Serve more clients without more preparation hours.
An accounting or tax practice spends much of its capacity before any advice happens: chasing documents, coding transactions, reconciling accounts and reading letters from the tax authority. We look for the workflows where agents can do that preparation and checking, with controls that catch errors before a return or a set of accounts is signed. Tax advice, judgment positions and every sign-off stay with qualified people.
Six workflows that grow with every new client.
These are the workflows we would examine first in an accounting firm, tax advisory practice or bookkeeping service. They run through the practice's accounting and tax software, client portals, email and bank feeds.
| Workflow | Manual today | AI-native version | Stays human | Verification check |
|---|---|---|---|---|
| Client document collection | Staff send checklists, chase missing receipts and statements by email, and sort what arrives by hand. | An agent tracks what each client still owes, sends reminders, reads incoming documents and files them to the right client and period. | Client conversations about gaps and any decision to proceed with missing records. | Each document is matched to a client and period. Unmatched documents go to a queue, never a guess. |
| Transaction coding and bookkeeping | Bookkeepers code bank lines and receipts one by one, using memory of each client's habits. | Rules code the clear cases. An agent proposes coding for the rest with a confidence score and its reasoning. | Low-confidence items, VAT treatment questions and anything unusual for the client. | Coding must use the client's chart of accounts. VAT code must be valid for the account. Low confidence goes to a person. |
| Client bank and balance reconciliation | Staff tick bank statements against the ledger and carry unexplained items forward. | An agent matches statement lines to ledger entries and explains the remaining differences. | Adjustments, write-offs and journal approval. | Matched totals must equal statement balances. No journal posts without approval. |
| Tax return preparation | Preparers re-key figures from documents and prior-year files into the tax software and check them by hand. | An agent pre-fills the return from source documents and the prior year, and lists every figure with its source. | Tax positions, elections, reliefs claimed, review and filing sign-off. | Every figure links to a source document. Year-on-year changes above a threshold are flagged for the reviewer. |
| Tax authority correspondence | Staff open letters and notices, work out what they mean and the deadline, and route them to the right person. | An agent reads each notice, links it to the client and return, proposes the deadline and drafts a summary for the advisor. | Every response, objection and deadline confirmation. | A person confirms every deadline. Unconfirmed notices escalate daily. |
| Client onboarding and identity checks | Staff collect engagement letters, ID documents and ownership details, and run checks one client at a time. | An agent collects and validates documents, checks completeness for the risk level and runs the screening. | The risk assessment, acceptance of the client and every screening hit. | Required evidence must be complete for the risk tier. Screening hits are never cleared automatically. |
Six numbers that show whether it worked.
The Audit measures these before anything is built. The Pilot is measured against the same numbers.
| Metric | Definition |
|---|---|
| Preparation hours per client | People time per client per period on collection, coding and reconciliation, before review. |
| Auto-coding rate | Share of transactions coded with no manual correction at review. |
| Review corrections | Changes the reviewer makes to prepared accounts or returns, per 100 items reviewed. |
| Records complete on time | Share of clients whose documents are complete by the internal cut-off date. |
| Deadline exceptions | Filings or responses made late or close to the deadline, per quarter. |
| Clients per staff member | Active clients served per full-time preparer or bookkeeper. |
What can go wrong in a practice, and the control for each.
Tax advice and filings have legal consequences for clients, so they stay with qualified people. Agents prepare, match and flag. Every figure carries its source, and every agent step is written to an audit trail.
| Failure mode | Control | How we'd detect it |
|---|---|---|
| Wrong figure in a return or accounts | Every pre-filled figure links to a source document. A qualified reviewer signs off before filing. | Review corrections tracked per preparer, agent and cause. |
| Wrong VAT or tax treatment | Treatment rules are versioned. Unusual items and low-confidence coding go to a person. | Sampled review of coded items, and variance analysis against prior periods. |
| Document filed to the wrong client | Matching on client identifiers, not names alone. Unmatched documents wait in a queue. | Alert on any document whose identifiers conflict with its assigned client. |
| Deadline missed | A person confirms every deadline the agent proposes. Unconfirmed entries escalate. | Daily report of notices received with no confirmed deadline. |
| Screening hit cleared without review | Hits always go to a person. Clearance needs a recorded reason. | Audit trail of every hit, decision and reviewer. |
| Client financial data exposed | Agents work on one client at a time and see only the fields a task needs. EU-hosted deployment is possible. | Access logs per agent, per client and per task. |
Transaction coding, from diagnostic to pilot.
Illustrative example — not a client result. The numbers below are assumptions chosen to show the method and the arithmetic. Your Audit replaces them with measured values.
| 01 | Diagnostic | Suppose an accounting practice with a large bookkeeping client base picks "monthly transaction coding" on the Other / General Business track. The Snapshot might show a High leverage signal with Medium confidence, friction at the Processing and Verification stages, and Level 2 (workflow augmentation) as the indicated AI depth. |
| 02 | Audit | Suppose the Audit measures 48,000 bank lines and receipts a month across all clients, at 30 seconds each. That is 48,000 × 0.5 = 24,000 minutes, or 400 hours a month, about 2.5 FTE at 160 hours. It also finds that about 85% follow a pattern the rules and agent can code with high confidence. |
| 03 | Pilot | Suppose the 85% high-confidence items are coded automatically, with a person checking a 5% sample at 30 seconds each. The 15% low-confidence items take 1 minute each, because they need a closer look. The Pilot measures whether that holds and whether review corrections go down. |
| Line | Assumption | Hours per month |
|---|---|---|
| Baseline | 48,000 items × 0.5 min | 400.0 |
| High confidence, after | 40,800 items, 2,040 sampled × 0.5 min | 17.0 |
| Low confidence, after | 7,200 items × 1 min | 120.0 |
| After, total | 17.0 + 120.0 | 137.0 |
| Difference | 400.0 − 137.0 (about 1.6 FTE) | 263.0 |
At an assumed loaded cost of €38 an hour, 263 hours is about €9,990 a month. For a practice, the bigger question is often capacity: whether the same team can take on more clients in the busy season. The Audit would also measure review corrections, because coding errors caught late cost far more than the coding itself. It would also count new costs: compute, accounting software integration and sample review. If the numbers don't hold, the recommendation is "don't proceed".
Where our proof comes from.
We'll be plain about this. Our production proof is in AdTech, not accounting or tax. We have not yet run these practice workflows for a client.
The closest relative is partner reconciliation, which we run in AdTech production: matching what was delivered against what was billed, with a verification check before anything commits. The architecture is the same: typed intake, deterministic checks, every figure traced to its source, bounded retries, then a person. Our case studies show the production work, and five AI plugins built on it are live at reviveadservermod.com.
Our engineering team and delivery network have built related systems: invoice and business document AI with layout-aware parsing and data extraction, and an assistant that answers from internal procedures. See Selected builds.
That is why practice work starts with the free diagnostic and then the Audit. The Audit measures your own baseline and records what must never be automated, before anything is designed.
Questions practice owners ask first.
Will an agent give tax advice or file returns?
No. Tax advice, tax positions and filing sign-off stay with qualified people in your practice. Agents prepare figures and drafts, each linked to its source, for a person to review.
Do we need to change our accounting or tax software?
No. We work with the software you have, through its interfaces or exports. The Audit records which systems hold the data and where it is re-keyed.
Our software already suggests coding. What is different?
Built-in suggestions help inside one system. We look at the whole workflow, from document collection to review, and put checks and a confidence threshold on each step, so reviewers see only what needs them.
Find out which practice workflow is worth redesigning.
Start with the free diagnostic on the Other / General Business track. Move to the Audit only if the Snapshot points to something worth measuring.
| 01 | Free AI Operational Diagnostic | 10 steps, 8–12 minutes, General track. Snapshot on screen and as a PDF. |
| 02 | Operational Leverage Audit, from $1,500 | Fixed fee, one workflow, 10–15 working days after data access. Measured baseline, verification architecture, benefit model with stated assumptions, go / no-go. |
| 03 | Pilot implementation, from $6,000 | Fixed scope. The workflow built and run in your environment, with an audit trail, measured against the Audit baseline. Typically 4–8 weeks. |