There's a number sitting in row 12 of your model that you've never questioned.

‍

Gross margin: 80%.

‍

You didn't calculate it. You inherited it. It's the number every SaaS template ships with, the number every investor deck quotes, the number that made software the best business model of the last twenty years.

‍

And if you've shipped an AI feature in the last eighteen months, it's fiction.

‍

What the data actually says

‍

ICONIQ Capital surveyed around 300 executives at companies building AI products for its 2026 State of AI snapshot. Those companies expect average gross margins of roughly 52% in 2026 — up from 41% in 2024, but nowhere near the 70–80% that traditional SaaS has always targeted.

‍

Read that twice, because both halves matter.

‍

The optimistic half: margins are improving. Model routing is getting smarter, teams are pushing the bulk of workloads to smaller fine-tuned models and only escalating hard problems to frontier models, and token prices keep falling. Anyone telling you AI economics are permanently broken is selling something.

‍

The uncomfortable half: 52% is not 80%. And the gap between those two numbers is the gap between a business that can fund its own growth and one that can't.

‍

Ben Murray, The SaaS CFO, lays out the arithmetic in about as plain a form as it gets. Take £100 of revenue with £20 of traditional COGS — hosting, support, DevOps. That's an 80% gross margin. Now add AI features costing £15 in inference, model routing and infrastructure. Revenue is still £100. COGS is now £35.

‍

Your gross margin just went from 80% to 65%. You didn't lose a customer. You didn't cut a price. You shipped a feature everybody asked for.

‍

Why this breaks more than one row

‍

Gross margin isn't a vanity metric. It's the constraint that sets everything downstream.

‍

It determines how much you can spend on R&D, sales and G&A per pound of revenue. Which means it determines CAC payback. Which determines burn efficiency. Which determines runway. Which determines when you have to raise, at what price, and from a position of what strength.

‍

Drop 15 points of gross margin and the effects cascade:

• CAC payback stretches. If you were recovering acquisition cost in 14 months at 80% margin, you're now closer to 17 at 65%. Every cohort takes longer to turn cash-positive.
• Your startup runway calculation is wrong. Not slightly. If AI COGS scale with usage and usage is growing, your burn grows with your success. That's the opposite of how your model behaves.
• Rule of 40 quietly slips. Growth stays the same; profitability doesn't.
• Your valuation multiple is exposed. Investors price software on margin structure. A 52%-margin business and an 80%-margin business are not the same asset class, however similar the ARR chart looks.

‍

The dangerous version of this isn't the founder who knows their margin fell. It's the founder whose blended number still looks fine.

‍

If you have a healthy legacy subscription business at 82% and an AI-heavy layer running at 45%, your blended number might read 71% and nobody panics. Meanwhile the fastest-growing part of your revenue is the least profitable part, and every pound of growth pulls the average down. You'll only find out when the blend crosses a threshold that matters — usually in a diligence process, usually at the worst possible moment.

‍

The fix is boring, which is why it works

‍

You cannot manage what you cannot see, and most early-stage companies have buried AI costs inside a single "software and hosting" line in Xero because that's where the OpenAI invoice landed.

‍

Three moves, in order:

‍

1. Isolate AI COGS now. Break it out as its own cost layer. You want to be able to answer: AI COGS as a percent of AI-related revenue, AI COGS as a percent of total revenue, cost per active AI user, and cost per workflow or request. If you can't answer those, you're flying on instruments you haven't calibrated.

‍

2. Calculate margin by revenue stream, not just overall. Subscription, variable/usage, services — separate revenue lines with separate matched COGS. Blended gross margin hides exactly the problem you need to see. This is the single highest-leverage change most SaaS founders can make to their management accounts this quarter.

‍

3. Watch the tail, not the average. Average cost per user can look completely healthy while a handful of power users torch your margin. Track top-decile user cost against median. AI COGS problems almost never show up in the mean — they show up in the heavy user, the unconstrained feature, and the customer who found a way to run your agent in a loop.

‍

Then price accordingly. That doesn't mean charging by the token — most founders sensibly keep pricing simple early. But if cost scales with usage and price doesn't, you've built a machine that loses more money the better it works. A platform fee plus a usage or credit layer, tiered buckets, soft caps, or heavier AI workflows gated to higher tiers — pick the mechanism that fits your buyer. Just pick one.

‍

What investors actually want to hear

‍

Nobody expects a 2026 AI-native company to run at 80%. What they expect is that you know your number and can show the path.

‍

Better model routing. Caching. Prompt optimisation. Smaller models where they're good enough. Higher ACVs. Better attach rates on paid AI features. Falling token prices doing some of the work for you.

‍

That's a margin roadmap, and it's a far better answer than a made-up 80% that collapses the moment someone opens your ledger. Boards don't need perfection. They need to believe you understand your own economics.

‍

We build models for UK founders that treat gross margin as a set of moving parts rather than a single assumed number — because a financial model for startup planning that can't survive contact with your actual cost data isn't a model, it's a hope.

‍

Not sure what your real gross margin is? Grab a free hour and we'll pull it apart with you — no obligation, no jargon. For founders that don't do boring numbers.

‍

Sources: The SaaS CFO (Ben Murray), "Your AI Feature Is Quietly Destroying Your Gross Margin," thesaascfo.com ; ICONIQ Capital, "2026 State of AI: Bi-Annual Snapshot," iconiq.com ; SaaStr, "ICONIQ's Latest State of AI Report: The 10 Most Important Data Points for SaaS Founders," saastr.com

‍