Your AI Feature Is Eating Your Gross Margin

It's the quarterly business review.
Slide four is gross margin.
Down two points.
The CFO stops the room: "Why?"
Heads turn to infrastructure. Cloud costs get blamed. Someone promises a vendor renegotiation.
Nobody looks at the AI feature you shipped in Q1.
The one that went out "included free." The one customers love. The one using AI inference on every session.
Adoption has doubled since launch. Nobody in the room connects the two slides.

And still — those two points came from somewhere.

It feels like a cost problem. Infrastructure's job. Finance's spreadsheet.
It's not.
It's a pricing decision — one your team made without noticing they were making it.
The day the feature shipped bundled with the main product, its price was set: zero.
Nobody signed off on zero. Zero just happened.

AI added a COGS line to software that never had one — and most product teams priced it at nothing.

Not because they ran the unit economics and chose zero.
But because "everyone bundles it," and parity felt safer than a pricing conversation.
Parity is not a pricing model. It's the absence of one.

Understand what actually changed, because it's significant.
Classic SaaS had one beautiful property: the marginal cost of serving one more user rounded to nothing.
That is where the ~80% gross margins came from.
It's what your valuation multiple was built on. It's why software ate the world profitably.
AI inference breaks this rule.
Every AI interaction costs real money. Usage up, cost up. Forever.
Not a one-time build cost you amortize and forget.
A running meter, attached to your most engaged users.
And the industry bill is already visible: SaaS gross margins have slid from roughly 80% to 50–60% — inference cost added without matching value or price.
Your product decisions now go directly into COGS.
Nobody told product leaders that.
Finance got it. Investors got it. The board reads it in every portfolio review.
At 80% gross margin, product could afford to ignore cost of goods.
At 60%, product is the cost-of-goods conversation.

One product leader asked me the honest version of this: AI features are more costly, so how much more care is needed before adding them?
More care, yes. But not the kind people mean.
Not another feature and architecture review before or during implementation.
A pricing decision, made at the same table as the implementation decision, by the same people.
The care isn't caution. It's economics.

Here's the part your CFO already knows.
Profit margin is the leading indicator. Rule of 40 is the lagging one.

Product leaders confuse the two — protecting the aggregated score of Rule of 40 while the input quietly slips.
Which means the CFO sees your AI feature eating margin quarters before it dents the number the board tracks.
You won't be in the room when the pattern gets spotted.
You'll just get the question. On slide four. In front of everyone.
There's a version of that moment where you have the answer ready.
It requires having asked the question first.

Now take the two questions every product leader is asking this year.

Both usually get answered by looking sideways — at competitors, at the market, at whoever shipped their AI announcement last week.

Wrong direction.

Bundle-versus-charge is not a feature decision. It's a margin and capital-allocation decision.

Pricing by parity answers one question: what is the market doing?

Pricing by unit economics answers three harder ones:

If value lands and you can show it, the capability can carry a price, a tier, a meter.

If you can't answer those three, bundling isn't generosity.
It's an unfunded liability accruing on every session of your product.

A mispriced AI add-on can cost more than it earns.

Not as a rounding error. As a feature with negative gross margin, scaling with adoption.
The better it performs, the worse it gets.
Success becomes the problem — every new enthusiastic user deepens the hole.

And notice what parity pricing can't see.
Parity tells you what a competitor charges. It says nothing about what the feature costs them — or you.
Their margin structure isn't yours. Their model contracts aren't yours. Their runway isn't yours.
Copying a price without copying the economics underneath it isn't strategy. It's a naive hope.
To be clear — this is not an argument to price off cost.
You price against value. You always price against value.
But the floor moved. Zero marginal cost made the floor theoretical.

Inference makes the floor real — and a price below a real floor isn't a pricing strategy. It's a subsidy with no end date.

Watch the difference in the room.

One product leader tells the QBR: "AI adoption keeps climbing. Customers love it. It's becoming a differentiator in deals."

The other says: "The AI capability now costs more to serve than we charge for it. We're moving it to a priced tier for the accounts consuming the most inference — here's the margin impact and the churn risk we're accepting."
The first is celebrating usage.
The second is allocating capital.
Only one of them gets trusted with the next pricing decision.

The pattern is bigger than one feature.
Product teams treat pricing as someone else's room — a finance artifact bolted on after the build.
So the largest economic decision attached to an AI capability gets made by default.
By parity. By silence. By the ship date.
Defaults are still decisions. They're just decisions with no owner.
The feature decision was debated for weeks.
The pricing decision attached to it was never held at all.

Keep that default and the sequence runs on its own.
Margin drifts. The CFO investigates. The AI line gets found.
And the remediation is far uglier than the original decision would have been.
Repricing a capability customers already have reads as a take-away. Churn risk enters every renewal conversation.
Leadership pushes back. Decisions stall. Momentum fades.
Worse: your next AI proposal gets read through the margin hole the last one left.
You're no longer the leader who ships AI. You're the leader whose AI costs money.
That reputation compounds too.

The board hears it. The CFO repeats it. Your next headcount ask meets it at the door.

From "AI as a feature decision" — To "AI as a COGS line with a pricing decision attached."

Every AI capability now arrives in two parts.
The part customers see. And the unit economics underneath it.
Ship the first without deciding the second and you haven't postponed the pricing decision.
You've made it — at zero, with no analysis, in writing.
Because a price customers can see is the most public analysis you'll ever publish.

Ask yourself this week: for the AI capability we shipped most recently, do I know what it costs to serve — and did anyone decide, on purpose, that its price is zero?

If the answer is silence, you already know what slide four looks like next quarter.
The margin story is being written either way.
Your CFO sees the erosion before you do. The only question is whether you arrive with the pricing decision — or the excuse.

A question to think about:

Hit reply and tell me — I love hearing your thoughts.

Talking about your product in margin, cost-to-serve, and unit-economics terms — the CFO's language — is the type of decision-making work I develop further with product leaders inside my cohort.

From PM to Product Executive — a Maven cohort for senior PMs, Directors, and VPs learning to think and decide like product executives.

Next cohort starts July 27.

Until next time,
Elena Leonova

Maven cohort: https://maven.com/elena-leonova/from-pm-to-product-executive
LinkedIn: https://www.linkedin.com/in/elenleonova
Website: https://elenleonova.com