The starting point
A mid-sized CPA firm — roughly 40 people across tax, audit, bookkeeping, and administration — asked Node8 a direct question: where would AI actually save us time? Not “should we buy licenses,” but which of our workflows are worth automating, in what order, and what does that require?
That is the right question, and answering it took a structured discovery: a working session walking through the firm’s real workflows on real screens — tax software, spreadsheets, the document management system, CCH Engagement on the audit side. This page is the hub for what came out of it. The discovery process itself is documented in What an AI Discovery Looks Like for a CPA Firm, and Node8’s broader accounting practice is at /acct.
What discovery surfaced
CPA firms are document factories. Almost every finding fell into one of four buckets.
Tax document data entry
The single largest time sink was manual data entry from source documents into tax software. The firm processes on the order of 5,000 K-1s and 1,000 W-2s a year. W-2 import mostly works; 1099 import is inconsistent because of formatting variation; K-1s are entered almost entirely by hand because correct box-to-field mapping takes real expertise, and a single return can carry more than twenty K-1s. Conservatively, K-1 entry alone consumes over 15,000 staff-minutes a year, spread across roughly fifteen people.
This is a near-ideal AI target: high volume, structured-but-messy documents, and mapping rules that experienced staff can articulate. The automation shape is document extraction plus a mapping layer that encodes the firm’s expertise, feeding the tax software through its import formats rather than through human retyping.
Trial balance and financial statement preparation
The second bucket was spreadsheet work. Each year, ending balances get copied into next year’s beginning balances across hundreds of entities — up to an hour per client. Client QuickBooks exports get pasted into firm templates, and inconsistent client account names get manually remapped to the firm’s standard chart of accounts every single time. On the audit side, CCH Engagement automates the roll-forward and linking, but adding accounts, copying work papers between clients, and fixing mappings is still manual.
Here AI helps in two ways: scripted roll-forward of the mechanical copying, and AI-assisted account mapping that learns the firm’s normalization rules instead of forcing a person to re-derive them per client.
Document management and client communication
Finalized returns are saved into the document management system by hand — a few minutes per return, multiplied by hundreds of returns. Client documents arrive by email attachment and shared links because the portal was never fully rolled out, and clients regularly upload the wrong document into the wrong slot. The fixes are unglamorous but compounding: trigger-based filing when a return is marked final, and AI-assisted validation that catches a credit card statement uploaded as a bank reconciliation before it wastes a preparer’s time.
Drafting and routine responses
Audit and compliance work generates recurring, near-templated writing: standard client queries, compliance checklist responses, document request lists. AI drafts these well when it is grounded in the engagement’s actual context — with a human reviewing before anything goes out.
Where AI does not belong
Discovery is as much about drawing the line as finding opportunities. Three boundaries came through clearly:
- Sign-off stays human. Nothing AI produces is a substitute for preparer and reviewer judgment, and nothing gets filed or issued without it. AI compresses preparation; CPAs attest.
- Confidentiality precedes convenience. Staff at the firm were already using ChatGPT and Claude individually, without a firm policy. Step one of any rollout is replacing that with a firm-managed subscription, data-handling terms the firm has actually reviewed, and explicit rules about client data.
- Some ideas fail the trust test. Recording client phone calls for AI transcription, for example, was set aside over legitimate privacy and client-relationship concerns. A discovery that never says “not this” is not doing its job.
The prioritization logic
Every candidate got the same rough math: annual volume, times minutes per occurrence, times the number of people doing it — against implementation difficulty. That ranks K-1 extraction and trial balance roll-forward at the top, document filing automation as a fast follow, and portal and communication improvements as a second phase. It also gives the firm’s partners something a tools pitch never does: a defensible, numbers-based reason for what happens first.
What happens next
The engagement is at the discovery-to-recommendation stage: Node8 is researching the specific automation options against the firm’s actual software stack and will return a prioritized proposal — quick wins first, a pilot scoped to one workflow with measurable time savings, and a coordinated AI-tooling rollout for around forty staff instead of the current informal, everyone-for-themselves usage.
For the discovery method itself — what we ask, what we measure, and how a pilot gets chosen — see What an AI Discovery Looks Like for a CPA Firm. For Node8’s work with accounting and finance teams more broadly, see /acct.
Work with Node8
If your firm is asking where AI would actually save time — rather than which licenses to buy — a structured discovery is the honest place to start. Get in touch and we’ll walk through your workflows the same way.