HomeGuides › Vertical-AI gross margin
Vertical AI

Vertical-AI gross margin: why the bar is lower, and the path-up story

A pure-software company that shows a 55% gross margin at seed is behind. An applied-AI company that shows the same 55% is ahead of its bar. Same number, opposite read — because inference is a genuine per-request cost, and a partner who knows the category grades you on a different scale. The mistake is presenting your margin as if you were a SaaS company. Here's the scale you're actually being held to, and how to tell the path-up story so the COGS line reads as a plan, not a flag.

The gross-margin rule used by the IR Narrative engine · ~7 min read

Why the AI bar sits ~15–20 points lower

Classic B2B software has near-zero marginal cost: once the code runs, serving one more customer costs almost nothing, so gross margins settle in the 70–80% range. Applied AI breaks that because every request buys tokens — you pay for inference (your own GPUs or a model-API bill) on each call, and that cost scales with usage rather than disappearing at scale. So the software-grade 70%+ bar doesn't apply. A partner who has funded AI companies knows this and adjusts; the engine encodes the adjustment as a separate ai-vertical margin band rather than grading you against the SaaS bar.

The point isn't that a low margin is fine. It's that "low for SaaS" and "on-track for applied AI" are two different judgments, and you lose the room by inviting the wrong one — leading with a bare 52% and no context lets a partner file you as a weak-margin software business instead of a healthy AI one.

The bars, side by side

These are the "clears vs. strong" bands the engine grades gross margin against. Below the software-grade band on the left; the applied-AI (vertical-AI) band on the right. Read each pair as "the low end is where you stop looking under-margined; the high end is where the number is a selling point."

StageSoftware-grade bandApplied-AI bandGap
Pre-seed— (omit)— (omit)
Seed50–70%40–52%~18 pts
Series A65–75%45–55%~20 pts
Series B70–80%50–60%~20 pts

Two things to read off this table. First, at pre-seed, gross margin is graded omit — you have too little revenue for the number to mean anything, so leading with it wastes a slide (see metrics by stage for the full lead/support/omit map). Second, the applied-AI band still climbs stage over stage. Investors don't expect SaaS margins, but they do expect the trend to bend upward as you scale — a flat 45% from seed to Series B reads as "no operating leverage," which is its own problem.

Where the margin actually goes — a worked COGS build

Say a vertical-AI product charges $400/month per seat and each active seat drives roughly 1,200 model calls a month. Build the COGS line the way a partner will:

LinePer seat / monthNote
Revenue (ACV / 12)$400list price, one seat
Inference (1,200 calls × blended token cost)$96the AI-specific line
Hosting / infra (non-inference)$28storage, egress, orchestration
Support / success allocation$36loaded, not just tooling
Gross margin$240 → 60%(400 − 160) / 400

60% lands inside the applied-AI Series-A band and near the top of the seed band — a healthy number for the category, and one a partner can rebuild from the lines. Notice the inference line is the single biggest cost and the one that moves most with scale, which is exactly why the path-up story hangs on it.

The path-up story (this is the part that de-flags the slide)

A margin number alone is a snapshot. What turns a below-software-grade margin from a flag into a plan is a credible, specific account of how it climbs. Generic "margins improve with scale" is worse than saying nothing — it's the phrase a partner has heard on every under-margined deck. Name the levers you actually have:

What good looks like: "Blended gross margin is 52% today. Inference is 24% of revenue; routing 70% of calls to a smaller model and a 35% cache-hit rate have taken inference cost down for three straight quarters, and we model 62% blended margin at 5× current volume." That's a number, a driver, a trend, and a target — a partner can underwrite it.
What kills the slide: a bare "55% gross margin" with no COGS line beneath it, or "margins will improve as we scale" with no lever named. Both read as either you don't know your unit economics or you're hoping the problem solves itself.

How the engine presents it

The gross-margin presentation rule is explicit: show the % with the COGS line beneath it, and if it's below software-grade (AI inference, payments, hardware), pair it with the explicit path up as you scale. The report grades your margin against the applied-AI band for your stage, tells you whether you clear the "on-track" line or the "strong" line, and — when you're below software-grade — flags whether your slide carries a path-up story or leaves the number naked. It's the difference between a partner filing you as a healthy AI business and a weak software one.

One honest caveat: these bands are institutional consensus for US venture, not a law. A vertical with unusually heavy inference (long-context, agentic, video) can sit below the band and still be fundable if the ACV and retention justify it; a thin-inference product can beat the band. The engine grades against the consensus band and says so — treat it as the default a partner starts from, then argue your specifics.

See where your AI margin lands vs. the bar

The report grades your gross margin against the applied-AI band for your stage and model, and flags a naked margin before an investor does.

Request the report — $490 See a full sample