The gross margin of a business is the test of whether each additional unit of revenue costs less to deliver than it earns. AI changes that test for some businesses, leaves it unchanged for others, and creates accounting illusions for a third group that the CFO must be alert to.
The framework below distinguishes the cases. It is not a forecast of AI's eventual margin impact. It is the working analysis a senior finance leader uses to decide whether margin movement attributed to AI is real, durable, and actionable.
What gross margin actually measures
Gross margin is revenue less the direct cost of generating that revenue, expressed as a percentage. The direct cost is the cost that varies with volume — the marginal cost of the next unit. Fixed costs, by definition, sit below the line.
The CFO's first discipline in AI margin analysis is to be precise about what costs are above the gross-margin line and what costs are below it. A new AI capability that compresses a fixed-cost activity — say, the corporate finance team — does not, by definition, change gross margin. It changes operating margin further down. The distinction matters for forecasting, for benchmarking, and for the narrative the entity tells the market.
Where AI moves the variable cost line
AI genuinely affects gross margin in businesses where the variable cost per transaction includes meaningful labour, judgement, or data processing — activities AI can perform more cheaply at marginal volume.
- Customer support per transaction. A business whose unit economics include support cost per customer can compress that cost materially with appropriate AI deployment. The compression flows into gross margin if support is in cost of revenue.
- Personalisation and pricing. A platform that adjusts price or experience per user, where the adjustment historically required human work, can deliver personalisation at materially lower variable cost.
- Content generation per transaction. A business that produces unique content per customer or transaction — listings, descriptions, recommendations — can produce more of it more cheaply.
- Fraud and risk screening. A business that screens each transaction for fraud or risk can replace some manual review with automated screening, with the labour cost moving from variable to fixed and the per-unit cost falling.
- Operational matching. A platform that matches supply and demand — bookings to inventory, products to buyers — can perform the matching at lower marginal cost with appropriate AI.
In each case, the test is whether the cost being reduced is a per-unit variable cost, whether the reduction is durable, and whether the AI cost displacing the previous cost is itself fully captured.
Where AI does not move gross margin
Several categories of business will see less or no gross margin expansion from AI, even where the technology is genuinely deployed.
Businesses where the variable cost per unit is dominated by physical goods, logistics, payment processing, or licensed inputs will see limited gross margin movement from AI. The technology may make the back office more efficient, but the back office is not in the gross margin line.
Businesses where the variable cost is already minimal — software-as-a-service at scale, for example — have less room to compress further. The marginal benefit of AI is more likely to appear in growth, in retention, in expansion revenue, than in gross margin per existing transaction.
Businesses where the AI investment itself materially increases the cost of goods sold — through compute, through licensed model access embedded in the product — can see gross margin decrease before it improves, particularly during the build-out phase.
The accounting trap: capitalisation and the apparent margin
One of the more subtle distortions of AI on reported margin is the capitalisation of development cost. Costs incurred building an internal AI capability may be capitalised as an intangible asset under IFRS or US GAAP, depending on the facts. Capitalised costs do not hit the income statement immediately; they are amortised over the useful life.
A business that capitalises AI development and reports operating costs that exclude the development effort can show improving margin during the build that reverses, partially, when the amortisation begins. The cash position tells a less flattering story than the margin line.
The CFO's discipline is to disclose the capitalisation policy explicitly, to present margin both inclusive and exclusive of the AI development effort in management commentary where material, and to ensure that the board's view of the underlying economics is not distorted by the capitalisation accounting.
The pricing question that follows the cost shift
When AI compresses a variable cost, the margin expands only if the price holds. Competitive markets do not generally allow the price to hold indefinitely against a cost reduction available to all participants. The expanded margin is, in many cases, transitional.
The CFO must distinguish:
- Margin expansion that is durable because the AI deployment produces a genuine cost advantage that competitors cannot replicate quickly.
- Margin expansion that is transitional because the AI deployment is ahead of competitors today but will be matched, with the margin then competed away.
- Margin expansion that is accounting rather than economic, driven by capitalisation, attribution, or comparison to inflated prior-period costs.
The board's expectations should be set against the durable case, not the transitional one. The market will eventually distinguish, and the entity that promised durable margin expansion that turned out to be transitional will be marked accordingly.
A CFO's test for genuine margin expansion
A useful test for any margin improvement attributed to AI has the following steps.
- Identify the specific variable cost line that moved, in monetary terms, per unit of revenue.
- Identify the corresponding AI cost — full-stack, including compute, infrastructure, governance, and refresh — that displaced the prior cost.
- Calculate net cost movement at the unit level, not at the aggregate.
- Test the movement against alternative explanations: pricing change, mix shift, volume effect, capitalisation, reclassification, currency.
- Project forward: what does the AI cost line do at the next scale step, and what does the displaced cost do if volume grows or shrinks?
- Compare against the counterfactual: where would the cost line be if AI had not been deployed and the prior trajectory had continued?
The output is a margin attribution that the CFO can defend, to the board, to the auditor, and to the market. It will not always show the headline number the technology team is reporting. It will show the number that survives examination.
This piece sits inside the CFO in AI framework. See also the full-stack cost of AI and the CFO as AI capital allocator. Lorna writes from practice at IMPT. The verified page records what is and isn't published here.
Lorna Mason is CFO of IMPT, Dublin. The verified public record is on the Verified page. Contact: lorna@impt.io