Big 4 Accounting Software vs a Small Firm's AI Agents

By Jude Lee · · Comparison

Two accountants reviewing a trial balance and variance report on a monitor in a small firm office

What accounting automation actually means now

Automation in accounting used to mean one thing: rules. If a bank feed description contains “AMEX EPAYMENT,” code it to the card clearing account. Deterministic, auditable, cheap. That layer is still the backbone of most firms and it isn’t going anywhere.

What changed is that two new layers sit on top of it:

AI-assisted features — the categorization suggestions, anomaly flags, and draft narratives now bundled into ledgers and practice-management tools. You don’t build these; you turn them on.

AI agents — an assistant that takes multi-step actions across systems rather than answering a single question. Pull the trial balance, compare it to prior period, list every account with a variance over your threshold, check the GL detail behind each one, draft the flux commentary, and post it to the close workpaper for a human to sign. That’s a sequence, not a chat.

The practical difference: a chatbot gives you text; an agent changes the state of your systems. Which is exactly why access control and review gates matter more than model choice.

The software the largest firms actually run on

Each of the Big 4 publicly markets its own audit platform — EY Canvas and EY Helix, Deloitte Omnia, KPMG Clara, PwC Aura and Halo — and you can read their descriptions on the firms’ own sites. Beyond the branded platforms, it is a general industry pattern (not something verified firm by firm here) that large practices also layer in data preparation and analytics tooling, robotic process automation, and enterprise document management. The specific vendor mix varies by firm, service line, and territory, so treat any list you see online as directional rather than authoritative.

Here is the part worth stealing. Those platforms are not magic software; they are two ideas, industrialized:

  1. A governed data layer. Client data is extracted once, normalized, and made queryable — so procedures run against a consistent structure instead of whatever the ledger export happened to look like.
  2. Standardized, versioned procedures. The same test is executed the same way by every team, with the evidence and sign-off captured automatically.

A small firm that gets those two things right will outperform a small firm that bought five more subscriptions.

The Big 4 advantage was never the software brand. It was that every engagement ran the same procedure against the same shape of data.
— Accounting Ops Guide

Three ways a smaller firm can get there

Enterprise-style automation platform

What it is: RPA plus workflow orchestration plus a data warehouse — the small-firm echo of the Big 4 stack.

Fits when: you have genuinely high volume, stable inputs, and a person who can own the platform. Deterministic, fully auditable, no model variance.

Breaks when: inputs are messy or bespoke per client. Screen-scraping bots snap when a UI changes, and someone has to fix them during busy season.

Real cost: licenses plus the specialist to maintain it. See hire, upskill, or outsource before assuming the labor is free.

MCP-connected AI agent

What it is: an assistant like Claude given governed, least-privilege access to your ledger and practice-management systems through MCP (the Model Context Protocol, an open standard for connecting AI to tools and data), plus skills — packaged instructions that make it perform a specific job the same way every time.

Fits when: the work is judgment-adjacent and language-heavy: variance narratives, PBC follow-ups, workpaper prep, exception triage.

Breaks when: you need bit-exact reproducibility, or the task is pure arithmetic a formula already handles better. Also expect confidently wrong categorizations stated in the same tone as correct ones — the reviewer catches these only if the output carries source references and a confidence flag. Expect silent drift when a client’s chart of accounts changes and the skill still maps to the old accounts, which a periodic mapping check catches. And expect context limits on large GL extracts: the agent may quietly work from a truncated slice, so the human should verify row counts and period coverage against the ledger before signing.

Real cost: integration and governance work up front, then ongoing skill maintenance. Start with connecting an assistant to QuickBooks or Xero via MCP.

How CPA firms are actually using AI today

Stripping out the hype, the deployments that hold up are narrow and boring:

Notice the pattern: the agent does retrieval, assembly, and drafting. The human does judgment and sign-off. Anyone selling autonomous close is selling something — the sign-off is a professional judgment attached to a named person’s responsibility, and that responsibility cannot be delegated to a model no matter how good the draft is.

A worked example of accounting automation, end to end

  1. Define the job narrowly

    Say a firm with 22 monthly bookkeeping clients (an illustrative figure — use your own): “Prepare the month-end close package for our recurring bookkeeping clients through the point of preparer review.” Not “automate the close.”

  2. Expose data under least privilege

    Read-only access to trial balance, GL detail, and prior-period comparatives. Write access, if any, confined to a draft workpaper location — never to posting journal entries.

  3. Package the procedure as a skill

    Thresholds, account groupings, tick-mark conventions, and the exact output format. This is what makes run 12 identical to run 1.

  4. Insert the review gate

    The agent’s output arrives as a draft with every source reference attached. A person approves before anything leaves the firm. More on designing these in AI review gates, not autopilot.

  5. Log everything

    Every query, every action, every approval, timestamped. If you can’t reconstruct what the agent did, you can’t defend the workpaper.

The full sequence, with what to keep human at each stage, is laid out in month-end close automation with AI agents.

Modeling the payback with your own numbers

Don’t trust anyone’s published time savings, including ours — we don’t have your client mix. Build the estimate yourself:

hours × rate
Baseline: measure current prep hours per close before automating anything
Worked example — use your own timesheet data
× clients × 12
Annualize per-client savings across the recurring book
Worked example
− build − upkeep
Subtract integration cost and recurring maintenance
Worked example

Name the upkeep line concretely or your estimate will run optimistic. It includes: revising skills whenever a firm procedure, threshold, or chart of accounts changes; re-testing connections and outputs after ledger or practice-management API updates; periodic access and permission reviews; spot-checking a sample of agent output against source data; and staff time spent clearing the exception queue, which never goes to zero.

Then add the two effects most firms forget. First, reallocation: recovered preparer hours are only worth something if they move to billable or advisory work rather than evaporating. Second, capacity: if the constraint is hiring, the return may be taking on clients you’d otherwise decline — plug in your own realization rate and decide.

When the honest answer is “don’t build”

Skip the custom agent if the task runs a handful of times a year, if the process changes every cycle, or if a spreadsheet formula and a saved report already close the gap. Skip it if nobody in the firm will own the skill after the consultant leaves — an unmaintained agent degrades quietly, which is worse than no agent.

And skip it if your ledger vendor is about to ship the same feature. Buying beats building when the workflow is generic; building wins when the procedure is yours — your thresholds, your tick marks, your client-specific quirks — and that’s precisely the part off-the-shelf AI can’t know.

The firms that get real leverage from this aren’t the ones with the biggest tool list. They’re the ones that wrote their procedures down clearly enough that software could follow them.

Not sure where to start?

Get a free automation audit: we map your bookkeeping, month-end close, client onboarding, document collection, and AP/AR — and show you what's worth automating before you spend a dollar.

Get a free automation audit

Related guides