How you need to think about product strategy in the age of AI

Context

Tim Herbig spent eight years as a PM before becoming an advisor, coach, and the author of Real Progress. In this episode, he explains what product strategy should look like now that AI has made feature shipping dramatically cheaper. His argument is not that PMs need a brand new strategy playbook. It is that the old fundamentals matter more than ever: making real choices, saying no, and measuring whether those choices changed customer behavior and business outcomes.

The Big Idea

AI changed the speed of product work, not the core job of product strategy. When shipping gets cheap, a strategy only counts if it forces real trade-offs, sharpens where to say no, and ties success to behavior and business impact rather than output.

Tim's core point is that many teams are mistaking faster execution for better product thinking. AI can help research markets, pressure-test choices, and speed delivery, but it cannot replace the judgment required to pick the right audience, the right winning move, and the right metric.

Key Insights

Cheap shipping punishes bad bets

• AI has made it much easier to ship features, especially for greenfield products and increasingly for established products too.
• That sounds like a pure win, but Tim's warning is the opposite: when features get cheap, wrong features get more expensive.
• The reason is simple. Every shipped feature still has to be adopted, understood, supported, and justified in the product. More output creates more clutter if the choices behind it are weak.
• For PMs, the new bottleneck is less often "can we build this?" and more often "should we build this, and will it matter?"

Real strategy enables no

Tim's simplest test for whether something is actually a strategy is whether it helps teams say no.

• A strategy is not a deck, a one-pager, or a template filled out correctly.
• A strategy is only real if it helps teams and stakeholders intentionally say yes or no to opportunities over a time horizon like six to eighteen months.
• Tim's practical signal: if most of a team's energy is not going into work connected to the strategy, then the strategy is probably not doing its job.
• A list of 14 priorities fails this test because it does not narrow choices. It just renames the backlog.

Questions beat templates first

Tim argues that teams often start in the wrong place. They begin with a framework, canvas, or polished strategy format before doing the actual strategy work.

The better sequence is to answer the core questions first:

• Narrative: why act now?
• Playing field: for whom are we solving what problem, and who else is solving it?
• Winning moves: how will we deliver value and differentiate?

Only after those questions are answered should a PM worry about whether the output fits a business model canvas, a strategy deck, or an internal template. Tim's point is that those artifacts are only windows into the work, not the work itself.

AI is not the moat

Tim is explicit that AI, by itself, is not a durable value proposition.

• Frontier models are widely accessible. Most companies can use the same underlying technology.
• That means "we have AI" is rarely a defensible strategic choice.
• The better question is how AI can amplify the reason customers already choose you.

He uses two examples to make the point:

• PhotoAI: Tim's take is that founder brand and distribution help make the product work, not just image generation itself.
• ChatPRD: Tim's read is that Claire Vo's product leadership experience is layered on top of AI, which creates the value customers perceive.

His broader conclusion matches what he heard from others like Brian Balfour and Ravi Mehta: the differentiators that still matter most are brand and data.

Audience choices drive downstream focus

Tim recommends getting very concrete about audience because audience choices create downstream clarity on problems, value propositions, roadmap calls, and metrics.

He uses a fictional marketplace called Fair Mile, a used-car marketplace, to show how this works.

• Start broad: all car buyers in Europe.
• Avoid getting absurdly narrow: first-time buyers of vintage Mercedes cars in southern Germany.
• Slice the market by dimensions that actually matter, such as region, first-time versus repeat buyers, and car value.
• Land on a segment specific enough to create focus, such as repeat buyers of cars worth more than 100K in Germany.

Once that choice is made, many other choices become easier:

• which buyer problems matter most
• what seller supply matters most
• what trust or service model matters most
• what roadmap items deserve priority

AI helps here by speeding up market sizing and research, but it does not make the audience decision for you.

Metrics must mirror choices

Tim thinks many teams jump too quickly into OKRs or dashboards. He prefers starting in plain language.

Ask two questions first:

• What will we do differently next quarter or half?
• How will we know that difference worked?

Only then should a team translate the answers into a formal metrics framework.

The key is to measure outcomes tied to the actual strategic choices, not generic company KPIs.

In the Fair Mile example:

• If the strategy is to focus on repeat German buyers of cars worth more than 100K, then a useful metric could be share of cars sold above 100K.
• If the winning move is a white-glove inspection and delivery service, then a useful metric could be recommendation score above 8 among German buyers of cars above 100K.
• A vague metric like revenue growth is too noisy, because many unrelated factors can move it.

Behavior matters more than tokens

Tim sees many AI product teams drifting into machine-centric metrics like token usage, model efficiency, or internal output counts.

His objection is not that those numbers are useless. It is that they are often measuring the wrong thing.

• Tokens used are part of the value exchange with the model provider, not the value exchange with the customer.
• A good strategy metric should explain whose behavior needs to change and by how much.
• If an AI feature ships 20 times faster but customer behavior still takes nine months to move, PMs should ask what the faster shipping actually achieved.

For product managers, this is a sharp reminder that AI metrics are often operational inputs, not proof of product value.

Mental Models & Frameworks

Strategy's three core questions

Tim reduces product strategy to three reusable questions:

• Why now? This is the strategic narrative.
• For whom, and against what alternatives? This is the playing field.
• How do we win? This is the set of winning moves, including value proposition and differentiation.

Use these questions when a team is getting lost in format debates. If a strategy artifact cannot clearly answer all three, it is incomplete no matter how polished it looks.

Roger Martin's stupidity flip

This is Tim's favorite pressure test for strategy quality.

• Take one strategic choice.
• Flip it to the opposite.
• Ask whether the opposite is obviously stupid.

If the opposite is obviously stupid, then the original statement probably was not a real strategic choice.

Examples from the episode:

• "Make salespeople more efficient" fails, because the opposite, making salespeople less efficient, is obviously stupid.
• "Make salespeople more efficient by shortening meeting preparation time" is better, because there are legitimate alternatives for how efficiency could be improved.
• In the Fair Mile example, focusing on first-time buyers versus repeat buyers is a real choice because both are plausible.

Tim recommends running the stupidity flip on individual strategy components, such as audience, value proposition, or differentiation, not only on one giant strategy sentence.

Audience tree for focus

The audience tree helps a team move from a broad market to a strategically useful segment.

• Start with the broad audience.
• Define one overly narrow version.
• List the dimensions that matter for segmentation, such as region, buyer type, or transaction value.
• Recombine those dimensions until the segment is specific enough to support decisions, but large enough to support a business.

Tim's Fair Mile example moved from all European car buyers to repeat buyers of cars worth more than 100K in Germany. The value of the exercise is not the exact answer. The value is that it gives a PM something concrete to exclude.

Seven Powers moat filter

Tim uses Hamilton Helmer's 7 Powers to test whether a supposed differentiator is likely to last.

He highlights several powers discussed in the episode:

• Brand: still durable, and hard to accelerate.
• Network effects: still strong, especially in marketplace models.
• Switching costs: real in products like iOS or Spotify, but weak for a car marketplace.
• Process power: can matter, but competitors can often catch up.
• Counter-positioning: works when incumbents are blocked by their own business model, like Blockbuster relying on late fees.
• Captured resource: weak when the same AI models are broadly available.

Use Seven Powers when a team claims an AI feature is defensible. Tim's view is that features like AI search may improve the experience, but they rarely become the reason a company wins.

Strategy to metrics loop

Tim describes strategy and measurement as a loop, not a handoff.

• Make a strategic choice.
• Define the metric that should move if that choice is correct.
• Watch whether the metric actually changes.
• If the metric does not move, revisit the strategy choice.

This matters because generic KPIs are often too delayed and too noisy to tell a PM whether a specific strategic bet worked. The loop only closes when the metric is tightly connected to the choice.

Product strategy stack

Late in the episode, Tim refers to Ravi Mehta's product strategy stack to explain how strategy changes by level.

• Company strategy sets broad direction.
• Product strategy narrows that direction for a product area.
• Roadmaps and goals should then express that narrowed choice in execution terms.

Tim's practical advice is to ask, at each level, what can be inherited from above and what still needs to be narrowed further so the team can actually say no inside its own scope.

Decision Principles

Principle: Test for real choices

• When: a strategy statement sounds good but feels too broad to guide decisions.
• Why: a strategy that cannot survive Roger Martin's stupidity flip is usually a slogan, not a choice. If the opposite is obviously stupid, the original statement still leaves too many options open.

Principle: Inherit, then narrow

• When: a PM is operating inside a company strategy or product-area strategy they did not create.
• Why: not every level needs to reinvent strategy from scratch. A PM should inherit what is already specific enough, then narrow only the parts that still do not help the team say no.

Principle: Use AI as leverage

• When: a team is deciding how AI should appear in product strategy.
• Why: AI is best used as a resource that strengthens an existing reason to choose the product. Treating AI itself as the moat will usually lead to weak differentiation because the underlying models are widely available.

Principle: Measure behavior, not activity

• When: a team is proposing success metrics for an AI initiative.
• Why: internal output and model metrics can be useful for operations, but they do not prove customer value. A stronger metric shows which customer or user behavior changed because of the product choice.

Trade-offs & Nuance

Speed versus strategic clarity

AI gives teams a real speed advantage in research, synthesis, scenario generation, and delivery. Tim is not arguing against using it.

The trade-off is that more speed without more clarity can flood the product with low-value work. Faster shipping only helps if the strategic choices behind it are sound.

Focus versus full coverage

Tim makes an important nuance that strategy does not need to cover everything a product team does.

• Choosing paid users as the strategic focus does not mean turning off the product for free users.
• Choosing repeat buyers as the focus does not mean pretending first-time buyers do not exist.
• The real requirement is that the segment choice fits the narrative behind the strategy.

This is useful for PMs because many teams resist focus out of fear that focus means abandonment. Tim's point is that focus means priority, not denial of reality.

Leading signals versus true outcomes

Tim does not present one perfect metric rule.

• Highly customer-centric metrics can be too slow to guide decisions.
• Highly leading metrics can be easy to track but detached from customer value.

The practical trade-off is to find the earliest signal that still has a believable link to the behavior or business outcome the strategy is meant to change.

AI features versus durable moats

Tim is not saying AI features are useless. He is saying they are often easy to copy.

A team may still build AI search, AI verification, or AI-assisted recommendations if those help the experience. The nuance is that those features should usually be treated as supporting moves, not as the core moat.

Common Mistakes

Mistake: Writing all-yes strategies

Tim's Fair Mile example included a deliberately bad strategy: throw AI everywhere, expand to all countries, expand to all categories, and accelerate growth.

Why it fails:

• it does not exclude anything
• it does not guide trade-offs
• it encourages more output without sharper judgment

The better approach is to make a choice specific enough that the team can explain what it will not pursue.

Mistake: Starting from templates

Many teams open a canvas, fill boxes, and mistake completion for strategy.

Tim's warning is that strategy work is messy by nature. The neat deck or framework should come at the end, after the team has worked through audience, alternatives, value, and differentiation in a more exploratory way.

Mistake: Treating AI as differentiation

A product that says "we use AI" has usually not said anything strategically meaningful.

Tim's view is that durable differentiation still comes from things like brand, data, network effects, or other advantages that take time to build. AI should strengthen those advantages, not substitute for them.

Mistake: Measuring model activity

Tim sees teams reporting metrics like tokens used, language model efficiency, or raw feature output as if those prove product progress.

Those metrics can matter internally, but they can distract from the only question that matters strategically: did the product change user behavior and improve the business?

Practical Application

Pressure-test current strategy

Take the current strategy for your product area and run Roger Martin's stupidity flip on each component.

• Flip the audience choice.
• Flip the value proposition.
• Flip the differentiation claim.
• Check whether each opposite is still plausible.

If the opposite sounds obviously stupid, rewrite the original statement until it describes a real choice.

Build an audience tree

Before the next planning cycle, map one product area from broad audience to decision-ready segment.

• Start with the widest reasonable audience.
• Define one version that is comically narrow.
• Slice the audience by 2 to 4 dimensions that matter, such as region, user maturity, company size, or transaction value.
• Pick the version that best supports saying no while still being commercially meaningful.

Use AI to speed up the market sizing and research work, but make the segment decision yourself.

Stress-test differentiation claims

List every reason your team thinks customers choose the product, then run those claims through Hamilton Helmer's 7 Powers.

• If a supposed differentiator depends mainly on access to AI models, treat it as weak.
• If a supposed differentiator depends on brand, data, network effects, or counter-positioning, test whether it is actually visible to customers and hard for competitors to match.

This exercise helps prevent AI feature excitement from being mistaken for strategic advantage.

Rewrite strategy metrics

For the next strategy review, force the team to answer two plain-language questions before any OKR writing starts:

• What will we do differently?
• How will we know that difference worked?

Then translate those answers into metrics tied to the strategic choice. Replace generic success metrics like revenue growth with choice-specific metrics whenever possible.

Map roadmap items back

Review the next quarter's roadmap item by item and ask what strategic choice each item supports.

• If an item cannot be linked to a specific audience choice, winning move, or strategic metric, treat it as suspect.
• If an item can be linked, ask what evidence says it will move the metric tied to that strategic choice.

This is a practical way to stop roadmaps from drifting away from strategy.

Questions to Consider

• Which statement in your current product strategy would fail Roger Martin's stupidity flip because the opposite is obviously stupid, and therefore does not describe a real strategic choice?
• Which audience for your product is still too broad to help the team say no, and what 2 to 4 dimensions could you use to build an audience tree that creates sharper focus?
• If every competitor had access to the same frontier AI models tomorrow, what would still make customers choose your product over the alternatives?
• Which metric in your AI roadmap is measuring model activity, such as token use or output volume, instead of measuring a change in customer behavior or business impact?
• For each major roadmap item next quarter, can you point to the exact strategic choice and the exact strategic metric that item is supposed to support?

Bottom Line

Tim Herbig's message is that AI makes product strategy more important, not less. The teams that win will not be the ones that ship the most AI features. They will be the ones that make sharper choices about audience and differentiation, use AI as leverage rather than a moat, and measure success through behavior and business change instead of output.

Resources Mentioned

Case Studies Mentioned

PhotoAI and ChatPRD

Tim uses PhotoAI and ChatPRD to argue that successful AI products are rarely winning on model access alone.

• In PhotoAI's case, he points to founder brand and distribution as part of why the product works.
• In ChatPRD's case, he points to Claire Vo's CPO experience being packaged through AI.

The lesson for PMs is that AI often delivers the capability, but brand, expertise, and distribution shape why customers trust and adopt the product.

Netflix versus Blockbuster

Tim uses Netflix and Blockbuster to explain counter-positioning.

• Blockbuster earned meaningful revenue from late fees.
• Netflix removed late fees.
• That made it hard for Blockbuster to respond without hurting its own business model.

The lesson is that some strategic moves work not because they are clever in isolation, but because incumbents are trapped by the economics of their current model.

Personio's evergreen benefits

Tim shares a concept from a friend at Personio, described as Europe's largest HR software operating system.

The idea of evergreen benefits is to ask what the company is already unusually good at, then make lower-level product strategy build on that strength instead of starting from zero. Examples could include distribution, integrations, or existing technology assets.

For PMs, the lesson is that product-area strategy should often inherit and exploit company strengths rather than trying to invent entirely new ones.

People to Follow

Tim Herbig

Tim Herbig is a former PM turned advisor, coach, and the author of Real Progress. He focuses heavily on making product strategy practical, especially for PMs who need clearer ways to connect strategic choices to execution and measurement.

Ravi Mehta

Ravi Mehta is referenced twice in the episode: once for the quote that prioritization problems are often strategy problems, and once for the product strategy stack. He is relevant here because his work helps PMs separate company strategy, product strategy, roadmap choices, and goals.

Hamilton Helmer

Hamilton Helmer is the creator of the 7 Powers framework that Tim recommends for evaluating differentiation. PMs working on strategy can use his work to pressure-test whether a claimed advantage is actually durable.

Notable Quotes

"difficulty prioritizing is often a strategy issue, not an execution issue." (Ravi Mehta)

"AI by itself, in your strategy, it's not a value proposition." (Tim Herbig)

"The point is, did it help you say no to things?" (Tim Herbig)

"The only sustainable differentiators and product strategy are the ones that take the longest to build." (Tim Herbig)