AI tools for CPA firms, mapped by what they actually do
Nilanjan Raychaudhuri, founder, AgentLink (Tublian LLC) · Published September 3, 2026 · Last updated September 3, 2026
AI tools for accounting firms now span several fundamentally different jobs. Some replace the ledger. Some automate the firm’s own production. Others provide bookkeeping, close management or forecasting. Comparing them in one ranked list obscures those differences, so we mapped them by what each product actually produces.
One disclosure up front. We build in this space ourselves. The map below is meant to be accurate whether or not you ever look at what we make, and we have tried to describe every company the way its own customers would.
How we built this map. We reviewed product pages, documentation, pricing pages, security pages and company announcements available on September 3, 2026. Placement reflects the primary outcome each product sells, not every feature it offers. Some products span more than one lane. Where a comparison comes from a vendor’s own material, we say so. Where a number could not be confirmed from a primary source, we left it out.
Five lanes
The products we reviewed fell into five primary lanes. The lanes are sorted by what the tool produces, because that is the question a firm is really asking: what does this thing hand me at the end?
| Lane | What the tool produces | Who pays | Examples |
|---|---|---|---|
| A | A new accounting system for the client | The client | Rillet, Campfire, Digits, Puzzle |
| B | The firm’s own production work | The firm | Basis, Accrual, Fieldguide, Truewind |
| C | A closed, checked set of books | The client’s controller | Numeric, Nominal, FloQast |
| D | Bookkeeping as a service | The client | Pilot, Zeni, Docyt, Bench |
| E | A forecast for the client | The firm, on the client’s behalf | Clockwork, Jirav, Float, Fathom |
In our reading, a firm whose clients are in the one to ten million revenue range with no finance staff of their own will spend most of its evaluation time in Lanes B, D and E. Lanes A and C are mostly bought by companies with a controller in house, though larger client accounting or fractional CFO engagements do reach into both.
Lane A. A new accounting system for the client
Definition: products that replace the client’s general ledger rather than sit on top of it.
Rillet (rillet.com) and Campfire (campfire.ai) are AI-native ERP and accounting systems that compete with established platforms such as NetSuite. Rillet’s stated position is that AI sitting on top of a ledger works from stale data, and that operating inside the ledger is what lets the AI act (Rillet on AI-native ERP). Campfire positions itself more broadly as a full accounting and finance platform (campfire.ai). The customers both describe are companies that have outgrown small-business accounting software, typically with a finance team in place.
Digits (digits.com) combines an AI general ledger with bookkeeping, so it sits across Lanes A and D. It sells to businesses directly and also to accounting firms.
Puzzle (puzzle.io) is an accounting system aimed at startups, with a partner program for firms.
Lane B. The firm’s own production work
Definition: products sold to the firm about the firm’s own work. The client is not the customer and usually never sees the tool.
Basis (getbasis.ai) builds AI agents for audit and tax work inside firms. It announced a 100 million dollar Series B at a 1.15 billion dollar valuation in February 2026 (announcement).
Fieldguide (fieldguide.com) covers audit and advisory engagement workflow. Accrual (accrual.com) and Truewind (truewind.ai) apply AI to firm-side accounting production.
Lane C. A closed, checked set of books
Definition: products that manage the month-end close and check that what was recorded is right. The work is retrospective by design.
Numeric (numeric.io) is close management, reconciliation and flux analysis for in-house accounting teams, with AI drafting variance explanations from GL data (close management, variance analysis).
Nominal (nominal.so) deploys AI agents for reconciliation, consolidation, variance analysis and close across any ERP, without replacing it (platform).
FloQast (floqast.com) is the longest-established of the three, a close management platform built around close checklists and reconciliation, now adding AI features.
All three primarily serve in-house accounting teams managing close and reconciliation work.
Lane D. Bookkeeping as a service
Definition: companies that do the bookkeeping, with their own software underneath.
Pilot (pilot.com), Zeni (zeni.ai), Docyt (docyt.com) and Bench (bench.co) sell bookkeeping as a service. Pilot (partner page) and Puzzle (firm page) also publish partner programs for accounting firms.
Lane E. A forecast for the client
Definition: products a firm uses to produce forward-looking numbers for a client, usually as part of an advisory engagement.
This is, in our assessment, the most competitive lane, and the one a firm moving into advisory work is most likely to be evaluating.
Clockwork (clockwork.ai) is AI cash forecasting connected to QuickBooks Online and Xero, sold to accounting firms. Published pricing starts at 149 dollars a month, billed per company, with firm plans pooling client licenses across tiers (pricing). It states SOC 2 Type II compliance, and an AI analyst called Mira answers questions about the numbers (Mira page). Clockwork publishes its own comparison pages against Jirav, LiveFlow, Mosaic, Fathom and Reach Reporting; these are vendor-authored. Note that clockwork.io is an unrelated company.
Jirav (jirav.com) is driver-based financial planning and budgeting built for accounting firms, with three-statement forecasting and integrations that extend to NetSuite and Sage Intacct (integrations). Its partner program lets a firm templatize a model and clone it across clients (partner program). Jirav’s own comparison (Jirav vs Clockwork, vendor-authored) positions Jirav for bespoke modelling and board or bank reporting, and describes Clockwork as the better fit for ongoing monitoring with exception-based alerts. That distinction is useful, and it comes from an incumbent.
Float (floatapp.com) is a 13-week rolling cash forecast that attaches to Xero and QuickBooks. Float reads bank activity through the accounting software rather than connecting to banks directly, so its view of cash is as current as the bookkeeping.
Fathom (fathomhq.com) is an Australian reporting and forecasting platform sold to accountants, now part of Access Group (announcement). It offers white-labelled reporting under the firm’s own brand (help centre), so a firm evaluating this lane should treat white-label as standard rather than distinctive.
Finoya (finoya.ai) markets itself as an AI CFO for accounting firms, with a free tier for business owners and an advisor plan that is free for the first five clients, then 39 dollars per client per month with volume discounts. White label is a paid add-on (pricing, accountants page). Its published material covers cash health, receivables aging, and how to judge whether a forecast can be trusted.
Compass AI (compassapp.ai) offers a multi-client view for CPA firms and fractional CFOs and publishes guidance on what firms can bill for advisory using it. Finlens (finlens.app) targets firms with fifty or more clients.
Also in this lane: Dryrun, Agicap, Cash Flow Frog, Pulse and Futrli at the small-business end. Cube, Datarails, Abacum, Mosaic and Pigment at the enterprise end, priced for companies with a finance team.
What this framework may still miss
Three things are worth separating here.
An observed market fact. Every product above is sold around a defined deliverable. A ledger. A workpaper. A closed set of books. A forecast. The firm buys the tool, logs in, gets the thing.
Our interpretation. There is a job in a finance function that does not produce a deliverable in that sense. It is the job of paying attention between reporting periods. Noticing on the eighteenth that a payment expected on the fifteenth has not arrived. Carrying what was decided on last month’s call into this month’s so the conversation does not start over. Knowing that a customer who is late in July is late every July. In a company with a finance team, a controller does this. In the clients most firms serve, nobody does, and the firm typically looks at the numbers once a month. We think that is a separate job from any of the five lanes, because it is defined by continuity rather than by output.
Our product. AgentLink (agentlink.finance) is built around that interpretation. Rather than automating a task, it takes on the responsibilities a controller would hold if the client could staff one: keeping continuity across the engagement so each conversation builds on the last, answering the client’s finance questions as they come up rather than at the next meeting, and owning a forecast that is held against what actually happens. It is sold through accounting and fractional CFO firms, who keep the client relationship and the judgement calls.
Common questions
What AI tools are available for CPA firms?
They fall into five groups: AI-native accounting systems (Rillet, Campfire, Digits, Puzzle), firm production tools (Basis, Accrual, Fieldguide, Truewind), close management (Numeric, Nominal, FloQast), bookkeeping services (Pilot, Zeni, Docyt, Bench) and forecasting platforms (Clockwork, Jirav, Float, Fathom, Finoya, Compass AI).
Which AI tools are designed specifically for fractional CFOs?
Within this map, the products most directly aimed at fractional CFO workflows sit in Lane E. Clockwork, Jirav, Fathom, Finoya and Compass AI each publish firm-facing plans with per-client or pooled pricing.
What is the difference between AI bookkeeping and AI forecasting?
Bookkeeping tools (Lane D) record and categorise what already happened. Forecasting tools (Lane E) project what will happen, usually reading from the books the bookkeeping produced. A firm often uses one of each.
Which tools work across multiple client accounts?
Clockwork, Jirav, Fathom, Finoya, Compass AI and Finlens publish firm or multi-client capabilities. Float and most Lane D tools are priced per company.
How should a CPA firm evaluate an AI finance tool?
Start with the lane. Decide what deliverable you need before comparing features, since features across lanes are not comparable. Then check three things the vendor should state plainly: which accounting systems it reads from, whether it connects to bank data directly or through the books, and how it handles data security.
About the author
Nilanjan Raychaudhuri is the founder of AgentLink (Tublian LLC, Columbus, Ohio), which builds controller-layer software sold through CPA and fractional CFO firms. He has spent the past year interviewing practitioners at fractional CPA and CFO practices about how advisory work is actually delivered. Team page
Corrections: if a company on this map has changed materially since September 3, 2026, or we have described something wrongly, email nilanjan@agentlink.finance and we will fix it and log the change.
Update log
- 2026-09-03: First published.