What does AI actually do inside an accounting or tax firm?

It works the administrative load that swells every busy season and has nothing to do with judgment: chasing the specific documents each client has not sent, answering 'is my return done?' from your workflow status, capturing new-client inquiries you cannot take in March, and making the firm findable for its real specialties. Preparation, review and advice stay with your staff. The system moves paper and messages so they have more hours for the work only they can do.

Will the system answer clients' tax questions?

No. A question about whether something is deductible, how a transaction should be reported, or what a client owes is professional advice, and it depends on facts a chat window does not have. The system recognizes those questions, records them accurately with the client's details, and routes them to the right person with a stated reply time. Answering them automatically would put an unreviewed opinion in your firm's name.

How does automated document chasing know what a client is missing?

It starts from what the firm already knows. Last year's return shows which forms the client usually receives — W-2s, 1099s, K-1s, mortgage interest statements — and your portal shows which have arrived. The difference is a specific list, so the reminder names the document and the issuer instead of asking for 'your tax documents'. When a K-1 typically arrives late, the system expects that and does not nag the client for something they cannot have yet.

Is it safe to use AI tools with client tax information?

It can be, if the build is designed around the rules first. Tax preparers are covered by the FTC Safeguards Rule and the IRS expects a written information security plan, and section 7216 of the tax code restricts how return information is used and disclosed. We keep return data inside systems the firm controls, limit what any outside tool can see, and leave the consent and policy decisions with the firm and its own advisers.

Does this work with our practice management and tax software?

That is the first question we answer. Practice management platforms such as Karbon, TaxDome, Canopy and Financial Cents differ in what an outside system can read and write, and desktop tax software is usually more closed still. We establish what your stack allows before designing anything. Where a direct connection is not possible, the build produces clean, structured tasks your staff can act on, rather than pretending to a sync that does not exist.

When should an accounting firm start building this?

Before the season, not during it. The document-chasing and status layers need to be running when the first organizers go out, and nobody has time to configure anything once returns start arriving. For a calendar-year tax practice, that means November and December. A firm that misses that window is usually better off starting with the new-client capture layer, which can go live at any time without disrupting the workflow.

AI for Accounting Firms: Busy Season Is a Document Problem, Not a Staffing Problem

By Mike Evan — Founder, Social Media Strategy HQ•Updated November 2026

Most of a tax firm’s lost season hours go to waiting on and chasing documents, not preparing returns. Social Media Strategy HQ builds four systems for accounting firms: client-specific document chasing, return status answered from your workflow, capture for inquiries you cannot take in March, and findability for your real specialties.

Ask Where the March Hours Actually Went

Every accounting firm owner has a theory about why busy season breaks them, and it is usually staffing: not enough preparers, not enough reviewers, not enough seniors who can sign off. Sometimes that is true. But sit with a firm’s staff for a week in March and count where the hours go, and a different picture appears. A large share of the day is not preparation or review. It is email asking for a brokerage statement, phone calls asking whether a return is finished, hunting through a portal to see whether the second W-2 arrived, and re-opening files that stalled waiting on one document.

That work is real, it is necessary, and none of it requires a credential. It is the part of the season an AI layer can carry. The professional work — preparation, review, judgment, advice — stays exactly where it is. The point is to give those people back the hours the paperwork was eating.

One practical note before anything else: this is November. A calendar-year tax practice has roughly ten weeks before organizers go out, and that is the window where these systems can be built and tested. Once returns start arriving, nobody has the attention to configure anything.

Layer One: Chase the Specific Document, Not “Your Tax Documents”

The standard reminder — “please upload your tax documents” — is almost useless, because the client does not know what you are missing and assumes they sent everything. The firm, on the other hand, already knows. Last year’s return lists every information return the client received: the W-2s, the 1099-INT from a particular bank, the 1099-B from a particular brokerage, the K-1 from a particular partnership, the 1098 from a particular lender. The portal shows which of those have arrived this year. The difference between the two is a precise list.

What the chasing layer does with that list

It sends reminders that name the document and the issuer, on a cadence that escalates gently, and stops the moment the document lands. It understands timing, too: brokerage statements are often corrected and reissued, and partnership K-1s commonly arrive late in the season, so the system expects them later and does not nag a client for something they cannot have yet. It also asks the one question prior-year data cannot answer — did anything new happen this year, such as a new job, a sale of property, a new rental, or a business started? — and routes the answer to the preparer. The plumbing is our business automation work.

Why the file stays open longer than it should

A return that is ninety percent complete and waiting on one document is the most expensive object in a tax office, because someone has to pick it up again and remember where they were. A chase list that tells the preparer exactly which file is now complete, the day the last document lands, cuts that re-entry cost more than any speed-up in the preparation itself.

Layer Two: “Is My Return Done?” Should Never Reach a Person

Status questions are the purest overhead in the season. The answer already exists, as a stage in your practice management system: received, in preparation, waiting on documents, in review, ready to sign, filed. The client cannot see it, so they call or email, and someone stops working to look it up.

The status layer pushes a short, plain message when a client’s file changes stage, and answers status questions by phone, text or chat from the same source. It says what is happening and what, if anything, the firm is waiting on from the client — which turns many status calls into the document they were supposed to send. Anything beyond status goes to a person, with the context attached. The conversational piece runs on our AI customer service layer, built with a visible route to a human, for the reasons in our report on why customers resist AI customer service.

Layer Three: The Client You Cannot Take in March

Many firms stop taking new individual tax clients partway through the season, which is sensible. What happens to the inquiries that keep coming is usually not. They hit a voicemail, an unmonitored inbox or a form that says nothing, and they are gone. Some of those people needed a return this April and would never have fit. Many wanted bookkeeping, a business entity conversation, or a planning relationship — exactly the year-round work most firms say they want more of.

The capture layer answers every inquiry, states plainly what the firm is and is not taking right now, asks the handful of questions that sort the inquiry (individual or business, which services, which states, what timeline), and puts it in a queue the firm works when capacity opens, with an honest reply time. That capture work sits inside our AI lead generation build. It is the request pattern from our report on the booking gap: when an appointment cannot honestly be offered, a request with a kept promise beats silence.

The Advice Line, and the Data Line Under It

Nothing we build tells a client whether something is deductible, how to report a transaction, what they will owe or whether to make an estimated payment. Those are professional judgments that depend on facts a message thread does not contain. The system recognizes those questions, records them with the client’s details, and routes them to the right person. It is the same boundary we hold for law practices and insurance agencies.

Under the advice line sits data. Tax preparers are covered by the FTC Safeguards Rule and are expected to keep a written information security plan, and section 7216 of the tax code restricts how return information may be used and disclosed. That last rule has a consequence firms often miss: using return data to market services other than tax preparation can require the client’s written consent. The chasing layer uses return data to do the engagement the client hired you for; anything beyond that is a policy decision for the firm and its own advisers, made before launch.

Layer Four: Being Named for What You Actually Do

People increasingly ask an AI assistant for an accountant who handles crypto, multi-state returns, rental property, S corporations, or a particular industry. Assistants can only name firms whose websites state those specialties plainly. Most firm websites list the same generic services as every other firm — tax, accounting, advisory — and leave out the specialties that actually decide who gets the call. The fix is unglamorous: one page per real specialty, written so a sentence like “we prepare multi-state returns for remote employees” can be lifted and quoted, plus the same specialties stated on your Google Business Profile and professional listings. Our explainer on why ChatGPT does not recommend your business covers how assistants pick names. We handle the specialty pages as SEO and answer engine optimization work on a website the firm can edit itself.

What We Do Not Do

To be explicit: we do not build tools that prepare returns, give tax advice, estimate liabilities or recommend positions. We do not put return information into tools the firm has not approved, and we do not decide your consent policy. We do not write marketing claims your professional standards may restrict; that review belongs with the firm. Nor will you hear a forecast of new clients or search positions from us; anyone who offers one is guessing.

Where We Start

We open with an inventory rather than a pitch: which systems hold your client list, your documents and your workflow stages, and which of them will accept an outside connection. Alongside it, a rough split of last season’s staff time between preparing returns and chasing them. Document chasing usually comes first because its payback lands in the first month of the season; status messages follow, then closed-season capture, with the search work running underneath all three. Because we build with Claude Code, a firm can be live before January rather than next autumn, and every client-facing message is approved by a partner before it goes anywhere. A firm whose chasing is already disciplined gets told so, and we start somewhere else.