AI now does real work in accounting — the repetitive, high-volume work that used to fill the bill. It still cannot exercise professional judgment. Here is exactly where we draw that line on your engagement.
Book a 30-Minute ReviewThe Dividing Line
Most firms will either tell you AI changes everything or that it changes nothing. Neither is true, and the useful answer is specific.
AI is very good at high-volume, repetitive, rule-shaped work: reading documents, classifying transactions, matching receipts, tying numbers between schedules, and scanning an entire population for anomalies rather than checking a sample. That work used to consume the majority of the hours on an accounting or audit engagement, and it is the work clients hated paying for.
What AI cannot do is exercise professional judgment. It cannot decide whether a grant restriction has been met, whether an accrual is reasonable, whether a related-party transaction smells wrong, or whether the evidence in front of it is sufficient. It also fails quietly — it produces a confident answer on the edge case it should have escalated.
That is why our position is simple: the machine does the volume, a person carries the responsibility, and we tell you exactly where the line sits on your engagement.
Three Service Lines
The role AI plays is different in each of these, and so is the point at which a human takes over.
We build an accounting system around your bank feeds, chart of accounts and invoicing, then clean up as much as five years of back books. The AI handles the grind and asks you about the rest in plain language. See how it works →
We test the whole population, not a 40-item sample. Journal entries, duplicate payments, Benford's Law, ratio and trend analytics, cash-flow tie-out — run across every transaction, with preparer, reviewer and partner sign-off on the workpapers. See how we audit →
Document intake, extraction from K-1s, 1099s and W-2s, prior-year comparison and scenario modeling are automated. Positions, elections and judgment calls are not, and a CPA signs the return. See the tax approach →
How We Operate
Automation is only useful in accounting if it is governed. These four rules apply to every system we build and every engagement we run.
Every system we build names a specific person who reviews AI output before it becomes part of the books. Not “someone on the team” — a name. AI can be confidently wrong, and it fails quietly on the unusual transactions rather than the obvious ones.
No system we set up is given authority to initiate a payment, transfer, or disbursement. Automation reads, classifies, flags and drafts. A person authorizes.
Allocation of shared costs, revenue recognition judgments, functional expense classification — these are policy decisions. They get written down by a person and the software follows the policy. The software does not invent the policy.
If a machine touches your books, an auditor will ask how it works and who checks it. We document that from day one, so the answer already exists when the question arrives.
Independence
We will build your accounting system, or we will audit you. Never both for the same organization.
If a firm builds and maintains the records it later audits, it is reviewing its own work. Professional standards treat that as a self-review threat, and an opinion issued without independence is worth very little to a board, a funder, or a federal agency reviewing a Single Audit.
So we ask early which one you need. If you are an audit client and you want the AI accounting build, we will refer you to another provider and stay your auditor. If we build your system and you later need an audit, we will refer you to another firm and stay your accountant.
You keep the choice either way, and you never end up with an opinion someone can discount.
Common Questions
Straight answers, including the ones that are less flattering to the technology.
No. AI replaces accounting tasks, not accountants. It handles categorization, matching, extraction and full-population testing — the repetitive work. It cannot exercise professional judgment, and it fails quietly on exactly the unusual transactions that need a person. The AICPA's position is that AI does not absolve a practitioner of professional responsibility; output is reviewed like staff work.
Four places, mainly: transaction categorization and receipt matching in bookkeeping; document extraction from K-1s, 1099s and W-2s in tax; anomaly detection across an entire transaction population in audit; and reconciliation. In each case a named human reviews the output before it becomes part of the books, a return, or audit evidence.
It can do most of the routine work — categorizing transactions, matching receipts, catching duplicates and odd charges, and reconciling feeds. It cannot set your accounting policy, decide how shared costs are allocated, or judge whether a revenue item has been earned. Those are written human decisions the software then follows.
It can do the mechanical preparation. Industry analysis puts more than eighty percent of the mechanical work on an individual return within reach of automation — intake, extraction, prior-year comparison. It cannot choose a tax position, make an election, or represent you under examination. A paid preparer signs the return and is responsible for it.
No, but it changes what an auditor spends time on. Analytics can test every transaction rather than a forty-item sample, which means the team examines items the analysis flagged instead of items a formula selected. The opinion is still signed by a CPA who is accountable for it, and the use of the tool is documented in the file.
It is safe under three conditions, and risky without them: a named person reviews AI output before it enters the books, no automation is ever given authority to move money, and your accounting policy stays a written human decision rather than something the software invents. Every system we build is set up that way.
Yes, and it is a good fit. Nonprofit and Single Audit engagements involve high transaction volume against restricted funds and federal award requirements — the pattern anomaly detection handles well. Compliance checklists are tied to Uniform Guidance, and preparer, reviewer and partner sign-offs are recorded on every key item.
Why Take Our Word For It
Most writing about AI in accounting is written by people who do not sign anything.
Summit Advisory performs nonprofit, Single Audit, 401(k), government, PCAOB and business audits, plus tax and advisory work. The AI questions on this page are ones we answer for our own engagements before we answer them for yours.
Our audits are performed on an AI-assisted engagement platform with phase-gated workflow, standards-linked checklists and three levels of sign-off. See how that works →
We set up AI accounting systems for organizations and clean up as much as five years of back books — so we know where the technology breaks, not just where it demos well. See the build →
We publish what AI cannot do alongside what it can, and we do not claim an automation percentage we cannot substantiate. If a vendor number is doing the work in a claim, we say whose number it is.
Everything on this page condensed into a single PDF you can forward to a board or a partner — the three service lines, who signs, the pricing, and the limits.
Get the One-Page Guide →