KeyBanc Capital Markets just published the 17th edition of its private software benchmark. For years it has been one of the better sources of private B2B data. This year KeyBanc added a large AI section covering pricing, AI costs, agents, internal adoption, and whether customers are building their own replacements.

keybanc-report-cover

The survey ran from July through September 2026. 81 companies took part, and 62 of the answers came from CFOs and CEOs. Median ARR was $22M in 2025, so most of the sample is $10M-$50M B2B companies.

The sample is small. Some of the cuts below rest on 4 to 15 answers, so read the medians as directional.

The top takeaways:

  • The median company grew 24% in 2025, plans 23% in 2026, and plans 40% in 2027. Only 6% met the Rule of 40 in 2025. 62% plan to meet it in 2027.
  • 43% of vendors say customers are already using AI coding tools to rebuild their product or features. At companies under 100 employees, it’s 55%.
  • Only 41% charge for AI. The most common model is bundling it into the core price. Where it is charged, the median premium is 20%.
  • Only 8% report seat compression from AI.
  • 30 of 68 companies are over budget on AI spend, mostly because token usage beat the plan. Companies spending more than 5% of revenue on AI run 66% gross margins. Light spenders run 78%.

1. The Median Plan Goes From 23% Growth in 2026 to 40% in 2027

The median trajectory, using management’s own forecasts:

  • ARR growth: 24% (2025) → 23% (2026E) → 40% (2027E)
  • EBITDA margin: -3% → 1% → 8%
  • Rule of 40: 16% → 20% → 44%
  • Share of companies at Rule of 40: 6% → 18% → 62%

kbcm-growth-plan-2027

kbcm-rule-of-40-share

The median company plans to speed growth up by 17 points and add 7 points of EBITDA margin in the same year. That would follow a 2026 with flat growth.

These operators did beat their own plan last year. Last year’s survey forecast 20% growth for 2025, and the actual was 24%. EBITDA was planned at -4% and came in at -3%. A 4-point beat is still a long way from a 17-point acceleration.

The 12 companies with “very high” confidence in their forecast plan about 30% growth for 2027. The companies with “moderately high” or “average” confidence plan 40-41%.

The GTM side of the plan assumes every efficiency metric improves together:

  • New-customer ARR per AE: $411K → $450K → $487K
  • Quota attainment: 70% → 75% → 80%
  • Gross magic number: 0.76 → 0.79 → 0.97
  • CAC payback: 21 → 19 → 17 months

Meanwhile, 84% say AI has had no material effect on how long deals take. 9% say cycles got shorter, and 7% say they got longer. Median cycles are 2 months in SMB, 3 in mid-market and 8.5 in enterprise. KeyBanc titled that slide “Buyer Process, Not Vendor Efficiency, Sets Deal Pace.”

On one of these boards, I’d underwrite the 30% the high-confidence operators plan. Then I’d ask management for a quarter-by-quarter bridge to the other 10 points.

2. 43% of Vendors Say Customers Are Rebuilding Their Product With AI Coding Tools

43% of respondents report customers using AI coding tools to recreate their product or features. Exposure falls with company size:

  • Under 100 employees: 55%
  • 100-249 employees: 40%
  • 250+ employees: 25%

By sector, vertical software is the most exposed at 48%. Horizontal and infrastructure are both at 39%. KeyBanc analyst Jackson Ader noted that 43% is far more than the share of customers building their own versions 5 years ago.

kbcm-customers-vibe-coding

When asked for their single biggest AI threat, respondents ranked:

  1. Incumbent competitors shipping AI features faster (15)
  2. AI reducing the need for the product (12)
  3. Customers vibe coding (11)
  4. AI-native startups (10)
  5. Hyperscalers (5)

AI-native startups ranked fourth, behind customers needing the product less and customers building it themselves.

The churn data points the same way. In 2025, the median company lost 10% of revenue: 7 points from customers leaving entirely and 3 points from downsell. Ader ties part of that full-logo churn to customers trying AI-native alternatives or building their own.

Contract length has the biggest effect on churn in the data. Median gross dollar churn by initial contract term:

  • 12 months: 9%
  • 24 months: 7%
  • 36+ months: 4%

kbcm-churn-by-contract-term

I use Replit every day and have built production apps on it without an engineering background. A decent internal version of a $20K/year point tool takes a weekend now. The vendors that are hard to replace hold proprietary data, run deep inside workflows, or own integrations nobody wants to rebuild. 64% of vendors in the survey build on proprietary or permissioned data. KeyBanc tells investors to focus diligence on “difficulty of replicating core functionality.”

3. Companies Charging for AI Grew 27%. Companies Bundling It Grew 17%.

How AI is monetized today:

  • 41% charge for AI products
  • 29% are testing or piloting monetization
  • 30% don’t charge for AI separately

The main pricing model, out of 69 respondents:

  • Included in core product price: 31
  • Add-on / separate product: 16
  • Hybrid: 12
  • Usage or consumption: 4
  • Outcome-based: 1

The AI price premium has a median of 20%, a 25th percentile of 4%, and a 75th percentile of 26%. KeyBanc’s summary: “customers have not settled on what AI is worth.”

Growth by monetization stage:

  • Charging for AI: 27% growth in 2025, 30% planned for 2026
  • Testing / piloting: 24% → 22%
  • Not charging separately: 17% → 21%

kbcm-growth-by-ai-monetization

AI-driven ARR shows the same split. Companies with no AI-driven ARR grew 11% in 2025. Companies with up to 25% of ARR from AI grew 31%.

Net retention sits at roughly 100-104% in all three groups, so charging for AI hasn’t moved retention yet. Some of the growth gap is likely selection: companies with AI good enough to charge for probably had stronger products to begin with.

AI-driven ARR was 4% of total ARR in 2025. The plan is 12% in 2026 and 17% in 2027. Planned 2026 growth in AI ARR is 128%, up from 74% in 2025. Only 21 companies answered that question, and KeyBanc tells investors to “diligence how AI-driven ARR is defined.”

4. Only 8% of Companies Report Seat Compression From AI

I’ve argued all year that seat-based pricing is under pressure. This group doesn’t show it yet.

  • 92% report no seat compression from AI. 8% do.
  • Fixed seat-based is the most common model in use (33 of 64 companies). Usage/consumption is close behind at 30.
  • Of the 30 companies moving to usage-based pricing, 53% are piloting, 27% have partially moved, and 20% have fully moved.

Ader again: “‘Consumption’ can be a scary word for a CIO or CFO.”

This is a mid-market group with a $22M median. Seat compression shows up first at large platforms with very large seat counts in support, sales and service, where one agent can replace a team. A $20M ARR vendor selling 40 seats to a mid-market customer hasn’t seen those cuts yet. Most companies are adding usage pricing on top of seats while keeping the seat base.

Companies whose core product is AI lean further toward usage: 38% of their pricing models are usage/consumption, against 25% for companies embedding AI.

5. 30 of 68 Companies Are Over Budget on AI Spend. Only 4 Are Under.

Year-to-date AI spend against budget:

  • More than 20% over: 16
  • 0-20% over: 14
  • Roughly in line: 34
  • Under: 4

kbcm-ai-spend-vs-budget

Of the 33 companies that explained their overruns, 25 cited higher-than-expected token usage and 16 cited more adoption of AI features. Per-unit inference cost was flat for 47 of 67 companies. The overruns came from volume.

AI spend intensity against the P&L (2025 medians):

AI spend as % of revenue Under 2% 2-5% Over 5%
ARR growth 24% 23% 17%
Gross margin 78% 72% 66%
ARR per FTE $205K $174K $159K
Opex per FTE $171K $166K $83K

The over-5% group is only 4 to 7 companies. In 2025, its AI spend showed up as a 12-point gross margin gap without a growth payoff. Those companies also run at half the opex per employee of everyone else, and still produce less ARR per employee.

Companies selling AI as their core product compared with companies embedding AI in a broader product:

  • Growth: 34% vs 22%
  • EBITDA: -12% vs -1%
  • ARR per FTE: $117K vs $201K
  • Spending more than 5% of revenue on AI: 38% vs 17%

Half of the Core AI companies report a last valuation of 16x revenue or more. Under a third of all other companies do.

On gross margin overall, 16 of 56 companies say AI has hurt it, 22 say no effect, and 18 say it improved. We’ve seen the same split in public companies this year. Figma took the AI cost in gross margin. Box, which sells the content layer underneath the AI rather than buying inference, barely moved.

6. 91% Use AI in Engineering. 65% Have Cut Zero Headcount Because of It.

Internal AI use by function:

  • Engineering 91%, customer support 63%, sales 63%, back office 61%, IT 56%, marketing 54%
  • 24 of 40 companies say AI writes half or more of their new code (2 say all of it)
  • 92% say AI has shortened development cycles
  • R&D as a share of revenue is planned to fall from 29% in 2025 to 24% by 2027

65% report no AI-related headcount reduction or reallocation. Planned cuts over the next 12-24 months are concentrated in R&D (45%), customer service (37%) and marketing (21%). 16% expect cuts in sales.

Jason Celino at KeyBanc says companies are putting the engineering savings into shipping faster, and few are taking them to the bottom line.

32% have no formal way of measuring AI productivity, and fewer than a third track any financial metric. Companies running 6+ internal AI use cases had $188K ARR per FTE in 2025, against $141K for companies with 0-5. They plan $295K by 2027.

We run SaaStr with 3 humans and 20+ agents in production, across events, sales, marketing and support. Most of our leverage came from agents outside engineering.

7. Bootstrapped Companies Posted a 36% Rule of 40. VC-Backed Companies Posted -3%.

The bootstrapped group is only 7-9 companies. The 2025 numbers:

VC-backed Bootstrapped
ARR $19M $24M
ARR growth 25% 24%
EBITDA margin -13% 13%
Rule of 40 -3% 36%
Gross magic number 0.73 1.08

Growth was nearly identical. EBITDA margin differed by 26 points. The bootstrapped companies added about 48% more net new ARR per dollar of sales and marketing.

The financing data is tough. 16 of 38 companies had less than 12 months of runway at their last VC or growth raise, and 27 of 38 had less than 24 months. Recent raises cluster at modest multiples, and much of the 16x+ group still rests on 2021-2022 marks. 27% say they’re open to being acquired. Active acquirers are mostly in the >$50M ARR group.

5 More Interesting Learnings

  • New logos carry the growth plan. New customers were 56% of net new ARR in 2025 and are planned at 63% in 2026. Median NRR is 102%, planned at 101% in 2026.
  • Infrastructure software is accelerating. It grew 33% in 2025 and plans 54% for 2026. Horizontal software went from 19% to a planned 17%. KeyBanc’s Eric Heath credits data-stack and cybersecurity spending pulled along by AI.
  • Customer-facing agents are mostly supervised. 48% of vendors offer semi-autonomous agents, 33% copilots, and 20% fully autonomous agents. 43% require human approval for agent actions.
  • 57% of vendors use multiple models for customer-facing AI. Model strategy made almost no difference to net retention.
  • 8% of vendors report broad, scaled customer deployment of their AI features. ROI uncertainty and data privacy tie as the top adoption barriers (24 each), with budget next (22).

Related Posts

Pin It on Pinterest

Share This