How B2B SaaS prices AI — credits beat add-ons, premiums hide
Should your AI feature cost extra — and if so, how? We read the pricing pages of 286 B2B software companies and checked every AI mention by hand, so you can see how 106 of them actually charge for AI: by the credit, as an add-on, bundled into every plan, or not priced at all.
Executive summary
AI is everywhere on B2B SaaS pricing pages (202 of 286, 70.6%) but the AI price usually is not. Of 106 hand-coded pages, 15 (14.2%) mention AI only as marketing — nav links, hero copy, testimonials — with no priced offer attached. Among the 91 genuinely priced AI offers, metered credits dominate (32, 35.2%), bundled-in-every-plan AI is second (23, 25.3%), and the classic flat add-on trails (11, 12.1%). No per-seat AI add-on was observed once.
The headline for buyers: an "AI premium" is almost never a number you can read off the page. Only 3 of 91 offers state an AI price against a priced base tier (20%, 31%, 60%) — everywhere else AI is metered inside usage pricing, bundled across plans, or reserved for higher tiers without a separable charge. Any index that claims a market-wide AI premium from public pages is inventing it.
The seven ways AI is priced (and the eighth, which is marketing)
| Model | Offers | Share | 95% CI |
|---|---|---|---|
| Metered credits / tokens AI sold by the unit: credits, tokens, prompts, minutes, actions | 32 | 35.2% | 26.1-45.4% |
| Included in every plan AI present from the free or entry tier up, no separate charge | 23 | 25.3% | 17.5-35.1% |
| AI is the product No separable AI premium: the product itself is AI, sold per seat or by usage | 15 | 16.5% | 10.3-25.4% |
| Flat add-on Fixed-price AI add-on bolted onto any plan | 11 | 12.1% | 6.9-20.4% |
| Reserved for higher tiers Full AI gated above the entry plans | 8 | 8.8% | 4.5-16.4% |
| Separate AI plan Standalone AI product plan beside the base product | 1 | 1.1% | 0.2-6.0% |
| Announced, no price AI agent unveiled; request-a-quote only | 1 | 1.1% | 0.2-6.0% |
Worked examples per model
- Metered credits —
chatwoot.com(Captain AI 0/300/500/800 credits by tier);noloco.io(30/250/custom credits per month);thunkable.com(2K/20K/50K/100K AI tokens per tier). - Included everywhere —
mixpanel.com("All plans include Mixpanel AI");mezmo.com("No AI surcharges");querypie.com(AI Extensions checked in Free and paid rows). - Flat add-ons —
readme.com(Ask AI $150/mo);zoko.io(AI agents $24.99/mo + $0.09/resolution);salarybox.in(AI Addons block). - Reserved higher —
epsilon3.io(AI Procedure Import in Pro $200, absent Basic $100);freshpaint.io(full AI from Growth up);zeroheight.com(full AI + workflows gated above Free).
AI presence by billing model
Matched against our pricing-transparency study of the same companies: which billing models mention AI, and which ones put a price on it.
| S1 billing model | Mention AI | Priced AI |
|---|---|---|
| Usage-based (n=41) | 87.8% | 53.7% |
| Per-seat (n=44) | 84.1% | 45.5% |
| Flat / tiered (n=85) | 67.1% | 36.5% |
| Contact-only (n=90) | 71.1% | 18.9% |
| Page, no numbers (n=23) | 30.4% | 4.3% |
The free AI ration
Metered AI starts with a free allowance and expands by tier. Stated allocations from the evidence:
| Vendor | What is metered | Stated allowances |
|---|---|---|
chatwoot.com | Captain AI | 300 credits, 500 credits, 800 credits |
retool.com | pooled AI credits | 250 credits, 1,000 credits |
archbee.com | AI add-on tokens | 20M tokens, 40M tokens, 100M tokens, 1M tokens |
thunkable.com | AI tokens | metered, amounts in page |
trymito.io | AI completions | metered, amounts in page |
weweb.io | AI tokens | metered, amounts in page |
noloco.io | AI credits | 250 credits |
taskade.com | Taskade AI credits | 50,000 credits |
The pattern rations intelligence the way APIs ration calls: a free tier that proves value (250–500 credits, 150 completions, 2,000 tokens, 1M tokens), a paid tier roughly 4–100x larger, and an "unlimited / custom" top. Founders setting allowances: the cohort's free rations cluster low enough to trial and high enough to habituate — copy the shape, tune the numbers to your marginal cost.
Stated unit prices: a reference table
Every per-unit AI price actually printed on a pricing page in this sample. None of these has a base tier to divide by — they are rates without a premium, published here so buyers can compare vendors unit-for-unit.
| Vendor | Unit | Stated rate | Note |
|---|---|---|---|
evidence.dev | Credit overage | $0.01 / credit | stated overage, no base tier to compare |
archbee.com | Token overage | $1.2 / 1M tokens | stated overage, no base tier to compare |
hyperbrowser.ai | AI Extract output | $30 / million output tokens | metered extraction |
plivo.com | Voice AI agent minute | $0.03 / min (excl. telephony) | models + transcription + orchestration |
zoko.io | AI agent resolution | $0.09 / resolution + $24.99/mo | flat + per-outcome hybrid |
candidately.com | AI resume export | $1 / exported resume | per-outcome unit |
nyckel.com | AI invoke | $0.005 / invoke after included use | metered inference |
nanonets.com | Credit packs | $100 / mo per 100 credits | pack pricing on usage product |
When the product is AI
15 of 91 priced offers (16.5%) have no separable AI premium because the product itself is AI. They still split two ways:
| How AI-natives charge | Vendors | Count |
|---|---|---|
| Per-seat / tiered AI products | aitechpacks.com, hudson-labs.com, plai.io, finlens.app, getdot.ai, tavus.io (PALs), tiriel.ai | 7 |
| Pure usage-metered AI APIs | exa.ai, nyckel.com, openlayer.com, sapling.ai, promptloop.com, contraforce.com (per-incident) | 6 |
| Mixed / unstated on page | inscribe.ai, mindee.com | 2 |
The lesson for AI-native founders: even without a legacy product to attach to, two-thirds still reach for seats or tiers first and meter second. Pure meter-only pricing is the minority even among AI companies.
The hype gap: AI mentioned, nothing priced
15 of 106 coded pages mention AI with no priced offer attached. Three recurring shapes:
| Fluff shape | Examples | Count |
|---|---|---|
| Nav-menu AI (product link, no offer) | cosmicjs.com, formance.com, rootly.com, draftwise.com (news), getshogun.com (integration note) | 5 |
| Hero/marketing claim, no priced offer | bik.ai, tesorio.com, zuddl.com, arist.co, superb-ai.com, useagave.com, nestorup.com, postscript.io | 8 |
| Sales-led page, AI unpriced | ironcladapp.com, lattice.com | 2 |
Plus 96 more pricing pages mention AI only in passing with no pricing signal at all. The hype gap cuts both ways: buyers should treat unpriced AI claims as roadmap language, and vendors with genuinely included AI should say so on the page — bundled AI you do not state gets you no credit.
Small companies meter; nobody starts with add-ons
| Team size | Metered | Bundled | Add-on |
|---|---|---|---|
| 1-10 employees (n=24) | 11 | 5 | 0 |
| 11-50 employees (n=35) | 11 | 15 | 5 |
| 51-200 employees (n=26) | 8 | 8 | 5 |
| 200+ employees (n=5) | 2 | 2 | 1 |
Who this is for
Price the AI like the cohort does
Metered credits are the default buyers already understand — 35.2% of priced AI offers work this way. A flat add-on is the exception, and a per-seat AI surcharge appears nowhere in 91 offers. If you charge differently, say why on the page; buyers will compare you against credits.
Ask which bucket the AI sits in
"Included", "metered", "add-on" and "higher tiers only" have very different expansion dynamics. A vendor that will not say which one its AI falls into is telling you the answer is metered.
Three premiums exist — study them
Readme (60%), Zoko (31%) and Unlayer (20%) are the only vendors that state an AI price against a priced base tier. Whatever your model, these three pages are the closest thing to a public benchmark.
Do not compute a market premium from public pages
Three unambiguous cases cannot generalize. Anyone publishing a market-wide "AI premium" from pricing pages is imputing, not measuring — cite this study for why.
Methodology, deviations & reproducibility
What we read: 286 live pricing pages, captured October 2026. 202 mentioned AI somewhere on the page; 106 mentioned it in a pricing context — a tier card, a comparison row, an add-on block, or a credit meter attached to a price or plan name. Each of those 106 was read and classified by hand. Mentions in nav menus, testimonials, blog teasers and job ads were counted as marketing, not pricing.
What changed from the plan: we wrote the classification rules before reading a single page and logged three deviations. (1) We did not cross-check changelogs or blogs, so no claim is made about when any AI feature launched. (2) No before/after history — archived pages mislead too often, so this is a snapshot of October 2026 only. (3) Two extra buckets were added during coding and counted openly: products that are AI (no separable AI price exists) and marketing-only mentions (never folded into paid-model shares).
Downloadable datasets: ai-pricing-taxonomy.csv (taxonomy counts) and companies-ai-pricing.csv (per-company codes with evidence notes).
python3 scripts/research/s7_capture.py # source HTML (S7 capture)
python3 scripts/research/build_s2_page.py # rebuild this page from codes
Disclosure and corrections
DataDab sells AI-visibility work and paid inclusion in its research pages. A research mention costs $400–$1,200. Payment does not change a figure in this study. DataDab is not part of the measured set. Corrections are published on the page; send factual corrections with the URL to amit@datadab.com.
Covering this research? Please link the primary source so readers can verify the data and methodology:
Markdown: [The B2B SaaS AI Pricing Premium Index 2026 by DataDab](https://www.datadab.com/research/b2b-saas-ai-pricing/)
Disclosure: DataDab sells AI-visibility work and paid inclusion in its research pages. A research mention costs $400-$1,200. Payment does not change a figure in this study. DataDab is not part of the measured set. Corrections are published on the page (amit@datadab.com).
What this study does not show
- Launch dates. The January-2023 launch criterion was not verified per product; no claim is made about when any AI feature launched.
- Disallowed and failed pages. 114 robots-excluded domains plus 79 failed S7 captures are unmeasured.
- A market-wide premium. Three computable cases cannot generalize; credit overages without base comparisons are never converted.
- Single reviewer. Hand-coded with quoted evidence; no independent second check yet.