"No way." That was the answer from the owner of an Estonian accounting firm — 11 staff, around 200 SMB clients on Merit — when I asked whether her sub-€1M clients would pay for an AI interface over their books. Not "interesting, but later." No way.
Then she added the part that matters: "Above €1M, it's a different conversation."
I heard a version of that line from three independent Estonian firms over six weeks. It points to something the category's TAM slides tend to flatten: a discontinuity, somewhere around €1M (~$1.1M) in annual revenue, where AI in finance shifts from hard-to-justify to potentially worth paying for. This piece looks at the evidence on both sides of that line, and why it's easy to misprice.
Disclosure: I'm building Artifi, an AI tooling layer for finance teams, so I have a stake in this market. The figures below are presented as I found them.
"SMB" is not one market
"SMB" covers a sole proprietor doing €80K of consulting and a 150-person manufacturer doing €40M. They aren't the same buyer. The incumbents' own numbers show where the economics sit. Xero reports 4.4M subscribers with ARPU around $26.50/month and a deliberate push up the size curve, because the micro-business band is saturated; its most common customer profile is 10–50 employees and $1M–$10M in revenue, per one market analysis (Porter's Five Forces analysis). QuickBooks reports 7M+ users with SMBs at roughly 62% of the base (DataCaptive), and Intuit's growth narrative centers on mid-market expansion, where ARPU rises (Intuit IR).
The AI-finance cohort sits even higher. Vic.ai positions itself around 1,000+ invoices/month on NetSuite, Sage Intacct, or Dynamics, and points smaller buyers to BILL or Ramp (accounting AI tools). Brex exited the SMB segment in 2022 to sell enterprise (Wikipedia). Pilot's entry tier is $499/month, rising to $1,500+ for a real company (Pilot). Ramp, the most SMB-shaped of the group, reports ~2,200 of its ~70,000 customers contributing $100K+ ARR each — a small share of accounts producing much of the revenue (Ramp). In practice, most of these companies make their money above the line, whatever the TAM slide says.
Below €1M: the math is hard
Under the line, the owner is usually the bookkeeper. By various estimates a large majority of small-business owners handle their own books, and many operate without any accountant at all; among one-employee businesses, only around 58% use financial software (NewityMarket, Compass). Standard guidance is to hold off on a bookkeeper until revenue is consistent — typically $100K–$150K, with a practical ceiling around $500K–$750K or 100–150 monthly transactions before you have to bring someone in (Steph's Books).
Run the ROI at that scale and it rarely closes. The "AP team" is one person who is also the CEO; invoice volume is 20–40 a month; learning a tool, wiring it to the bank, and reaching steady state takes weeks. Bill.com Essentials is $49/user/month plus per-transaction fees (Bill.com); Pilot starts at $499/month. Even "free" tools like Ramp earn their margin on spend volume a sub-€1M business doesn't generate. The automation savings are often smaller than the cognitive cost of adopting the tool — which is what "no way" tends to mean. It is a statement about the math, not about AI.
Above €1M: where it starts to pay
Cross €1M and three things tend to happen together. A dedicated finance person appears — part-time bookkeeper at €1M, full-time by €2M, a controller by €5M — so the time being saved finally has a name and a salary. Transaction volume crosses the 100–150/month range where automation pays off in hours per week rather than minutes per month (Steph's Books). And the business gets categorically more complex: payroll, multi-entity, sales tax across jurisdictions, FX, project accounting, inventory — each a new failure mode.
The same outreach behind my notes on 100+ CFO conversations showed the pattern: engagement among 50–200-employee European companies ran around 13%, and fell off sharply below 50 employees — not from hostility, but because there is often no finance lead to engage. A ~€10M Estonian transport group on Business Central was evaluating layer tooling in the €500–1,000/month range — the band where, with three finance staff, the math begins to close. A fintech running 100,000+ Stripe transactions a month, far above the line, had explicit demand for tax-code automation, because at that volume a misclassification compounds.
The threshold is best read as a start, not a niche. €1M is where a finance function first exists; above it, each increment of complexity adds a workflow, and a single-task tool tends to widen into a function-wide need. The demand doesn't cap at the line — it begins there and compounds upward.
Why the line gets mispriced
If the line is this clear, why do vendors still pitch a 30M-business TAM to sub-€1M owners? Largely because of how rounds are raised: "AI for the 33M US small businesses" fundraises better than "AI for the ~3M businesses above €1M with a real finance function." The predictable result is a cycle — launch broad, struggle to convert sub-€1M owners at sane CAC, quietly retarget mid-market, rewrite the site. Brex did it explicitly in 2022 (Wikipedia); Ramp's ICP language is drifting upmarket as a small share of customers produces most ARR (Ramp); Vic.ai stated its 1,000-invoice floor outright. A secondary factor: sub-€1M client work is low-margin for accounting firms too, so the channel has little incentive to push adoption there. The economics of the channel mirror the economics of the customer.
What this means for each group
For vendors: the data suggests two distinct buyers wearing one label. Below the line is a single human (often the founder) for whom install time and cognitive cost dominate; above it is a finance function for which integration, security review, and per-seat ROI dominate. The pricing and onboarding that suit one tend not to suit the other.
For SMB owners under €1M: much of the "AI is transforming finance" framing describes the transformation of finance functions — which a sub-€1M business usually doesn't have yet. The historically effective stack at that size has been accounting software plus a payments processor plus a periodic accountant; a single agent for a genuinely disliked task (receipt capture) is the plausible exception.
For investors: the realistically addressable market is the share of SMBs above the revenue line where a finance function exists — a small fraction of the headline business count, but the large majority of the addressable spend. A flat SMB TAM tends to overstate the reachable demand.
The takeaway
The €1M line is less a marketing segment than the point where the buyer for AI in finance begins to exist — a person whose time can be saved, in a business with enough volume to save it. Below the line, the pitches mostly outrun the math. Above it, the buyer is real, the math works, and — because complexity keeps compounding with size — the entry point tends to widen rather than cap. The useful exercise is drawing that line on the whiteboard before drawing it on the TAM slide.