How Penny Reduces Your Client Review Workload

If you are an accountant managing bookkeeping oversight for multiple clients, you know exactly how the review cycle works. Transactions come in. You check them. Most are correct and unremarkable. A few need adjustment. Occasionally something unusual requires investigation. Then you document your review, prepare workpapers, and move on to the next client.

The problem is not that any individual review is particularly difficult. It is the cumulative volume. When you are managing fifty, eighty, or a hundred clients, each with hundreds of transactions per quarter, the sheer volume of routine checking becomes the bottleneck that limits your practice's capacity. This is the problem Penny was designed to solve.

Penny is Accounted's AI bookkeeper. She categorises transactions, processes receipts, handles client queries via WhatsApp, and — crucially for accountants — identifies which items need professional review and which do not. The result is a fundamentally different review experience: instead of checking everything, you check only what matters.

What Penny Does Before You See Anything

To understand how Penny reduces your workload, it helps to understand what happens to a transaction before it ever appears on your screen.

When a bank transaction arrives via the client's Open Banking feed, Penny processes it through a three-tier categorisation system. The first tier applies deterministic rules: recurring payments to known suppliers, transfers between the client's own accounts, and transactions that match established patterns exactly. These are categorised with confidence scores of 0.95 or above and auto-applied without requiring any human input.

The second tier handles transactions that are recognisable but not perfectly matched. A payment to a new branch of a known supplier, a variable-amount payment to a regular payee, or a transaction whose description contains identifiable keywords. These receive confidence scores typically between 0.80 and 0.94. Depending on the client's confidence threshold settings, they may be auto-applied or flagged for review.

The third tier uses large language model reasoning for genuinely ambiguous transactions. An unfamiliar payee with a vague description, a transaction that could plausibly be categorised in multiple ways, or a payment whose context requires understanding of the client's business. These receive lower confidence scores and are almost always flagged for review, with Penny providing her reasoning alongside her suggestion.

By the time you open a client's records, this processing has already happened. The routine transactions are categorised and documented. The ambiguous ones are identified and queued with context. Your starting point is not a raw bank statement but a curated list of items that specifically need your expertise.

The Numbers: Before and After Penny

The impact on review time is substantial and measurable. Here are the numbers we see across practices using Accounted.

Transaction Volume Reduction

For a typical sole trader client with 150-200 transactions per quarter:

• Without Penny: you review all 150-200 transactions, checking categorisation, VAT treatment, and completeness. Estimated time: 2-3 hours per client per quarter.
• With Penny: you review 15-30 flagged items (the 10-15% where Penny's confidence is below threshold or where she has identified an anomaly). Estimated time: 20-45 minutes per client per quarter.

That is a 70-85% reduction in review time per client. Across a practice of fifty clients, this translates to roughly 75-100 hours saved per quarter. At a chargeout rate of £100 per hour, that is £7,500-£10,000 worth of time reclaimed every three months.

Accuracy of Auto-Applied Categorisations

A natural question is whether the auto-applied categorisations are actually accurate. The data from across our accountant user base shows:

• First month: Penny's accuracy on auto-applied transactions (those above the 0.90 threshold) is typically 94-96%. The remaining 4-6% are minor categorisation differences (such as "office supplies" versus "stationery") rather than material errors.
• After three months: accuracy rises to 97-98% as Penny learns from the accountant's review corrections.
• After six months: accuracy typically exceeds 99% for clients with regular transaction patterns.

These figures reflect the learning loop built into Penny's design. Every time you amend a categorisation in the review queue, that correction feeds back into Penny's model for that specific client. She learns your preferences and applies them going forward.

Five Specific Ways Penny Reduces Your Workload

Beyond the headline numbers, there are specific workflow improvements that compound to create the overall time saving.

1. Receipt Matching

Traditionally, matching receipts to bank transactions is a manual process. The client provides receipts (in varying states of organisation), and you or your staff match each receipt to the corresponding transaction, verify the details, and flag any missing receipts.

Penny automates this entirely. When a client uploads a receipt via WhatsApp or email, Penny reads it using OCR, extracts the supplier name, date, amount, and VAT, and matches it to the corresponding bank transaction. If the match is confident, she attaches the receipt and categorises the transaction. If no matching transaction exists (perhaps the payment has not yet cleared), she holds the receipt and matches it when the transaction appears.

For you, this means receipts are already matched when you review the client's records. Your role shifts from matching receipts to verifying matches, which is dramatically faster.

2. VAT Treatment

VAT categorisation is one of the more time-consuming aspects of bookkeeping review because the rules are complex and the consequences of errors are financial. Standard-rated, reduced-rated, zero-rated, exempt, outside the scope — each transaction needs the correct treatment, and getting it wrong affects the VAT return.

Penny applies VAT treatment based on the supplier, the category, and the specific circumstances of each transaction. She understands the difference between zero-rated food and standard-rated catering, between standard-rated office furniture and zero-rated children's clothing. When she is uncertain (which happens most often with mixed-supply purchases or suppliers who sell both VATable and exempt goods), she flags the transaction with a specific VAT uncertainty flag.

This means your VAT review focuses on the genuinely ambiguous items rather than verifying the obvious ones. The standard-rated broadband payment does not need your attention. The mixed purchase from a supplier who sells both business equipment and personal items does.

3. Client Communication

A significant portion of review-related workload is not the review itself but the communication around it. Chasing missing receipts, asking clients to clarify unusual transactions, and explaining why certain expenses cannot be claimed — these conversations consume time.

Penny handles the initial layer of client communication via WhatsApp. When a receipt is missing, she prompts the client. When a transaction is unclear, she asks the client for context before the item reaches your queue. When a client asks whether something is deductible, Penny provides guidance for straightforward cases and escalates complex queries to you.

By the time an item appears in your review queue with a "requires further information" flag, Penny has already asked the client for clarification. If the client responded, the context is attached to the transaction. If they did not, the flag tells you that the automated prompt was unsuccessful and you need to follow up personally. For more on structuring your client communication approach, our dedicated guide covers both the automated and human elements. This division of labour reflects the evolving distinction between bookkeeping and accountancy — Penny handles the bookkeeping communication, and you handle the professional advisory communication.

4. Anomaly Detection

Beyond categorisation, Penny monitors transaction patterns for anomalies that warrant accountant attention. These include:

• Unusual amounts: a payment to a regular supplier that is significantly larger or smaller than the norm
• Frequency changes: a supplier that was paid monthly now being paid weekly, or vice versa
• Potential duplicates: two transactions of the same amount to the same payee within a short period
• Threshold proximity: transactions that bring the client close to a VAT registration threshold, a tax band boundary, or a CIS deduction threshold
• Pattern breaks: any change in the client's established transaction patterns that might indicate a change in circumstances

Without Penny, you would only spot these anomalies if you happened to notice them during a comprehensive review — which, given the volume of transactions, is hit and miss. With Penny, they are specifically identified and flagged, ensuring nothing significant slips through.

5. Workpaper Generation

The review process generates documentation automatically. Every transaction that Penny auto-applies is logged with the categorisation, confidence score, and the tier of reasoning used. Every item you review is logged with your decision, any amendments, and a timestamp. The result is a complete workpaper trail that documents the entire review process without you needing to prepare it manually.

This is not a minor saving. Workpaper preparation typically consumes 15-20% of total review time in practices that maintain proper documentation. Automating it eliminates that overhead entirely. Our post on automated workpaper generation covers the specifics of what is included and how the workpapers meet audit requirements.

How Penny Learns and Improves Over Time

The review workload does not stay static. It decreases over time as Penny learns from your review decisions. This learning happens at three levels.

Client-Specific Learning

When you review a flagged item and amend the categorisation, Penny learns that specific correction for that specific client. If you recategorise a transaction from "general expenses" to "marketing" for Client A, Penny applies that learning to similar transactions for Client A in the future. This is why flag rates decrease over time: the corrections you make today prevent the same items from being flagged tomorrow.

Practice-Wide Patterns

Penny also identifies patterns across your practice. If multiple clients in similar industries have the same type of transaction and you consistently categorise it the same way, Penny applies that learning to new clients in similar industries. This means a new client benefits from the accumulated learning of your existing client base, reducing the initial flag rate.

Platform-Wide Intelligence

At an aggregated, anonymised level, categorisation patterns across the entire Accounted platform contribute to the base model. If thousands of accountants consistently categorise "Amazon Web Services" payments as "hosting and servers," a new client's first AWS payment benefits from that consensus even before their own learning data exists.

The practical effect is that review workloads follow a characteristic curve: higher in the first quarter, declining in the second and third quarters, and reaching a steady state by the fourth quarter where only genuinely unusual transactions are flagged.

What Penny Does Not Do

It is important to be clear about the boundaries of Penny's capabilities, both for managing your own expectations and for explaining the system to clients.

Penny does not replace your professional judgement on complex tax matters. She can identify that a transaction looks unusual or that it might have tax implications, but she does not advise on tax planning, assess IR35 status, or determine whether a particular expense is allowable in ambiguous cases. Those are your domain.

Penny does not file returns on her own. She prepares the data and pre-populates returns, but submission requires accountant authorisation. This is a deliberate design choice: the professional sign-off remains with you.

Penny does not communicate with HMRC. All interactions with HMRC — whether querying a penalty, responding to an investigation, or seeking advance clearance — are human-to-human processes that require professional representation.

In short, Penny handles the volume work that consumes your time without requiring your expertise, and she surfaces the items that do require your expertise clearly and efficiently. She is a tool that makes you more effective, not a replacement that makes you unnecessary.

Getting Started

If you want to see the workload reduction first-hand, the fastest path is to connect a few clients and run through a review cycle. The difference between reviewing a client's records with and without Penny's pre-processing is immediately apparent.

You can set up your practice portal in approximately five minutes using our portal setup guide, and the review queue populates as soon as your first client's transactions flow through. Within a week, you will have enough data to assess the impact on your workflow.

For practices looking at the broader picture of how AI-assisted bookkeeping fits into their service offering, our post on why accountants should recommend AI bookkeeping covers the strategic perspective. And for a real-world example of the impact, the Accountancy Age coverage of AI in practice provides independent context on how the profession is adapting.

The maths is straightforward. If you can reduce review time by 70-80% per client, you can either grow your client base proportionally or reclaim the time for advisory work, business development, or simply a better work-life balance. Penny makes the maths work. The choice of what to do with the time saved is yours.

Visit our page for accountants to learn more about how Accounted supports practices like yours.

Accounted gives accountants a free practice portal — manage all your clients, file to HMRC, and let Penny handle the routine work. See the accountant portal →