Meet Caddi in personILTACONAug 23–27Future ProofSep 14–17
All articles
AI Use Cases · 2026

AI Use Cases for Accounting Firms in 2026

The administrative loops around the return, not the return itself, are where the payback is. Plus the §7216 question you have to answer before pointing any AI at a client's tax data.

The accounting profession is short of people in a way hiring cannot fix, and the work that is consuming your existing people is administrative. Accounting degree completions have fallen for eight consecutive years, the pipeline of new entrants has contracted by roughly a third over the past decade, and more than 300,000 professionals left the profession between 2020 and 2024. Meanwhile the hours that make busy season painful are not review judgment. They are chasing documents, filing them, assembling returns, chasing signatures, working rejects, and rekeying the same client detail into four systems.

0
consecutive years of declining accounting degree completions
0%
contraction in the pipeline of new entrants over the past decade
0%
of finance leaders report accounting and CPA talent shortages
Sources linked below. Figures describe the profession, not Caddi-specific results.

Sources: AICPA/NASBA Trends analysis; 2026 accountant shortage research; Controllers Council.

First, the part most AI articles skip: §7216

Before the use cases, the constraint. IRC §7216 imposes criminal penalties on a preparer who discloses or uses tax return information without proper consent, and transmitting that information to an AI vendor is generally treated as a disclosure unless an exception applies or the taxpayer has signed a compliant written consent. Layer on IRS Publication 4557, the FTC Safeguards Rule, and Gramm-Leach-Bliley, plus the Written Information Security Plan every paid preparer is required to maintain, and the picture is clear: pasting a client's K-1 into a general-purpose assistant is a consent problem before it is ever a quality problem.

This is where the architecture of the tool matters more than its benchmark scores. If AI is in the loop every time the workflow runs, then every run is a potential disclosure you have to paper. If AI is only present while the automation is being built, and the thing that actually touches the return at runtime is verified code calling your systems through their APIs, the consent surface collapses to something a WISP can describe honestly. Note also that the IRS WISP templates do not contemplate AI as a category at all, so the AI-vendor risk assessment is something your firm has to bolt on itself.

1. Source-document intake

Every 1040 season starts as a document avalanche: 1099s, W-2s, K-1s, brokerage statements, and closing documents arriving through the portal, the tax mailbox, and occasionally a photograph of a paper form. The automatable work is extract, name to convention, file to the right return, and tick the organizer checklist. Done well, the preparer opens a complete file instead of spending the first hour assembling one.

2. Open items and PBC chase

The single most reliable source of slipped deadlines is a request that nobody followed up on. Automating the chase across Suralink, Karbon, and TaxDome means outstanding items get pursued on a schedule rather than on someone's memory, responses are filed to the engagement, the item is closed, and only genuinely unresponsive clients escalate to a human.

3. Return assembly, delivery, and the 8879 chase

Assembly and delivery are mechanical, and the authorization chase is pure administrative overhead that peaks exactly when your people are most expensive. Automating package assembly, SafeSend delivery, signature follow-up, and status write-back to the engagement removes a job that consumes April and produces nothing a client values.

4. E-file rejects and batch extensions

A reject code is a routing problem with a known fix. Automation reads the code, sends it to the preparer who owns the return with the correction attached, refiles once resolved, and logs the outcome. The same foundation drives the batch extension run at the deadline so nothing slips because a queue went unwatched during the worst week of the year.

5. CAS monthly close

Client accounting services is the growth line for most firms and the most repetitive work in the building. Bank feed exceptions, transaction coding, reconciliations, and the close checklist across QuickBooks, Xero, Sage Intacct, and Bill.com are rules-driven by nature. Automating them turns the CAS team's month into reviewing a short exception list instead of clicking through every ledger to produce one.

6. WIP, billing, and realization

Realization leaks in the gap between work performed and work billed. Pulling time and WIP, assembling invoices to the firm's rules, routing for partner approval, and running the receivables follow-up sequence is the least glamorous automation on this list and frequently the one with the fastest cash payback.

7. Shared-mailbox triage and the compliance calendar

The tax@ and ap@ inboxes are a hidden queue with no owner. Classifying each message, filing attachments to the right client record, and routing the task to the owning preparer converts a constant interruption into a managed background process. The same mechanism drives the state registration and filing-deadline calendar so renewals stop being a surprise.

Built for the firm stack

  • QuickBooks
  • Xero
  • Sage Intacct
  • Karbon
  • Lacerte
  • Bill.com
  • DocuSign
  • SharePoint
The work spans tax software, practice management, the portal, document management, and the ledger, plus the everyday tools staff already use. Caddi connects across 150+ of them.

The real case: capacity is the scarce input

Each of these is worth doing alone. The strategic point is what happens when you sequence them. Burnout-driven turnover in public accounting runs 15% to 25% annually, and replacing a mid-level accountant costs $30,000 to $50,000 in recruiting, training, and lost productivity. Every hour of administrative work you take off a senior is an hour of capacity you did not have to hire, and a reason the senior stays.

There is a competitive dimension too. Private equity has been consolidating the profession aggressively: 177 direct investments drove 875 roll-up acquisitions between 2015 and 2025, and in 2025 each direct investment produced 7.6 additional transactions, with the consolidation index up fourfold since 2021. Sponsor-backed platforms are buying capacity and standardizing operations at speed. Independent firms competing against them for the same clients and the same staff need the capacity without the balance sheet.

The barrier has been that traditional automation means either a multi-year platform conversion or brittle bots that break the next time your tax software ships a UI change. Record-to-code is a different model.

Hit record
Screen-share the task once
Caddi writes it
As deterministic code
Runs unattended
Maintained for you
Record-to-code: someone on your team screen-shares the task once, Caddi writes it as deterministic code that runs over APIs with audit trails, built and maintained for the firm.

With Caddi, getting started is a screen share. Pick one high-volume loop, source-document intake or the open-items chase, record it, get it live with a baseline metric, then put it on a schedule. Reuse the same foundation for assembly and billing after that.

The firms that get busy season back will not be the ones that announced an AI strategy. They will be the ones that moved the filing, chasing, and rekeying off their people first, in a way their WISP could actually describe, and let the capacity compound while everyone else was still arguing about whether AI can prepare a return.

See these AI use cases built for your firm

Explore real workflows Caddi runs today, see the accounting firm overview, or book a demo to watch one of your own workflows built from a screen recording.

Do more with less

See Caddi in action

Tell us where to reach you and the calendar opens right here. In 30 minutes we'll show you how Caddi automates the back-office work that grows with your clients—built, run, and maintained for you.

Frequently asked questions

What are the top AI use cases for accounting firms in 2026?

The highest-ROI use cases are the administrative loops around the return, not the return itself: source-document intake from the portal and the tax mailbox, open-items and PBC chase, return assembly and delivery with 8879 follow-up, e-file reject handling and batch extensions, CAS monthly close, WIP and billing, K-1 tracking, and shared-mailbox triage. These are high-volume, rules-driven, and measurable, which is why they pay back inside a single season rather than across a multi-year platform project.

Can accounting firms use AI on client tax data legally?

Carefully, and the constraint is IRC §7216. Transmitting tax return information to an AI vendor is generally treated as a disclosure, which requires a specific, compliant written consent from the client unless an exception applies, and §7216 carries criminal penalties. On top of that, firms are subject to IRS Publication 4557, the FTC Safeguards Rule, and Gramm-Leach-Bliley, and every paid preparer must maintain a Written Information Security Plan that reflects the AI tools the firm actually uses. The IRS WISP templates do not contemplate AI as a category, so firms have to add the AI-vendor risk assessment themselves.

How bad is the accounting staffing shortage in 2026?

Structural rather than cyclical. Accounting degree completions have declined for eight consecutive years, the pipeline of new entrants has contracted by roughly a third over the past decade, and the profession lost more than 300,000 professionals between 2020 and 2024. Roughly 61% of finance leaders report minor or significant shortages of accounting, finance, and CPA talent. Burnout-driven turnover in public accounting runs 15% to 25% annually, and replacing a mid-level accountant costs $30,000 to $50,000. Hiring is not the lever it used to be.

Where should a firm start with automation?

Start with source-document intake or open-items chase. Both are high volume, both run every day of the season, both have an obvious owner, and both produce a metric you can show the partner group after one cycle: documents filed without a touch, or days to close an open item. Set the baseline before you deploy, keep a human on exceptions, and reuse the same foundation for assembly and delivery next.

How does Caddi automate firm workflows without changing tax software?

Caddi uses record-to-code: someone on your team screen-shares the workflow as they do it today, and Caddi writes it as deterministic code that runs unattended with a full audit trail across the systems you already own. AI is present when the automation is built, not when it executes, so tax return information is not handed to a model at runtime. It runs across CCH Axcess, UltraTax, Lacerte, Drake, Karbon, TaxDome, SafeSend, Suralink, QuickBooks, Xero, Sage Intacct, Bill.com, SharePoint, and DocuSign, with no conversion and no screen-scraping bots to maintain.