CPA Engagement Letter Automation: Call to Signed in Minutes

The advisor call ends. Now someone has to remember exactly what was discussed, work out what to charge, draft a scope of work, turn it into an engagement letter, and get it in front of the client before their interest cools off. At most CPA firms, that's not a 15-minute job. It's a chain of drafting, review, and back-and-forth that can stretch over several days — and the client relationship starts on a wait, not a win.

We built the system that closes that gap for a mid-market US CPA firm running Firm360 and UltraTax CS. It starts at the very first discovery call and ends with a signed engagement letter roughly 10–15 minutes after the advisor call is over — fully scoped, fully priced, without the advisor drafting a word of it.

Where the days actually go before a client signs anything

Most firms already have a system for what happens after the engagement letter is signed — document checklists, portals, onboarding software. We've written before about what a 3-day onboarding process looks like once that letter exists. What's usually missing is everything before it.

The typical path looks like this: a prospect has a discovery call with someone on the team, not the advisor who'll actually run the engagement. Notes go into a CRM field or, just as often, into someone's memory. Before the real advisor call happens, that person has to re-read or re-explain everything the prospect already said — the same discovery, run twice. During the call, the advisor is improvising pricing and scope in real time, or promising to "follow up" once they've had time to think about it. Afterward, a scope document and letter get drafted from a template, routed for review, and sent whenever someone has a free hour between client work.

Each of those handoffs is a place the thread can drop. None of it is a skills problem — the people involved know exactly what to do. It's a plumbing problem: the information from the first conversation isn't structured, searchable, or connected to whatever happens next. A prospect who had a great first call and then hears nothing for four days doesn't conclude the firm is busy. They conclude the firm is disorganized, and they keep the other quote they got.

The timing makes it worse, not better. New-client discovery calls don't pause for busy season — a referral comes in during the second week of March the same as it does in July, and that's exactly when the advisor has the least spare time to draft a custom scope document by hand. A gap that's a minor annoyance in the off-season becomes the reason a good prospect never becomes a client during the months a firm most needs new revenue in the pipeline.

What actually runs between the discovery call and the signed letter

Here's the sequence, in order:

• The discovery call is transcribed automatically. No one is taking notes by hand or trying to remember details three days later.
• The transcript builds a structured client profile — business name, entity type, reporting period, and the specifics of their situation — the moment the call ends.
• The system checks what's still missing against that profile and emails the client directly asking for it: a W-2, a K-1, whatever the situation calls for. No one has to remember to chase it.
• The client uploads documents to a secure vault in whatever format they have — PDF, JPG, PNG, spreadsheets. This includes handwritten notes and receipts, which matters more than it sounds: a lot of client records aren't clean digital files, they're a shoebox's worth of paper accumulated because nothing was organized along the way. The system reads handwriting and cross-checks it against everything else that's come in.
• Before the advisor call even happens, the system drafts a scope of work — which services this client actually needs, and a rough price — from the discovery call transcript and the uploaded documents. The advisor walks in already briefed instead of running discovery twice.
• After the advisor call ends, the engagement letter goes out automatically — typically 10 to 15 minutes later, while the conversation is still fresh for the client.
• The client signs electronically, the invoice goes out, and onboarding starts — which is where the rest of the pipeline picks up.

The point isn't that any single one of these steps is hard to do by hand. It's that a firm doing all seven manually, for every prospect, is running an assembly line with a person standing at every station — and every station is a place a Friday-afternoon prospect quietly goes cold before Monday.

Why the advisor still has to say yes

None of this sends anything binding without a human looking at it first. The scope of work that's ready before the advisor call is a draft for the advisor to walk in with, not a decision made on their behalf — they can adjust pricing, add or drop services, or throw it out entirely if the conversation goes somewhere the transcript didn't predict. The engagement letter that goes out 10–15 minutes after the call follows the same rule: it's built from what was actually agreed on that call, and the advisor's approval is what releases it, not the system's judgment about what should have been agreed.

That's a deliberate design choice, not a limitation we haven't gotten around to removing. A CPA firm's engagement letter is a contract. The AI's job is to remove the drafting delay, not to replace the judgment call about what the firm is actually agreeing to deliver.

The AI never actually knows whose data it's looking at

This is the part most "AI for accountants" pitches skip, and it's the part that actually matters to a compliance-minded partner: documents are read by classical OCR (optical character recognition — software that turns a scanned document or photo into structured text and numbers) before the AI ever touches them. Some of the strongest OCR results we've seen for messy scanned documents at this stage have come from Google's document AI models, alongside the Azure equivalents — either way, this step is deliberately not an AI reasoning step at all.

That ordering is the whole point. By the time the AI is reasoning about the file, it isn't looking at "this client's brokerage statement." It's looking at a set of structured figures — a fund value, a purchase date, a form type — with the identifying wrapper already stripped off. The AI can tell you that a client holds units in a foreign fund that trigger PFIC reporting (Passive Foreign Investment Company — a US tax classification for owning a foreign mutual fund or ETF, which forces extra IRS filings and often punitive tax math) without ever being told whose fund it is.

That's a different security posture than "we encrypt everything," which is table stakes and says nothing about what the model itself ever sees. It's an architecture decision: keep the raw, identifying documents out of the model's context entirely, and only ever hand it de-identified, structured data. For a firm handling client financial and tax information, that distinction is the difference between a real answer to a partner's question and a reassuring sentence that doesn't hold up under a second question.

It runs inside the software the firm already has — not next to it

The firm we built this for was explicit going in: they didn't want another login. They wanted their existing stack to get smarter, not a new tool their team would have to context-switch into on top of everything else. That's why this sits on top of Firm360 and UltraTax CS rather than replacing either one — we've written separately about what it actually takes to get AI vendor access into Firm360 and what we built in front of UltraTax CS for tax season document triage. This piece runs before either of those workflows starts: the discovery-call-to-signed-letter stage, not the return-prep stage.

That's also the more general lesson underneath all of this. There's no single tool that fits every CPA firm's stack, and there shouldn't be — the economics, software, and workflow of a five-partner firm and a fifty-person firm aren't the same business. The job is understanding what a specific firm already runs and building the connective tissue between the pieces, not selling them a new system to learn on top of the ones they've already paid for and trained their staff on.

In practice that means the actual build work is less about the AI model and more about the wiring: pulling a transcript out of whatever call tool the firm uses, writing a structured profile into whatever practice management system holds client records, and dropping a finished letter into whatever e-signature tool the firm already has a contract with. The AI step — reading the transcript, drafting the scope, flagging missing documents — is often the smaller half of the engineering work. Getting it to talk to Firm360 and UltraTax CS without the firm changing how it already operates is the bigger half, and the part a generic AI chatbot plugged into a browser tab can't do at all.

Frequently asked questions

Does the AI see our clients' actual financial documents?

No. Documents are processed by OCR first, which converts them into structured data — numbers, dates, form types — before the AI step runs at all. The AI works with that structured data, not the raw, identity-bearing document itself.

What if a client's records are handwritten or disorganized?

The system reads handwritten notes and receipts, not just typed or digital documents, and cross-checks them against everything else the client has uploaded to catch mismatches or gaps before they become a problem during the actual return.

Does this replace the discovery call or the advisor's judgment?

No. It removes the busywork around the call — note-taking, re-explaining context, drafting the scope and letter from scratch — so the advisor's time goes into the conversation and the decision. The advisor still approves the scope, the pricing, and the letter before anything goes out.

How does this connect to onboarding once the letter is signed?

Directly. This system hands off a signed engagement letter and an invoice; from there, the onboarding pipeline takes the client from signed letter to fully onboarded.

Does this work if our firm uses different software than Firm360 or UltraTax CS?

Yes — the pattern (transcribe, structure, detect gaps, draft, route for human approval) isn't specific to those two tools. What changes firm to firm is which systems it needs to talk to, which is exactly the part worth a real conversation before assuming a fixed price or a fixed build.

What this is worth finding out for your firm

If your firm is losing days between a good first conversation and a signed engagement letter, that's a specific, fixable gap in the plumbing — not a hiring problem, and not something clients will tell you about directly before they quietly go with someone else. Firms in CPA and tax practices can get a free read on where their own capacity is actually going with our CPA Tax Season Capacity Calculator, no email required.