👋 Hi, it’s Kyle Poyar and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups. Today’s newsletter is a ✨ bonus ✨ edition.

A quick update: Unhinged Perks just got even better. I’m adding five new offers exclusively for premium subscribers, bringing the total to 18 perks and $40,000+ in potential savings* from the likes of AirOps, beehiiv, Metronome, Retool, Rox, and much more. The new offers include:

^{*Terms and conditions apply and we may run out of codes at any time. More information here.}

We’ve grown accustomed to the traditional set of SaaS metrics as just part of how to operate a B2B business. Here’s the thing: the traditional SaaS metrics playbook can be extremely misleading when it comes to building an AI-native business.

ARR, for example, was the lifeblood of SaaS. It’s becoming meaningless untrustworthy for AI-native companies.

It isn’t just ARR that no longer feels as powerful as it once did. Let’s look at a few other examples:

I’ve been interviewing more than a dozen of the top AI founders for an upcoming report about how AI-native companies grow (stay tuned 👀). What struck me was how the metrics these founders obsess over have started to shift, although there was almost no consistency in metrics from company to company.

The reality is that the old SaaS metrics still have a place, especially for companies selling subscriptions with high retention and healthy gross margins. But we need to start looking for next-era metrics that better define success.

Today’s post unpacks seven alternative metrics that feel more relevant and urgent for this next era of AI-native companies. These include both operational KPIs (which help you identify where to focus) and investor KPIs (which help you communicate the health of your overall business model).

Perhaps the most unifying thing about AI companies is that they consume an immense amount of tokens. Tokens are a metric that’s hard to hide from. They show whether people are deeply using your AI product or if being AI-first is mere marketing jargon.

I was surprised when people began posting publicly about their Anthropic bills or their OpenAI “Token of Appreciation” plaques. There’s something a bit off about being proud of burning spending so much money on AI tokens.

Some investors, like Jamin Ball from Altimeter Capital, are pointing out that “the fastest-growing AI companies today are the ones sitting directly in the token path." Revenue is tightly linked with token consumption, and these companies are positioned to benefit as token use continues to explode (provided they have a product that makes these tokens more valuable).

I recently wrote about emerging shift to platform + tokens pricing at Clay and PostHog. In a sign of the growing popularity of token monetization, Stripe is now in private preview with a new LLM token billing feature. This can automatically pass through raw LLM token costs synced to the latest model prices with a consistent markup.

Token consumption could just be passing through your infrastructure costs with very little additional value. Or it could be delivering real value above and beyond the LLM costs. The counterpart to token consumption becomes the gross profit per million tokens (inclusive of all revenue streams).

Tokens are essentially fuel consumption. But gross profit shows how many miles you’ve driven with that fuel. As Tomasz Tunguz recently wrote, the gross profit per million tokens seems to correlate pretty closely with the multiples that AI companies are trading for.

This can start to become an interesting metric for sales teams, too. Historically reps have been compensated based on the ARR they’re bringing in. That makes sense in an 80% gross margin world. It doesn’t work as well if margins are 20-60% and could swing anywhere from -50% to +80% across the customer base.

If sales reps are allowed to discount your product, or if they can choose which monetization model is right for a given customer, you might wind up paying fat commissions for unprofitable deals. Adjusting compensation to align with gross margin dollars becomes a better representation of the rep’s actual contribution. A simplified version would be an earnings multiplier connected to gross profit, e.g. 1.5x for 80%+ margin, 1.25x for 65-80%, 1x for 50-60%, 0.75x for 35-50%, and 0.5x for 20-35%.

Everyone can produce AI outputs for customers. That doesn’t mean these outcomes are any good, or that they’re improving over time. There might be hallucinations, latency issues, generic slop, or inconsistencies.

I’ve talked to a number of companies that don’t have a good way of measuring the effectiveness of their AI. They might use certain benchmarks or evals prior to shipping improvements, yet those evals in testing don’t necessarily line up with the real world experience of users. Nor do these evals inherently translate into downstream benefits like higher conversion, more usage, or better retention.

I’m starting to see AI quality metrics become the AI-native analog of what the SaaS world considered product activation. My product activation framework looked like this:

Applying a similar lens to AI quality, potentially useful metrics could include explicit user feedback (ex: thumbs up) or implicit feedback based on actions taken by the user (ex: sharing AI outputs, downloading or copying the outputs, using the AI output as-is versus editing it). I recommend starting with a long-list of the plausible AI quality metrics that you can capture, then scoring these metrics based on the following factors:

This metric isn’t only for product teams. It can become part of your marketing and competitive differentiation, helping educate prospects about why your AI is better than what else is on the market.