By Geoff McDonald, CEO, Ambassador
The 2026 benchmark data came out and most people are going to read it as a recovery. Efficiency up, payback shorter, profitability better. Boards will celebrate. Bankers will use it to justify multiples.
Read one layer down and it is not a recovery at all. It is a split, and the two halves are moving in opposite directions.
Here's the short version: AI made software companies dramatically better at building and selling. It did not make them better at keeping anyone. Retention fell across the whole market while efficiency posted its best year in five, and almost nobody is connecting those two facts.
The 2026 SaaS and AI Performance Benchmarks, published jointly by Aleph and Benchmarkit on June 1, 2026, drew on 342 B2B SaaS and AI-native companies and reflects full-year 2025 actuals. On the efficiency side, it is genuinely good news.
Median CAC payback improved from 18 months to 16. That is an 11% gain year over year and the largest single-year improvement in the four-year trend the report tracks. Rule of 40 jumped from a 15% median to 25%, the biggest single-year gain in five years of benchmarking. ARR per employee is reported at $175,000, up 17%, driven almost entirely by AI-assisted engineering productivity.
If you stopped reading there, you would conclude that the industry has figured it out.
Gross revenue retention fell from 88% to 84% at the median.
Four points. Market wide. And the companies that are supposed to be insulated from market-wide pressure did not escape it either. The 75th percentile slid from 95% to 91%, which means the 95% gross retention figure that has functioned as the gold standard for a decade is now a top-quartile stretch goal rather than a competent baseline.
Growth decelerated for a fourth consecutive year in the same dataset, from a 30% median in CY-22 to 20% in CY-25.
So the picture is this. We got cheaper at acquiring customers and faster at building product, and we simultaneously got worse at holding onto the revenue we already had. Those are not two separate stories. They are the same story told from both ends.
Think about where the last two years of AI investment landed inside most software companies.
It went into engineering. Code assistants, test generation, faster shipping. That is the ARR-per-employee number.
It went into acquisition. Better targeting, automated outbound, content at volume, cheaper pipeline. That is the CAC payback number.
Almost none of it went into the part of the business that decides whether a customer is still there in eighteen months. Renewal risk, expansion timing, advocacy, the specific question of what to do next with a specific account. That work is still mostly a human reading a dashboard on a Tuesday and making a guess.
There is a reason the easy wins came first, and it is not laziness.
Acquisition is a fast-feedback problem. You spend, something happens within days, you adjust. That shape is ideal for automation because the loop closes quickly enough to learn from.
Retention is a slow-feedback problem. The behavior that predicts a churn in March shows up in November, buried in product usage, support tone, invoice timing, who stopped logging in, who stopped referring anyone. By the time the outcome is visible the decision window has closed. You cannot optimize your way out of that with a faster dashboard, because the problem was never speed of reporting. It was that nobody knew which of four thousand accounts to look at, or what to do when they got there.
That is the actual job. Not more reporting surface. A shorter distance between a signal and a decision.
If you need a reason to care beyond the operating pain, the market has already priced it.
McKinsey's analysis of more than 100 B2B SaaS companies found that companies in the top quartile of valuation multiples carried a median enterprise-value-to-revenue multiple of 24x, against 5x for bottom-quartile peers. The retention gap between those groups was 113% net revenue retention versus 98%.
Fifteen points of retention. Roughly a fivefold difference in multiple.
Now put that next to a market-wide four-point drop in gross retention and a top quartile that lost four points of its own. The asset class did not get cheaper because growth slowed. It got cheaper because durability slipped, and durability is the thing multiples are actually paying for.
Here is what I think happens over the next two years.
The efficiency gains from AI are going to keep arriving, and they are going to stop being a differentiator, because everyone gets them. When every company ships faster and acquires cheaper, neither one is an advantage anymore. They become the new floor.
What will not commoditize is knowing what to do next with the customer you already have. That requires the acquisition data and the retention data to live in the same place, which for most companies they do not. Ad spend sits in one system, product usage in another, support in a third, and the reconciliation is somebody's spreadsheet.
The companies that close that gap will hold retention while everyone else watches it slide, and they will get paid for it on a multiple.
The 2026 benchmarks are not a recovery story. They are a bifurcation story.
Efficiency improved because AI was pointed at production. Retention fell because it was not pointed at customers. If you are planning next year off the headline numbers, you are planning off the half of the report that is going to matter least.
Go find your own gross retention number, and then find out who owns it. In most companies the second question is harder than the first, and that is the whole problem.
Ambassador is the Customer Lifecycle Operating System. GROW. KEEP. PROVE.
Sources