow where it lives. The evidence bar is what moved. If the product involves AI, discovery now includes evals, meaning a written definition of what "good enough" means for your use case, tested before you commit a roadmap to it. Write the problem statement the way you would present it to a CFO: the problem, who has it, what it costs them, and the strongest evidence you have that they will pay for it to go away. If you are pre-revenue, this step is most of your strategy, and we cover the evidence side in depth in our founder's guide to product-market fit.
The eval requirement inverts the usual sequence, which is why teams resist it. The normal pattern is to commit to an AI capability, build it, then discover during QA that the model fails the cases that matter. Running the eval first, on your data and your edge cases against your accuracy threshold, costs days and saves quarters. If the best available model cannot clear your bar today, that is not a dead end. It is a strategy fact, and it belongs in the document with a date to retest.
2. Value Proposition and Differentiation
In 2019 we wrote that your value proposition "is not so much about what problems you solve, but how you solve them." That aged well, with one sharp edit: the model is no longer the how. As one CIO put it in the a16z survey, "for most tasks, all the models perform well enough now". Raw model capability is a commodity input. Your strategy has to name what you own that the next model release does not commoditize: proprietary data, workflow depth, distribution, integration into physical operations, or a validated evidence trail competitors would need years to reproduce. If a competitor with an API key could copy your differentiation section, it is not differentiation.
3. Build, Buy, or Orchestrate
This step barely existed in 2019 and now decides more outcomes than any other. Menlo Ventures' 2025 enterprise AI survey of roughly 500 U.S. enterprise decision-makers found 76% of AI use cases are now purchased rather than built internally, where a year earlier the split was 53% bought and 47% built. Put that next to MIT's finding that external tools reach production about twice as often, and the default has flipped. Buy the commodity capability, build only what differentiates, and orchestrate the seam between them.
Make the call capability by capability, and record two properties for each: reversibility and switching cost. The a16z CIOs named the trap themselves, "all the prompts have been tuned for OpenAI... changing models is now a task that can take a lot of engineering time". Model-agnosticism is a hedge against pricing power you do not control, which is what makes it more than an architectural nicety. Any capability you cannot reverse out of within a quarter deserves a written justification.
4. Users and Workflow
Personas told you who the user is. Workflows tell you where the product earns its keep, and in AI products the workflow question has teeth. Where does the model act autonomously, where does a human review, who can override, and what happens operationally when the model is wrong? A feature that saves ten minutes but needs five minutes of verification saves five minutes. Price and prioritize it that way. In regulated environments, human-in-the-loop is a claim you make to a regulator, and changing it later can reopen your submission, so decide it here rather than in sprint planning.
5. Goals, KPIs, and Kill Criteria
KPIs made our 2019 list. Kill criteria are the addition, and they separate a strategy from a wish. A kill criterion is pre-committed: if metric X has not reached Y by date Z, we stop. Mind the Product's 2026 AI product strategy guide proposes a useful economic version, the 3x Rule: an AI feature should generate at least three times its compute cost in measurable value. Its authors ask the question we now put to every client: "if a heavy user runs this feature 100 times a day, does our margin on that user go up or down?" Define success as reduced manual work and improved margin rather than engagement clicks. An AI feature that gets used constantly while losing money per invocation is a failure that looks like traction.
6. Regulatory and Risk Posture
Here is the step no other digital product strategy framework includes, and the one that decides winners in our industries. Treat regulation as a strategy input rather than a late-stage gate. The FDA has authorized more than 1,400 AI-enabled medical devices, and its Predetermined Change Control Plan guidance lets manufacturers pre-authorize future model updates, but only if you planned at submission time which changes you would want and how you would validate them. That is a product strategy decision made years before it pays off. The same logic runs through IEC 62304 for medical software, ISO 26262 and ASPICE for automotive, and IEC 62443 for industrial systems: the pathway you choose constrains your architecture, your data practices, and your release cadence. Outside device regulation the discipline holds too. An AI credit decision still owes every applicant a specific reason, and that explanation has to be designed into the model rather than reconstructed after a denial. Decide here which pathway, which standards, and what evidence you will generate continuously instead of reconstructing forensically.
7. Pricing and Unit Economics
In 2019, pricing digital products was mostly a psychology question, because marginal cost rounded to zero. AI ended that holiday. Every generative feature carries real COGS, which Mind the Product's guide splits into input tokens, output tokens, and retrieval. Your strategy has to state the price metric (per seat, per usage, per outcome), the expected COGS per user, and the margin at heavy usage, because flat per-seat pricing on usage-driven cost is how products die of their own success. Infrastructure placement is part of the same decision. We walked through the cost mechanics for high-volume connected products in our edge-versus-cloud analysis, and the conclusion generalizes: at scale, architecture is a pricing decision.
Strategy Under Regulatory Constraint
Most strategy content treats regulated markets as a special case. We treat them as the proving ground, because constraint is where strategy stops being theater. Take our work on the Canary Medical smart-knee platform, where implantable sensor data feeds recovery-trajectory intelligence for surgeons and patients. The decisive move was made early: treat the evidence trail as a first-class product deliverable. Verification and validation documentation and test automation were engineered into the development process from the start, and that body of evidence contributed to the client's own De Novo regulatory pathway. The regulatory outcome belonged to the client, as it always does. The strategy decision that made it reachable, generating evidence continuously instead of reconstructing it before submission, was made years earlier in exactly the step most frameworks omit.
That posture compounds. In healthcare and automotive, most of your competitors' AI initiatives will die in validation rather than in the market. A strategy that prices in the regulatory pathway from day one converts everyone else's compliance burden into your moat. Slower to first demo, dramatically faster to deployed and paid.
What This Looks Like in Practice
The AutoTrace fleet telematics platform processes tens of millions of vehicle events a day, a scale at which none of the seven steps stay abstract. The build-versus-buy line was drawn around differentiation: the telemetry ingestion and event-processing pipeline is the moat, so it was built and owned, while the commodity infrastructure beneath it was bought. Unit economics drove architecture, because at tens of millions of events a day the per-event processing cost is the margin, which made the pipeline design a pricing decision as much as an engineering one. And the KPI conversation was about operational outcomes for fleet operators rather than model benchmarks.
Notice what the walk-through did not include: a technology commitment made for its own sake. Every architectural and vendor decision traced back to differentiation, margin, or an operational outcome. That traceability is the practical test of a strategy document. If you cannot point at a line item on your roadmap and name the decision it serves, either the roadmap has drifted or the decision was never made.
The framework also forces an early call on who you build with. The economics apply whether the work happens in-house or with a partner, and we have written separately about how product studios differ from software vendors. Step 3's logic, own the differentiating capability and contract for reversibility, applies to partners exactly as it does to model providers.
A One-Page Digital Product Strategy Template
A strategy that does not fit on one page usually is not finished deciding. Here is the deployment-first one-pager we use, seven sections, one per step:
- Problem and evidence: two sentences on the problem and who pays for it, one line on the strongest evidence it is real, plus eval results if AI is involved.
- Value proposition: one sentence naming what you own that a model release does not commoditize.
- Build, buy, or orchestrate: a three-column list of capability, decision, and reversibility (how fast could we switch?).
- Users and workflow: who uses it, which workflow changes, where the human reviews and overrides the AI.
- KPIs and kill criteria: three KPIs tied to reduced manual work or margin, two kill criteria with metric, threshold, and date.
- Regulatory and risk posture: the pathway and standards that apply, and the evidence you will generate continuously.
- Unit economics: price metric, COGS per user across input tokens, output tokens, and retrieval, and margin at heavy usage.
Review the page quarterly and check the kill criteria monthly. Model economics move fast enough that steps 3 and 7 can flip inside two quarters, and a strategy document nobody reopens is a roadmap with a fancier name.
What Is the Difference Between a Product Strategy and a Product Roadmap?
A product strategy is the set of decisions, covering problem, value proposition, build-versus-buy, workflow, kill criteria, regulatory posture, and unit economics, that determine whether and why to build. A roadmap is the sequenced delivery plan derived from those decisions. If the strategy changes, the roadmap must change. If the roadmap changes with no strategy decision behind it, that is drift.
What Should a Digital Product Strategy Include in 2026?
Seven things: (1) the problem and its evidence, (2) the value proposition and what stays differentiated as models commoditize, (3) build-buy-orchestrate decisions with reversibility, (4) users and workflow including human oversight of AI, (5) KPIs and pre-committed kill criteria, (6) regulatory and risk posture, and (7) pricing and unit economics including AI COGS. The last three are what separate a 2026 strategy from a 2019 template.
Who Owns the Digital Product Strategy?
One accountable owner, typically the CPO, CTO, or founder depending on stage. Not a committee. In regulated companies the owner drafts it jointly with quality and regulatory leadership, because step 6 is a shared commitment, and a regulatory posture product leadership chose alone gets re-litigated at the worst possible moment.
How Often Should You Revisit a Digital Product Strategy?
Quarterly for the full document, monthly for kill criteria, and immediately on a triggering event: model pricing that breaks step 7, an eval result that breaks step 1, regulatory guidance that changes step 6. The cadence is the point. A strategy is a living decision system rather than an annual offsite artifact.
If you are taking an AI-enabled product into a regulated market, this framework is the conversation we have with every client before any code is written. You can see how we structure that work on our How we partner page, or go straight to the decisions and scope a First Build. About a month later you have the seven steps answered for your own product, with the unit economics and the regulatory posture written down rather than assumed.
Related Insights

Insight
How to Develop a New Product Guide Part 1
Thousands of products fail daily due to poor execution. Master the New Product Development (NPD) process with Part 1 of our guide and turn your idea into a market success.
Read More
Insight
UX Strategy - What is it and do you really need one?
Is UX strategy just a buzzword? Discover how bridging business goals with user needs drives innovation. Learn why your digital product needs a strategic roadmap today.
Read More
Insight
No Roadmap Survives First Contact With Users (And Why That’s a Good Thing)
"No roadmap survives first contact with the user." Discover why linear planning fails in production and how to build resilient, user-led digital products that adapt to real behavior.
Read More
Insight
A Closer Look at How Value-Based Product Development Works
Stop building features; start delivering value. Discover how value-based product development aligns your software strategy with business goals to maximize ROI and customer impact.
Read MoreFacing a problem like this one?
Tell us what the business needs to win and who owns it on your side. If it is a fit, a First Build starts within weeks. If it is not, we will tell you why.