06-reference/research

ai bookkeeping pricing channel split

2026-10-31·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)

The channel split is clean and the per-client price gap is roughly 20x: business-facing AI bookkeepers charge 11-67% of a Client Advisory Services practice's revenue per client, firm-facing tools charge 0.4-3.4%, and not one published firm-facing price is per staff seat

The question

"What do Pilot, Receipts AI, Karbon and the newer AI-bookkeeping entrants actually charge per client per month, and do they sell to the end business or into the CPA firm?"

Derivative of [[2026-09-01-cpa-bookkeeper-aging-cohort-tooling-spend]], which found that the displaceable prize in a small accounting firm is labor (28.7-34.4% of net client fees) and not tooling (4-8% at best, and that figure traced to a vendor blog rather than to the AICPA). The channel an entrant sells through determines which of those two lines it is competing for, and the displacement map is different in each case.

What we already know (from the vault)

What the web says

All vendor prices below were read off the vendor's own live page on 2026-10-31 and are VERIFIED-PRIMARY unless tagged otherwise. Per-client percentages are Ray arithmetic against the $1,489/month CAS revenue-per-client benchmark and are ESTIMATE, cross-survey.

Convergences and contradictions

Synthesis for RDCO

The channel question resolves cleanly, per vendor, and it resolves in favor of the parent brief's reframe. Selling to the end business (B2B-DIRECT): Pilot's pilot.com product, Fondo, Decimal, Numeric (to internal finance teams rather than small businesses), Inkle, and Receipt AI (receipt-ai.com). Selling into the CPA or bookkeeping firm (B2B2B-FIRM): Karbon, Double, Uncat, Black Ore, Fieldguide, Basis, and Pilot's Meridian. Both, through distinct products or tiers: ReceiptsAI, Puzzle (software direct plus an accounting-partners program where Puzzle bills the client and the firm pays nothing), Zeni, Bookkeeper360, Bookkeep, Truewind. The business-facing group is competing for the labor line and competing for it by taking the client outright, not by selling into the firm. The firm-facing group is competing on the tooling line on price while pitching the labor line in its copy. Notably, Pilot built a separate white-label brand rather than adding a partner tier to pilot.com, which is at least consistent with the two channels being unable to share a price list.

For the acquisition lane, the number that matters is not the tooling stack, it is the $299-$799. The Palm Beach CPA practice carries $150,000/year of monthly accounting across 20-30 clients ([[2026-08-30-shortlist-refresh-scored]]), which is roughly $420-$625 per client per month. Pilot Core at its lower expense bands, Fondo at $399, and the reborn Bench at $399 land inside that band as human-backed services, and Pilot Essentials at $99 lands at a quarter to a sixth of it for AI-only books. That is the competitive fact that belongs in diligence on any monthly-accounting book: the price floor for the service is now publicly posted by funded competitors, it sits at or below what the practice charges, and the practice's defense is relationship and tax adjacency rather than price. This strengthens the parent brief's conclusion that tax-heavy books are the safer acquisition while CAS-heavy books are the better margin story - the CAS-shaped revenue is exactly the revenue the business-facing entrants are priced to take.

The two deaths are the most useful finding for RDCO's own positioning. Bench realized roughly $4,083 per customer per year, about $340 per month (ESTIMATE: $49M revenue against the ~12,000 customers Employer.com acknowledged, which mixes a FY-2024 revenue figure with a Dec-2024 customer count), with human bookkeepers in the loop, and burned $135M without reaching profitability. It raised its combined bookkeeping-and-tax price 50% in its final quarter; that is suggestive about where management thought the problem was, but no Bench or Employer.com statement explains the failure, so motive is inference. Finally failed in the same shape roughly twenty months later. Meanwhile a CAS practice earns $1,489 per client per month, and MAP firms overall post a median $252,663 of net remaining per partner (all-firm median, not CAS-specific). The ~4.4x gap between Bench's realized revenue per client and a CAS practice's is the clearest quantity in this brief, and it is a pricing gap, not a demonstrated moat - no firm-failure data was gathered here, so "incumbents survive" is not established. What the evidence does support is narrower and still useful: two well-capitalized attempts to serve this segment at a third of the incumbent price died, which means the segment punishes low revenue per client and acquisition cost at least as hard as it rewards production efficiency. The firm-facing channel, where one sale reaches a firm that already owns 67 relationships, is the structurally better side of that trade even though its per-client price is roughly 20x lower.

Where this changes RDCO's own framing: per-client metering is the pricing model to copy. Double, Uncat, ReceiptsAI and Black Ore all independently landed on metering per unit of client work in a category whose incumbent is per-seat. The inference - and it is an inference, since no vendor states it - is that per-seat pricing caps vendor revenue exactly when the product succeeds: if the agent lets one accountant carry a $1M book, per-seat revenue falls while value delivered rises, and per-client metering captures the opposite. For any RDCO agent product sold into a services firm, that is the directly transferable lesson. It is worth noting the contrast with the flat $12K-plus-$500/month shape in [[2026-05-24-mostlymetrics-pizza-shop-ai]], which is a single-business flat fee rather than per-client metering, and therefore a counterexample rather than corroboration.

Why this is in the vault

It closes the parent brief's own open follow-up with verified list prices, and it supplies two things the acquisition funnel does not currently have: a published competitive price floor ($99 AI-only, $299-$799 human-backed per client per month) to test any monthly-accounting book against during diligence on the bookkeeping/CPA lane in [[2026-10-01-shortlist-refresh-scored]], and a bounded AI-tooling figure (2.2% of net client fees at full attach, from published prices) that keeps axis 1 of the 5-point fit test pointed at labor rather than software. It also retires the Bench uncertainty the parent carried as "recalled, not verified," and adds Finally as a second data point on the same failure shape, which bears directly on whether RDCO should ever price an agent-delivered service book on a flat monthly fee.

Open follow-ups

Related

Sources

Vault:

Vendor pricing pages (VERIFIED-PRIMARY, retrieved 2026-10-31):

Bench and Finally:

Paywalled, flagged and not retried: The Information, American Banker, Globe and Mail. GeekWire returned 403.