Software had a good week in the market. On October 6 the S&P 500 software and services index rose 1.3 percent to its highest level since November 2025, and the analysts explaining the move leaned toward a more comforting story. AI looked more like an enabler than the disruptor feared during the selloff earlier this year. One technology researcher told Reuters that the feared “SaaSpocalypse” had not happened nearly as fast as some on Wall Street expected, and the sector’s expected annual earnings growth rate for 2026 had been revised up to 20.6 percent, from 13.8 percent at the end of March (Reuters, via Investing.com).

I read a different number the same week, and it points the other way. Benchmarkit’s 2026 benchmark reports that median gross revenue retention in its GRR sample (n=226) fell to 84 percent in CY-25 from 88 percent in CY-24, with the 75th percentile sliding from 95 to 91 (Benchmarkit, 2026 B2B SaaS and AI-Native Metrics). Gross retention is one of the cleanest reads on whether your existing book is holding, because it strips out expansion and counts only what you kept. At a median of 84 percent, the benchmark implies a 16 percent annual loss of existing ARR to churn and contraction before expansion.

Both things are true at once. The market is pricing in resilience, and the operating data underneath is eroding. That gap is the subject of this piece.

What the multiple cannot see

A stock multiple is a belief about the future. It does not know whether your customers renewed at the same dollar value, and it cannot see the shape of the retention curve beneath a blended growth rate. Net revenue retention hides this especially well. When expansion is strong, NRR can sit comfortably above 100 percent while the base quietly shrinks, because gains from healthy accounts mask contraction and loss elsewhere. ICONIQ’s 2026 State of Scaling makes the contrast visible: its Pacesetter cohort—AI-forward companies with at least top-quartile growth for their scale—reports net retention of 120 to 146 percent, well above the software norm (ICONIQ, 2026 State of Scaling). In a market this bifurcated, an average tells you very little about your own business.

Benchmark medians have the same limit. They tell you the market moved. They do not tell you why your book did, which turns out to be the whole game.

One number, three different problems

Here is the move that costs most leadership teams. A gross retention number comes down, and it gets labeled a churn problem in the next board deck. Churn is only one of the things that produces that number, and the label quietly decides the response before anyone has checked whether it is the right one.

A four-point decline in gross retention can come from at least three different places, and they do not share a fix.

Diagnostic frame

One number. Three different problems.

A four-point GRR decline can come from different mechanisms. Each calls for a different response.

  1. Mechanism

    Customer loss

    Value delivery and renewal

    Customers leave, or cut spend because the product is not delivering enough value to defend at renewal.

    Where to lookValue delivered before renewal

  2. Mechanism

    Seat rationalization

    Pricing and usage economics

    The customer stays, but a leaner team needs fewer licenses and the billed unit no longer tracks the value delivered.

    Where to lookA unit of value that still grows

  3. Mechanism

    Competitive displacement

    Positioning and differentiation

    Buyers consolidate overlapping tools or compare the vendor with an AI-native alternative or an in-house build.

    Where to lookThe work only you remove

One is genuine loss. Customers leave, or they cut spend because the product is not delivering enough value to defend at renewal. This is the explanation the word “churn” assumes, and when it is the real driver, the work lives in value delivery and the renewal motion.

Another is seat rationalization. The customer can stay and still spend less because its team got leaner and no longer needs as many licenses. Revenue can fall at renewal with no logo lost and satisfaction not being the primary issue. As organizations use agentic AI to complete work across systems, the number of users interacting with traditional software interfaces may fall as well (Gartner, July 2026). When this is the driver, the question is not how to make customers happier. It is whether your pricing is tied to a unit of value that still grows, or to a headcount that is now falling.

A third is competitive displacement through consolidation. Buyers can collapse overlapping tools and negotiate renewals harder as they weigh each vendor against an AI-native alternative or an in-house build. Worldwide software spending is forecast at $1.433 trillion in 2026, up 14.7 percent from 2025, so the category is still growing; that does not tell you which vendors will capture the spend (Gartner, February 2026). Vendors that cannot show what work they remove may learn that at renewal. When this is the driver, the first work is positioning and packaging, not an undifferentiated retention push.

That is one reported number sitting on top of three mechanisms, each of which may sit in a different part of the business and each calling for a different response. Run an excellent save motion against a problem that was actually seat rationalization, and you will spend real money and goodwill solving something that was not primarily about satisfaction.

Why four points is not a rounding error

The economics are worth making concrete. Take a company with a $50 million ARR base and hold expansion aside. At 88 percent gross retention it keeps $44 million of that base each year. At 84 percent it keeps $42 million. The four-point difference is $2 million of retained ARR in that period. If the lower retention persists, next year’s expansion builds on a smaller base, and the new logos added to backfill the hole carry acquisition costs that retained revenue never did. Those figures are illustrative, not a benchmark for any specific company. The shape holds regardless of the inputs: a retention number that slips a few points can read as a rounding error inside a growth story, when it is the base the entire growth story stands on.

My read

After years running post-sale and service organizations, my view is that the retention number is often not the diagnosis. It is the summary of several problems that live in different functions, and it looks like one problem only because it arrives as one line on a dashboard. The instinct to respond fast, usually with more customer success headcount or a sharper save motion, can feel like leadership and still spend the most valuable resource a team has, its attention, in the wrong place.

I also think gross retention is the number executives should be watching right now, more than net. Net retention can flatter you in a market where expansion is carrying the result, and flattery is the last thing you want from a metric in a year when the base is under pressure. If your NRR looks fine and you have not cut GRR by segment lately, you may be healthier on paper than in the book.

On pricing, the signal in the benchmark data is hard to ignore, but it is narrower than a GRR claim. Benchmarkit’s 2026 cut puts seat-only pricing at 98 percent median NRR versus 108 percent for usage-based models. I do not read that as a fad or a licensing trend. The unit that per-seat pricing charges for—a human logged into software—can come loose from the value the software delivers as leaner teams and automated workflows do the same work with fewer people in the tool. When the thing you bill for stops tracking the thing you deliver, the model can become a contributor to retention pressure; no amount of relationship work repairs a price pointed at the wrong unit.

How I work this problem

This is the work I built Lumaire to do. Your dashboards tell you what happened. Lumaire helps you understand why. The retention line gives you the outcome. It will not tell you which of the three mechanisms is driving it, and the response you choose is only as good as that distinction.

My method is a single lens I apply to any performance question. Start with the outcome, here the gross retention decline. Separate the drivers, the specific mechanisms producing it, using evidence the company already has. Identify the levers leadership can actually move, which differ sharply by driver. Then size the impact, so effort goes where it changes the economics most. I diagnose before prescribing, and the diagnostic stands on its own. It is scoped to a defined outcome, it produces a defensible answer, and it never assumes you buy anything after it.

For a retention decline, that means pulling GRR apart by pricing model, by segment, and by downgrade versus full logo loss, then reading the pattern against renewal timing, product usage, and the commercial terms underneath each account. Cut correctly, the data can point to its dominant cause. What leadership gets is a clear answer to which problem it actually has, and the evidence behind it, before a dollar of effort goes toward fixing the wrong one.

The window

The market will keep telling its story, and it may well be right about direction. Direction is not the same as timing, and neither one tells you what is happening inside your own retention curve. The companies that come out of this period ahead will read their own numbers honestly while the narrative is still comforting, and act on the specific driver instead of the general worry. Clarity creates momentum, and it starts well before the market makes up its mind, with a leadership team that knows which of the three problems it actually has.

Jennifer Williams is the founder of Lumaire, a business performance advisory. lumaireco.com

Sources