Things move faster now. And this means that everything around us had to adjust – the way in which we work, what teams we build, technology choices and also how we measure stuff. When times were slower, ARR – Annual Recurring Revenue – was the north star for SaaS businesses. Saas products used to evolve slowly, business models were stable and predictable, growth was measured in annual cycles – and more pertinent: founders and investors had the luxury of time.
With the rise of AI-native products that move fast, shift pricing models, and often monetise differently, it’s worth asking: is ARR still the metric that matters? Is ARR no longer a trustworthy signal? And if not then what?
The Problem with ARR in Fast-Moving AI
AI products are evolving at breakneck speed. Take ChatGPT from OpenAI. In a year, it’s moved from free research preview to Pro plans, enterprise licensing, API usage, and even partnerships powering Microsoft products. Where does ARR fit into that picture?
Or look at Midjourney, the AI art tool. It monetised early through Discord-based subscriptions, but the speed of user growth, community virality, and network effects mattered far more in the early phase than predictable annual revenue.
Even Perplexity.ai, the AI-powered search tool, focused heavily on product quality and daily active usage long before introducing paid tiers. Success wasn’t ARR – it was habitual use and trust.
What’s Going Wrong with ARR?
ARR was designed for:
- Stable SaaS businesses with predictable subscription cycles.
- Linear growth models.
- Enterprise sales motions.
But AI-native businesses often:
- Experiment with pricing models rapidly (freemium, usage-based, team tiers).
- Grow virally via network effects or API integrations.
- Prioritise defensibility through data flywheels or custom models.
This makes ARR a lagging indicator. It can be useful, but by the time you hit £1M ARR, you might have already missed key inflection points — or burned through your runway chasing the wrong target.
So What Should We Measure Instead?
We can reframe what matters, especially in the early and growth phases of AI-native product development. ARR isn’t dead per se but…
1. Engagement Depth > ARR (Early Stage)
Metrics to track:
- Daily Active Users (DAU) / Weekly Active Users (WAU) (proxy for stickiness)
- Sessions per user
- Time to ‘Aha’ moment
- Prompt success rate / Model satisfaction rate
👉 If people aren’t getting value daily, ARR is a vanity metric. Engagement shows you’re building something people return to.
2. Speed of Learning > Predictability of Revenue
Metrics to track:
- Experimentation velocity (How fast can you ship + learn?)
- Feature retention (Do new features stick or get ignored?)
- Customer feedback loop tightness (Time from feedback to iteration)
👉 AI tools thrive when they’re improving fast. Optimising for ARR can slow you down by pushing you toward premature optimisation.
3. Monetisation Readiness > Monetisation Now
Ask:
- Are users getting clear value?
- Is there a natural paywall or upsell?
- What’s the willingness to pay for power users?
Examples:
- Runway ML started with generous free usage before creating tiered plans that scaled with creator needs.
- Notion AI added AI as an upsell after widespread adoption of its core product.
👉 ARR is a side effect of building a product that’s worth paying for. Don’t mistake the side effect for the strategy.
4. Scalability of Marginal Cost > Pure Growth
In AI, cost scales differently. Running inference isn’t free.
Track:
- Inference cost per active user
- Model utilisation efficiency
- Cost-to-value ratio per feature
👉 You can hit £100k ARR and still die if your margins are negative on every active user. Sustainable AI products monitor this early.
So Is ARR Dead?
No – but it’s no longer the hero metric. In AI-native products, ARR is a milestone, not a compass. It tells you how far you’ve come – not where to go next.
The better question is: Are we building the right kind of value – fast – in a way that users want and are willing to pay for?