Strategy thought partner
Ask the product questions that shape your strategy
What new AI features could be our moat? Is anyone actually making usage pricing work? Ask the questions that decide your quarter and the answer comes back from what the market has already done, with the pricing page, launch or teardown behind every claim.
Across the 113 companies with digitised pricing, 77 sell AI in some form and the split is not what most teams assume. Bundling it into tiers you already have is the most common (32) — Figma gives every seat type on every tier, free plan included, its own monthly credit allowance 3. Usage metering is close behind (27). Gating it behind a paid tier, the option you are asking about, is a distant third at 12.
What the gate looks like when it is used
Slack unlocks its first AI summaries at Pro and adds more at every step above it 4; Atlassian puts Rovo in Standard with nothing in Free 5. Both gate a feature whose cost per use is small and roughly fixed. The metered group looks different — Linear ships its Agent on every tier including Free, then meters the expensive part, with coding sessions and Loops both carrying a “requires AI credits” footnote 1, and Notion bills Custom Agents at $10 per 1,000 monthly credits on top of tier access 2.
What this means for you
The real question is not which tier, it is whether your AI feature has a marginal cost that scales with enthusiasm. Slack’s summaries do not, so a tier gate holds. Linear’s coding sessions do, so a gate would either cap their best users or cost them money — which is why they metered instead. Check your own cost per active user first: if usage is unbounded, a flat tier is a bet on your customers being lazy.
LinearPricing capture · 22 Jul 20262
NotionPricing capture · 23 Jul 20263
FigmaPricing capture · 22 Jul 20264
SlackPricing capture · 22 Jul 20265
AtlassianPricing capture · 29 Jul 2026Every answer is cited. Every claim traces back to the exact signal, pricing capture or teardown behind it, so you can check the evidence before you repeat it in a meeting.
product signals every answer can draw on
sources behind the evidence, from changelogs to earnings calls
of answers carry the signals they were built from
The workspace
A complete strategy toolkit, grounded in real market evidence
Across the digitised pricing pages, bundling into existing tiers is the most common approach, with usage metering close behind. The split turns on whether your cost scales with enthusiasm.
Ask the questions your strategy depends on
Pricing, positioning, what to stop doing, whether an advantage survives contact with a competitor. Each one is answered from what leading product companies have already tried, and every claim carries the pricing page, launch or teardown it came from.
You take evidence into the room instead of a point of view.
Thought Canvas
Work a decision through on one canvas
Put the question in the middle and build the case around it: evidence from Product Intelligence on one side, the realistic options below, and a framework run over them on the other. A Next step guide takes you from question to decision, and the finished canvas becomes a deck or a document to share.
Evidence · UI example
AI add-on
A footnote clarifies that certain AI features like Coding sessions and Loops require AI credits, revealing metered usage hidden behind a flat per-seat price.
Linear pricingTeardown
Evidence · Signal
Pay-per-use model will become standard offering for AI agents, with upgrade paths
Perspective from Jeff Weinstein, Senior Product Manager.
Stripe · 18 Aug 2026Open
Evidence · Pricing teardown
Metronome AI pricing
Pricing tiers: Starter (unpriced: Billing volume billed at 0.8%, Events volume billed at $0.04/1k ingest events, Real-time usage metering and alerting…
Metronome · 14 Aug 2026Open
Evidence · Pricing teardown
LaunchDarkly pricing
LaunchDarkly uses tiered plans layered with usage-based pricing; it starts at $10 metered by usage, offers a free tier, and includes a full feature comparison table.
LaunchDarkly · 1 Oct 2026Open
The question
Should we move our AI features to usage-based pricing?
Option A
Keep flat subscription, fence AI with soft fair-use caps
Leave the per-seat plan as it is. Add a published fair-use ceiling on the heaviest AI features and a clear footnote, similar to the credits footnote in Linear’s teardown. Only accounts that exceed the ceiling get a conversation about an add-on.
- Upside
- Keeps the buying decision simple while protecting you from a few heavy users eating margin.
- Risk
- Caps feel arbitrary, and you may give up revenue from your most engaged accounts, who would pay more.
- Watch
- Share of accounts that hit the ceiling in a quarter, and whether they upgrade, grumble or churn.
Option B
Hybrid: seat platform fee plus prepaid AI credit packs
Keep the annual subscription as a platform fee with a bundled monthly credit allowance. Heavy AI actions draw down credits, and teams buy top-up packs when they run out. This follows Siena’s platform fee, LaunchDarkly’s tiers layered with metering, and Linear’s credits footnote.
- Upside
- Revenue scales with real AI cost and value while the base price stays predictable for procurement.
- Risk
- Credits add friction and bill-shock anxiety for enterprise buyers, and metering has to be built before other work they ask for.
- Watch
- Whether prospects ask about credit limits and overage before features, and the share of accounts buying top-ups.
Option C
Don’t move yet; price on seats and learn from usage
Hold pricing steady while usage data is thin. Instrument per-account AI usage and cost behind the scenes, and use the next round of demos and early renewals to see how teams actually use it.
- Upside
- Avoids locking in a model before you know which AI features customers value.
- Risk
- If a few accounts run up heavy AI costs, margin erodes quietly, and Stripe’s view that pay-per-use becomes standard could force a later price change.
- Watch
- AI cost per account as a share of its subscription, and any prospect asking for usage-based terms unprompted.
Framework · Pre-mortem
Option C ahead, with usage instrumented now so Option A or B can follow once demo and renewal data exist
- Option A: a year on, it failed because the caps were guesses. With no usage data, the ceiling was set too low and annoyed the best accounts, or too high and protected nothing. Linear’s credits footnote shows a fence can work, but Linear knows its cost curve.
- Option B: a year on, it failed because credits slowed the sales cycle. The evidence for B is real (Siena’s platform fee, LaunchDarkly’s tiers layered with metering, Stripe’s view that pay-per-use becomes standard for AI), but it shows where pricing is heading, not that these buyers want it today.
- Option C: a year on, it failed because one or two heavy accounts quietly ate margin. A real risk, but bounded and watchable: track AI cost per account as a share of subscription and set a trigger.
- Conclusion: Option C comes out ahead. A and B fail from acting on missing information; C fails only if you do nothing with the data. Treat it as a short, dated decision, not a default.
Next step
See what the market is doing
Adds the five most relevant signals, decisions and teardowns from Product Intelligence.
Drag to pan · ⌘/Ctrl + scroll to zoom · Shift-click to select several cards
One canvas for “Should we move our AI features to usage-based pricing?”: the evidence it found, the three options it laid out, and a pre-mortem that weighs them.
Ask it anything strategic
The strategic questions to shape your roadmap
Ask whatever is actually on your mind this quarter — how to price AI, whether to go upmarket, what to stop doing — and get it answered from what leading product companies have already tried, with the pricing pages, launches and teardowns behind it attached. Desk research that used to take a fortnight and a deck now lands in the twenty minutes before the meeting.
The framework library
Use powerful frameworks to solve your problems
Describe what you are facing and it picks the framework that fits — Jobs to Be Done when a feature request needs reframing, Pre-mortem before a risky bet, Blue Ocean when every deal comes down to discounting — then works your own case through it in a live tool. A structured read on your situation in an afternoon, and reasoning your team can pressure-test rather than a conclusion they have to accept.
Every framework in the library ships an interactive tool — a canvas, a scoring table, a board or a tree — so you run your own situation through it rather than reading about someone else’s.
It runs the framework, not just names it. Each one ships an interactive tool — a canvas, a scoring table, a board or a tree — so you work your own case through it rather than reading about someone else’s.
Each one is drawn, not described. A framework you have never run is legible before you open it, so you can pick the one that fits the problem instead of the one you already know.
The Decision Maker
Pressure-test a decision before you commit to it
Frame the call, add what you already know, and pick the lenses: a pre-mortem that assumes it failed, a devil’s advocate, a bull case, the second-order effects, and a read of what the market has already tried. It weighs them against each other and leads with a recommendation — so the strongest objection to your decision is one you have already answered, rather than one raised for the first time in the room.
The lenses run against real market data, not just the prompt. Wider market analysis pulls real signals, pricing captures and comparable decisions; every lens also reads your company profile. The synthesis then weighs them against each other before the recommendation is written.
Decision
Context
Frameworks
You choose which lenses to apply. Frame the decision, add the context, then pick from six — a pre-mortem when the risk is what worries you, the wider market read when you suspect someone has already tried this.
Should we meter AI usage or gate it behind a tier?
6 lenses applied · grounded in real market data · 5 citations
Recommendation
Meter it. Your AI cost per active user is unbounded and your top decile already uses 6× the median, so a flat tier gate prices for the median and loses money on the users you most want to keep.
Synthesis
The pre-mortem and the devil’s advocate both land on the same failure: a tier gate that has to be re-drawn within two quarters. The bull case survives it, but only under metering.
Pre-mortem
A year from now this failed because heavy users hit the tier ceiling, downgraded their usage rather than their plan, and the feature’s engagement numbers made it look unsuccessful.
Second-order effects
Metering makes usage visible to the customer, which suppresses exploratory use — the thing that drives adoption in month one. Expect a slower ramp and budget for a free allowance.
Devil's advocate
Credits are a second currency your customers have to learn. Both of the metered pages you are modelling on needed a footnote to explain theirs, which is a legibility cost you are choosing to take on.
Bull case
Usage pricing lets you ship the expensive feature to the free tier, where it does the most acquisition work, without the cost scaling with people who were never going to pay.
Wider market analysis
Grounded in real pricing data: of 113 digitised pricing pages, 32 bundle AI into existing tiers and 27 meter it by usage — against 12 that gate it behind a paid tier and 6 that sell it as a separate add-on.
The report leads with the answer. Recommendation first, then the reasoning, then the individual lenses that justify it — so you can act on it in a minute or defend it for an hour.
Case studies
Learn the frameworks from decisions companies actually made
Each case study takes a decision a company actually made and works it through the frameworks that explain it — what they were betting on, what had to be true for the bet to pay off, and what a pre-mortem or Porter’s Five Forces makes visible about the call. Read a few and the frameworks stop being theory and become something you can run on your own decisions.
NetflixBuild, buy & partnerBuying Its Way Into AI Production Instead of Licensing the ToolsNetflix spent $587M to own an AI filmmaking stack rather than rent capability from vendors, betting that production tooling is now core infrastructure, not a commodity service.Porter’s Five Forces · Pre-mortem
MicrosoftProduct portfolioOne Copilot to Rule Them All: Consolidating AI Bets Into a Single PlatformMicrosoft is folding its scattered Copilot products into one unified AI platform for consumers and enterprises alike. The move trades the safety of many small bets for the scale advantages of one dominant surface.Jobs to Be Done · Porter’s Five Forces
VantaProduct portfolioVanta Bets on an AI Agent That Replaces the Compliance Work It Used to Just TrackVanta moved from monitoring compliance evidence to generating it, launching an AI agent that writes audit-ready SOC 2 documentation instead of just helping customers assemble it.Jobs to Be Done · Porter’s Five Forces
HubSpotProduct portfolioPivoting to agentic AI as core product bet amid commoditization riskHubSpot is redirecting serious R&D capital toward proprietary agentic AI (Breeze), betting it can out-compete generic LLMs before its CRM moat commoditizes.Porter’s Five Forces · Lean Canvas
SnapchatProduct portfolioBetting on Glasses: Snapchat Wagers Its Future on AR Hardware, Not SoftwareSnap moved beyond its app to build standalone AR glasses, staking capital and brand on the idea that spatial computing, not the camera feed, is the next platform. It is a costly, multi-year bet with no guaranteed mainstream buyer yet.Porter’s Five Forces · North Star Metric
RobloxProduct portfolioRoblox Bets on a Second Front Door: Creation Without a DesktopRoblox launched Build, a standalone mobile app for making games on the phone, betting that the next wave of creators won't touch a desktop at all. The move trades platform simplicity for reach into a much larger, less technical creator population.Jobs to Be Done · Opportunity Solution TreeGenerated from the Decision Tracker, so the library grows with the market rather than being written once.
Frameworks, applied to decisions that really happened. Each one names the bet, the conditions it depended on, and the frameworks that read it — so you see how a framework behaves on a real call before you use it on one of your own.
Grounded and personalised
Get answers that are personalized to your company
Tell it who you are once and it holds that behind every answer afterwards — what you sell, which tracked companies are genuinely comparable, how you are organised, how your own pricing page scores.
The same question from a 40-person design tool and a 4,000-person enterprise platform comes back with two different answers, each drawn from the companies that one can actually learn from. Advice you can act on directly, instead of translating someone else’s situation into your own.
Every answer still carries its sources, so personalised never means unaccountable — you can see which signals it read about the market, and which facts it used about you.
See more of the platform