AI & Pricing

AI is usually discussed as a growth story for SaaS. In the retention data it reads as a risk. The mechanism is a paradox: AI improves gross retention by making software cheaper to run for the customer, and it caps net retention by removing the very units, seats, that expansion has historically been sold in. This is the most forward-looking section of the report. Its claims are tiered Inference rather than Verified, and the AI-cohort cut is a small sample (n=52), directional rather than definitive. Inference

The AI Paradox

AI-native companies run 87% gross dollar retention, better than the market, but only 100% net dollar retention, against 102% for the AI-interested group. The gross number improves for a clear reason: AI lowers the customer’s cost of operating the software, raises its return, and makes it less likely to leave. The net number stalls for the same reason. When AI lets a customer do more with fewer people, the account stops adding seats, and seats are how expansion has traditionally been priced and sold. The dynamic that keeps the customer is the dynamic that stops the account from growing. Lower churn, little expansion on top of it.

Source: KBCM/Sapphire Survey 2025 p21: AI cohort (n=52, blended Native+Enabled); small sample, directionalCross-Survey

The Threat Matrix

Four distinct threats follow from the same shift, ordered by severity: Inference

Figure sources: workforce-reduction selling point 15% (KBCM-2025 p25); AI monetization by subscription 58% (KBCM-2025 p24); evaluation-cycle shift 55% (KBCM-2025 p23, 30 of 55 companies); seat-based pricing 40% (below). The four severity ratings are the report’s own assessment.Verified

The Seat-Pricing Trap

The clearest structural exposure is that the market has held its bet on seats at exactly the wrong moment. Seat-based pricing has sat near 40% of companies across four survey editions (41% in the 2022 and 2024 surveys, a 33% dip in 2023, and 40% in 2025), through the same window in which AI began compressing seat counts and the 2025 edition’s own commentary named the model as at risk of disruption. Under a per-seat model, the arithmetic is direct: every 10% reduction in a customer’s headcount becomes a 10% reduction in what the customer pays, with no renegotiation required. The contraction is built into the contract. Companies have not hedged against AI-driven seat compression; two-fifths of the market remains priced so that the compression flows straight to revenue.

Source: KBCM/Sapphire Survey 2022–2025Verified

The Ouroboros

The threat matrix warns that its four exposures do not share a denominator, and they do not. They share something a grid cannot show: a circuit. Read as parallel threats they merely coexist; read as a loop they compound, because the industry is running the compression on itself.

Trace it. Fifteen percent of companies name workforce reduction as an AI selling point, the lowest-ranked of seven opportunity areas but the one the matrix rates critical, so a slice of the industry is actively selling headcount cuts. Those cuts land on seats: roughly two-fifths of the market prices primarily per seat (33% fixed plus 7% variable), where a customer’s 10% headcount cut becomes a 10% revenue cut with no renegotiation. And the cut arrives fast, because three-quarters of contracts run a year or less (67% annual, another 7% month-to-month), so every renewal re-counts the seats. The vendors on the receiving end respond the only way the cost structure allows: operating expense fell from 118% of revenue to 88% across the actuals, the deepest cuts in the headcount-heavy sales and R&D lines. The reduction the industry sells to its customers is the reduction it then performs on itself.

One station on this circuit is uninstrumented, and it has to be named: the survey never measures who buys the respondents’ software. That the buyer carries its own AI-efficiency mandate is the tightest version of the loop, and in a market where SaaS sells heavily to SaaS it is a reasonable read, but it cannot be traced in this dataset. The survey certifies every station on the circuit; it cannot certify the traffic between them.

The certified stations still close the loop on their own. The cohort selling the compression story hardest is the first to show its cost: the AI-Native and AI-Enabled cohort sits at exactly 100% net dollar retention against 102% for the AI-Interested group, even as it leads on gross retention. AI is keeping these companies’ customers and capping their growth at once. Which is why the hedge that follows is not a defense against a passing trend. It is a way of stepping out of a circuit the industry is otherwise running on itself. Inference

SuccessCOACHING interpretation, not attributable to KBCM/Sapphire. Certified anchors: workforce reduction 15% (KBCM-2025 p25); seat-based pricing 40%, derived 33+7, and contract length 67% / 7% (KBCM-2025 p27); OpEx 118%→88% actuals, 2022–2024 (KBCM-2025 p39); AI-cohort NDR 100% vs 102% and GDR 87% vs 85%, n=52 blended Native+Enabled (KBCM-2025 p21, transposition-corrected). Buyer composition is not measured; the circuit is a statement of mechanism, not of correlation.Inference

The Pricing Hedge

The exposure is structural, so the hedge is structural: move the value metric off the unit AI compresses. Consumption models attach price to work performed, such as transactions, records, API calls, and resolved tickets, so when AI lets a customer do more work with fewer people, usage holds or grows where a seat count would contract. Outcome-anchored models go a step further and price the result the customer bought. Neither is free: pure consumption transfers revenue volatility from the customer to the vendor, and a soft usage quarter is a soft revenue quarter with no contractual floor beneath it. That trade-off is the tension the seat-priced 40% will have to resolve: a committed platform or seat base holds revenue predictability, a usage or outcome component tracks the work AI adds, and the survey measures only the seat share, not how companies are balancing the two.

The survey measures the seat share, the 40% above, but does not certify outcomes for the alternative models, so the direction here is a structural argument, not a measured result. The timing argument, however, inherits the inflection math below: repricing takes sales cycles and renewal windows to propagate through a book, which is why the move belongs inside the 24-to-36-month window rather than after the compression is visible in NDR. Inference

The Inflection Point

The reason this does not yet show up in the headline retention numbers is timing, not safety. AI adoption is still early, and the downsell from workforce reduction lags: it appears at renewal, not at deployment. That lag is shrinking. As the contract data shows, the market is moving to shorter terms, which pulls every renewal, and every seat re-count, forward. The report places the inflection 24 to 36 months out, and projects that as AI reaches critical mass in customer organizations the resulting downsell could push median NDR below 98% and the Net Magic Number below 0.40.

Those figures are a projection, not a measurement, and are labeled accordingly. The structural point beneath them does not depend on the projection being exact: net retention is already at 101%, the seat-compression mechanism is real, and a growing share of the market is priced to absorb its full force. This is the one crisis in the report that has mostly not happened yet. The survey places it 24 to 36 months out. Inference

The survey’s own respondents forecast the opposite. Asked to project the AI-native cohort forward, they put its net dollar retention back at 104% by 2026E, a recovery from the 100% of 2024. That is the tension worth naming: the one cohort most exposed to seat compression is the one whose operators forecast the smoothest rebound. The report’s structural read favors the downside, because the seat-compression mechanism is concrete and the pricing exposure is measured while the 104% is a forward expectation. But the cohort is small, and that thinness should temper confidence in either figure, so the two are best held as a genuine live test rather than a settled call. The scorecard frames the same fork the same way: the data will show within an edition or two whether the operators’ optimism or the mechanism wins. Inference

Forward projection: editorial interpretation, not a measured figureInference