What automated client onboarding looks like at a CPA firm

What the manual version actually costs

Most small firms still run onboarding by hand. Layer3labs' 2026 guide to accounting client onboarding automation cites the AICPA's 2025 Practice Management Survey: roughly 65% of firms with fewer than 20 staff still rely on email-based onboarding with no automation at all. Layer3labs' guide also estimates a typical manual intake runs 12 to 18 separate steps across 4 to 6 different systems, with every handoff a place the file can stall for days.

The hours behind that add up. Layer3labs estimates firms with 2 to 20 staff lose 2 to 4 hours of staff time per new client to tasks a system could otherwise handle. Using its own example, Layer3labs notes a firm bringing on 150 new clients a year at 4.2 hours each spends roughly 630 staff hours annually just on getting people in the door, before any billable work starts. Layer3labs also reports that small firms which automate onboarding, document collection and billing commonly see 20 to 35% fewer admin hours per engagement, though the range depends on how much of the process actually gets automated. Most of that time sits inside the document chasing that eats every busy season, not the judgment calls only a CPA can make.

There is a client-facing cost too. TaxDome, which sells onboarding and practice management software itself, reports survey figures showing 71% of accounting clients value efficient onboarding, and one in five say they would pay up to 50% more for faster service. A vendor citing numbers that favor its own product is not neutral evidence, but the underlying point holds up: a client waiting on a checklist notices, and starts to wonder if their return will get the same treatment.

How the workflow actually runs

Picture the loop instead of a portal. A client signs the engagement letter in Ignition. That signature is the trigger: a system reads it, creates the client record and the first job in Karbon or Jetpack Workflow, and sends a document request built from the return type rather than a generic list. Nobody has to remember to start any of that by hand.

Then it waits, but not passively. If a document has not arrived in three days, it sends a reminder. If it still has not arrived after the second reminder, it escalates to a staff member instead of firing off a third automated email the client will ignore anyway. Deciding when a machine should stop nudging and hand the problem to a person is the part most client portals skip. The same pattern shows up in automated certificate of insurance requests: reminders work for a while, then a phone call works better.

Once the documents are in, a KYC or AML check runs against them, the same kind of identity and risk check compliance already requires before work can start. If the check clears, the first job opens on its own, and nobody has to remember to kick it off. If it flags anything, the workflow stops and waits for a person, every time. A compliance flag is exactly the kind of decision a system should never make by itself, because a wrong autonomous call there costs far more than a late document ever would.

Where this sits next to Karbon, Canopy, Ignition and Jetpack Workflow

None of these tools are the wrong choice. Karbon and Jetpack Workflow template the tasks and deadlines inside a firm. Canopy and Ignition handle the client side: e-signature and billing. Ignition's Solo plan starts at $39 a month and its Pro plan at $229, according to Layer3labs' guide, though pricing is worth confirming directly since plans change.

What most of these platforms do not do on their own is watch the space between them: notice that a signed engagement letter in Ignition has not produced a job in Karbon yet, or that a client's portal upload does not match the file type the checklist asked for. That watching-and-deciding layer sits on top of a firm's existing stack. It does not replace it. How workflow automation differs from a system that can decide and escalate on its own is worth reading before adding either kind of tool.

We build and run these loops the way we run our own operations: as scheduled jobs that check their own results and keep going without anyone watching. Right now that is 197 scheduled jobs across 78 live integrations, running continuously.

What still needs a person

Some parts of onboarding should stay manual past a certain point. Engagement letter terms that fall outside the standard template, or a KYC flag on a new client, need a partner's judgment, not another retry. A system that guesses on either one creates more cleanup than it saves.

The honest version of any of this starts with mapping the process first: which steps are genuinely repetitive and rule-based, and which ones need someone experienced to make a judgment call. How to evaluate an AI automation vendor covers the questions worth asking before handing any of this over, onboarding included.

A number worth checking yourself

Pull your last ten new-client files and count the emails: from a signed engagement letter to the first return actually starting work. Some firms will find three steps. Others will find something closer to Layer3labs' estimate of 12 to 18. Either way, that number, not an industry estimate, is the one that tells you where automating this would actually help this season.