CCH Axcess added agentic AI to its tax preparation workflow. The pitch, per CPA Practice Advisor, is that it reduces repetitive preparation work, increases capacity, and gives preparers a stronger starting point for review. That's a reasonable claim about the prep step. It says nothing about the review step — and the review step is where the actual bottleneck in most firms lives.
This matters because the industry conversation around AI in tax prep keeps treating "draft the return faster" as the hard problem. It isn't, anymore. The hard problem is what happens after the draft exists: who reviews it, in what order, against what checklist, with what sign-off, and what happens to that trail when a regulator or a client asks about it eighteen months later. An AI that drafts a return doesn't touch any of that. It just produces more drafts, faster, that still have to go through the same review process your firm runs today.
Agentic AI Changes What Gets Reviewed, Not How You Review It
"Agentic" in this context means the AI isn't just suggesting line items — it's working through the return with some autonomy, pulling data, applying logic, and producing something closer to a finished draft than a first pass. That's a real shift from earlier tax automation, which mostly handled data entry and OCR.
But a stronger draft doesn't change the structural question every firm already has to answer for every return: who is responsible for reviewing it, how do you know where it sits in the review pipeline right now, and what happens if the reviewer doesn't get to it before the deadline. Agentic AI makes the first stage faster. It does nothing for the second and third stages unless the firm already has infrastructure built for them.
Put plainly: if your review process today is a shared drive, a few emails, and someone's memory of whose returns are "almost done," an AI-drafted return just means more returns sitting in that same undisciplined queue, arriving faster than before. You haven't added capacity. You've added volume to a bottleneck.
The Capacity Argument Assumes Infrastructure Most Firms Don't Have
"Increase capacity" is the headline claim, and it's worth taking seriously, because capacity is the real constraint for most CPA firms during busy season — not the number of hours it takes to type numbers into a return. If agentic AI genuinely removes hours of repetitive prep work, that time has to go somewhere. It only becomes capacity if the firm can redirect it without creating a new jam downstream.
That redirection depends on three things most firms underinvest in:
- Status tracking. At any moment, can a partner see which returns are drafted, which are in review, which are stuck waiting on a client document, and which are overdue — without asking anyone? If the answer requires a status meeting or a phone call, the firm doesn't have a capacity problem to solve yet. It has a visibility problem.
- Review queues with assignment logic. A faster draft needs a reviewer who is notified the moment it's ready, routed based on workload and expertise, and accountable for turning it around in a defined window. Without that, drafts pile up waiting for whoever remembers to check.
- An audit trail. Who approved what, when, and based on what version of the return. This isn't optional paperwork — it's the thing that protects the firm when a client disputes a position, a regulator asks questions, or a staff member who handled the engagement has since left. An AI-drafted return with no record of who reviewed it and signed off is a liability, not a time saver.
None of this is about the AI. It's about whether the firm's practice workflow was built to absorb a faster first step. For firms running review on spreadsheets and email threads, it generally wasn't.
Why Spreadsheets and Email Break Down Specifically Here
A spreadsheet tracker for engagement status works fine at low volume with one office and a handful of staff checking it. It breaks down exactly where agentic AI is supposed to create value: at scale, across multiple preparers and reviewers, during the weeks when volume peaks.
The failure modes are familiar to anyone who has run a tax season:
- Two people update the same tracker with conflicting information, and nobody notices until a deadline is missed.
- A reviewer finishes a return and emails the next person, but that email sits unread for two days during a busy stretch.
- A multi-office firm has no single view of where every engagement stands, so each office manages its own version of "done," and partners can't see firm-wide risk until it's already a problem.
- When a question comes up about who approved a position on a return, the answer lives in someone's inbox, if it exists at all.
Faster drafts make all of this worse, not better, because the volume moving through the pipeline increases while the tracking mechanism stays the same. The firm experiences the AI as noise rather than capacity.
What a Real Review Pipeline Looks Like
We built an audit and tax engagement platform for a multi-office CPA firm handling more than 500 engagements a year, specifically because the firm's existing process — tracking spreadsheets, email approvals, and a patchwork of document tools — couldn't give partners a reliable answer to "where does everything stand" during peak season. The system we built includes a firm-wide deadline dashboard, DocuSign for sign-off, SmartVault for document management, and a client portal so clients aren't a bottleneck waiting on email replies.
The design principle that matters most here isn't any single feature — it's that every engagement has a status that updates automatically as work moves through it, every review step is assigned to a specific person with a visible deadline, and every approval is logged with a timestamp and a record of what was approved. That's the infrastructure that would let a firm actually benefit from a faster-drafted return. Without it, the draft just moves the backlog from "waiting to be prepared" to "waiting to be reviewed," which is not a win.
This is also where the build-versus-buy decision gets real for a lot of firms. Off-the-shelf practice management tools generally give you generic status fields and generic approval steps. They work if your review process matches the vendor's assumptions about how a firm operates. Multi-office firms, firms with non-standard sign-off chains, or firms that want the deadline dashboard to reflect how partners actually make staffing decisions usually find the generic version close but not quite right — and "close but not quite right" is exactly the gap that turns into spreadsheets and side-channel emails six months after go-live.
What to Build Before You Adopt AI-Drafted Prep
If your firm is evaluating CCH Axcess's agentic AI or any similar tool, the sequence matters. Adopting the AI first and figuring out review capacity later guarantees a bottleneck shows up downstream, usually during the exact weeks you can least afford it.
Before rolling out AI-assisted drafting at any scale, a firm should be able to answer, without a meeting:
- Where does every active engagement stand right now, across every office?
- Who is the assigned reviewer for each one, and when is it due?
- What happens automatically when a draft is ready — does a reviewer get notified, or does someone have to remember to check?
- If a client or regulator asks who approved a position six months from now, can you produce the answer in minutes?
If any of those require digging through email or asking a staff member to remember, the firm isn't ready to convert faster drafting into more capacity. It's ready to convert faster drafting into a bigger pile of unreviewed returns.
The Bottom Line
Agentic AI in tax prep is a real improvement to one step in a multi-step process. It is not, by itself, a capacity strategy. Capacity comes from the whole pipeline — draft, review, approval, delivery — moving in sync, with visibility at every stage and a record of every decision. Firms that already have that infrastructure will see real gains from faster drafting. Firms that don't will see the same review backlog they have today, just arriving under more pressure.
Before adopting the AI, audit your own review pipeline the way you'd audit a client's internal controls: where are the manual handoffs, where does status live only in someone's head, and where would an approval record not survive a serious question. Fix that first. The AI will still be there once you have somewhere for its output to go.