AI for accounting firms, starting with the documents before the accounting.
An engagement often slows down before the accounting begins. A statement arrives without a clear period, a client sends most of the files but misses one, and staff spend hours sorting, matching and chasing. We build AI that organizes incoming documents, matches them to the right client and period, and drafts the follow-up request, so your accountants spend their time on what the documents mean.
Where intake work eats a firm's week.
Sorting before accounting
Files arrive through a portal and a mailbox with unclear names and periods. Someone has to work out which client, which month and which checklist item each one is.
Chasing the missing item
A client sends nine of ten documents. Finding the gap and writing a specific request takes longer than it should, and generic reminders annoy clients.
Wrong client, wrong period
A correct summary attached to the wrong client is a serious failure. Similar company names, subsidiaries and multi-period files make matching risky.
Rules on client data
Tax preparation can fall under the FTC Safeguards Rule, and Section 7216 limits how tax return information is used and disclosed. AI tools have to fit inside both.
AI for the preparation work around client engagements.
Document intake and matching
Classifies each incoming file, matches it to the client and reporting period, and attaches it to the engagement checklist. Uncertain matches go to a person.
Missing item requests
Compares what arrived with the checklist and drafts a specific request for what is missing, with the evidence for each gap. A coordinator approves before it is sent.
Review-ready packets
Gives the accountant a summary of what arrived and what is unresolved, with source links, so review starts from a clear picture.
Close evidence packs
Once intake is reliable, assembles the evidence for a monthly close with open items flagged for the accountant.
Client status updates
Drafts a plain-language update on where an engagement stands, for staff to edit and send.
Connected to your tools
Works with your portal and practice management software, and uses their own features first where they already cover the job.
AI prepares the work. Your accountants make the judgments.
A complete checklist does not mean the evidence is sufficient. The assistant can surface unexplained transactions or a wrong entity, but the accountant decides whether the work can proceed. Calculations and reconciliations stay in structured software, and any explanation an AI writes has to match those computed numbers.
We start with one recurring service, usually monthly bookkeeping intake, and keep tax return information out of the first project until its processing has been reviewed under the FTC Safeguards Rule and Section 7216. Journal posting and tax positions are separate scopes with their own controls.
Document intake for a 70-person firm, as an example.
This is an illustrative scenario to show how a first project runs, not a QueryNow client result. Every number is an assumption to replace with your own.
- 01
Scope
Before the build800 recurring bookkeeping and close packets a month arriving through a portal and a mailbox. Tax return material is excluded at the start.
- 02
Build
Two weeksClassification, client and period matching, checklist comparison, duplicate and date checks, and drafted follow-up requests, all inside your environment.
- 03
Test
Before go-liveA locked set of about 200 packets, including mixed-client files, wrong-year records and empty scans. Proposed gates: at least 97% correct client and period matching, at least 95% of missing items found, and zero cross-client disclosures.
- 04
Measure
First monthsTime per packet including review, unnecessary client requests, and days waiting for documents, tracked separately for busy season.
How the monthly value is worked out.
Hypothetical figures from the example above. Your volume, minutes and review effort decide the real answer.
| Input | Example value |
|---|---|
| Packets a month | 800 |
| Minutes per packet, before and after including review | 18 before, 10 after |
| Hours recovered a month | About 107 |
| Loaded hourly cost | $50 |
| Share of recovered time that becomes useful work | 70% |
| Monthly operating cost assumed | $1,200 |
| Net monthly capacity value | About $2,533 |
| At half the volume | About $667 |
One workflow, built in two weeks, paid when it works.
We scope one recurring intake process with you, agree the acceptance tests before we start, and build it in your own environment in two weeks. You pay $10,000 only after every test passes, and client data stays in your systems.
Bring a redacted packet, your checklist and one follow-up that needed manual investigation. That is enough to see whether your volume and records justify a build.
Sources for the rules on this page.
Checked in October 2026.
What the first project gives your firm.
Matched packets
Each document attached to the right client, period and checklist item.
Specific requests
Follow-ups that name the missing item and why.
A clear review start
A summary of what arrived and what is still open.
Client boundaries held
Mixed-client files stop and go to a person.
Measured results
Time per packet and waiting days, before and after.
Room to grow
Close packs and status updates once intake is reliable.
From workflow to working tool in two weeks.
Scope and sign
We define the workflow, the deliverables, and the acceptance criteria, and sign an agreement on them before anything starts.
Build
We build the tool in your environment, with Claude Code and automated evaluation against your own data.
Pay when it works
$10,000, due only after every criterion in the signed agreement is met. Nothing before that.
One intake workflow built in two weeks in your environment, paid only after it passes the acceptance tests you signed.
What AI costs, including after the build →
Tell us the workflow →Questions buyers ask.
What is a good first AI project for an accounting firm?
Document intake for one recurring service, such as monthly bookkeeping. The AI sorts and matches incoming files, compares them with the checklist and drafts requests for missing items, and staff approve before anything goes to a client.
Can AI post journal entries or prepare tax positions?
Those are separate projects with their own authorization and reconciliation controls. A first intake project prepares documents and requests; accountants keep the judgments.
Does the FTC Safeguards Rule apply to AI tools?
Tax preparation firms can fall under the FTC Safeguards Rule, and a covered firm's security program has to account for service providers and AI data flows. Section 7216 separately limits how tax return information is used and disclosed.
Our practice software already has AI features. Do we need a custom build?
Use those features first where they cover the job. A custom workflow makes sense when your process spans several systems or needs firm-specific rules the existing tools cannot express.
How do we measure whether it works?
Track time per packet including review, how often client and period are matched correctly, unnecessary client requests and days waiting for documents, with busy season measured separately.
How much does it cost?
One workflow is $10,000, built in two weeks and paid only after it passes your acceptance tests. Model usage and hosting run on your own cloud bill.
Ready to ship AI that actually works?
Send us the workflow. We return a fixed scope, price, and acceptance criteria in 48 hours. You pay $10,000 only after it works.
Tell us the workflow →The mid-market AI advantage: How $50M to $500M companies out-ship the Fortune 500
Why $50M to $500M companies get AI into production faster than the Fortune 500.
Read it free →