AI Audit Evidence Gathering for Accounting Firms_

Audit teams use AI to automate the retrieval of evidence from client accounting systems, classify each document by audit area and financial statement assertion, select samples based on the audit plan’s methodology, and cross-reference supporting documents to workpaper requirements. The evidence gathering phase, which typically consumes the first 2-4 days of fieldwork on every engagement, compresses to 4-8 hours while producing better-organised, more complete evidence files.

The evidence gathering bottleneck

Every statutory audit begins with evidence gathering. Before the team can perform substantive testing, analytical procedures, or draw conclusions, they need the documents. For a standard SME audit, the evidence requirement spans every area of the financial statements:

Revenue: sales invoices, contracts, delivery notes, credit notes, and bank receipts for a sample of transactions. For cut-off testing, transactions around the year end. For completeness, dispatch records or delivery confirmations not matched to invoices.

Purchases: purchase invoices, purchase orders, goods received notes, and payment evidence for a sample of transactions. For accruals testing, invoices received post year-end relating to pre year-end expenditure.

Payroll: payroll summaries, employment contracts for starters and leavers, P11D returns, pension contribution schedules, and HMRC submissions.

Fixed assets: fixed asset register, purchase invoices for additions, disposal proceeds evidence, depreciation calculations, and impairment assessments.

Cash and bank: year-end bank statements, bank reconciliations, bank confirmation letters, and supporting documentation for reconciling items.

Debtors and creditors: aged debtor and creditor listings, post year-end receipts and payments, supplier statement reconciliations, and specific evidence for significant balances.

For a typical SME audit, this represents 200-500 individual documents. The traditional process involves sending the client a prepared-by-client (PBC) request list, waiting for responses, chasing missing items, and then organising the received documents into the audit file. This process takes 2-4 days of team time per engagement, often longer when client responses are slow or incomplete.

The delay has knock-on effects. If evidence gathering takes a week, the audit team can’t start substantive work during that week. Fieldwork extends, budgets overrun, and the team is pulled to other engagements before completing the work.

How AI evidence gathering works

Direct client system connection

The system connects to the client’s accounting platform (Xero, Sage, QuickBooks, IRIS) via API and extracts data and documents directly. Rather than sending a PBC list and waiting for the client to gather documents, the system pulls what it needs from the source.

Bank statements are downloaded from connected bank feeds. Sales and purchase invoices are extracted from the accounting system’s document store. Payroll summaries are pulled from the payroll module. Trial balance data, journal entries, and management accounts are extracted at the required level of detail.

For documents not stored in the accounting system (board minutes, contracts, loan agreements), the system generates a targeted request list covering only the items it can’t retrieve automatically. This request list is shorter and more specific than a traditional PBC list, leading to faster client response.

Classification by audit area

Each retrieved document is classified by audit area using the document type, accounting code, and transaction characteristics. A sales invoice for a revenue transaction is classified to the revenue audit area. A fixed asset purchase invoice is classified to the fixed assets area. A bank statement is classified to the cash and bank area.

Within each audit area, documents are mapped to the relevant financial statement assertion:

• Existence: evidence that assets and liabilities reported actually exist (bank confirmations, physical inventory counts, debtor confirmations)
• Completeness: evidence that all transactions and balances are recorded (cut-off testing documents, unmatched delivery notes)
• Accuracy: evidence that transactions are recorded at the correct amounts (invoices matching journal entries, calculation checks)
• Valuation: evidence supporting the valuation of assets and liabilities (impairment assessments, provision calculations)
• Rights and obligations: evidence of ownership or liability (title deeds, loan agreements, lease contracts)

The audit team receives organised evidence folders rather than an unsorted document collection. Each folder contains the documents relevant to that audit area with the assertion mapping visible.

Sample selection

The system applies your sampling methodology to the relevant population for each audit area. For revenue testing, it identifies the full population of revenue transactions and selects a sample based on the audit plan’s criteria: statistical sampling with a specified confidence level, or non-statistical sampling targeting high-value items, unusual transactions, and a random selection from the remaining population.

For each selected item, the system retrieves the supporting evidence chain: the sales invoice, the corresponding bank receipt, the delivery note or proof of service, and the accounting system entry. The auditor receives a complete evidence package for each sample item rather than having to trace each item manually.

Sample selections are documented with the methodology applied, the population size, the sample size rationale, and the selection criteria. This documentation supports ISA 530 requirements for audit sampling.

Cross-referencing and gap identification

The system cross-references retrieved evidence against workpaper requirements. For each workpaper in the audit file, it identifies the evidence needed and checks whether that evidence has been retrieved. Gaps are highlighted: “Revenue workpaper R3 (cut-off testing) requires invoices for 15 transactions around the year end. 12 invoices have been retrieved. 3 are pending from the client.”

This real-time completeness tracking means the audit team knows exactly where gaps exist before they start testing, rather than discovering missing evidence mid-way through a workpaper.

Evidence for analytical procedures

For analytical procedures, the system prepares comparative data: current year figures alongside prior year, budget, and industry benchmarks. Monthly breakdowns for revenue and cost categories enable the auditor to identify unusual patterns (a spike in December revenue, a drop in Q3 costs) with the data ready for investigation.

Integration with the audit workflow

Evidence is organised into the structure expected by your audit software. CaseWare, Inflo, and Pentana integrations file evidence directly into the correct workpaper sections. For firms using other platforms, evidence is exported in an organised folder structure that mirrors the audit file.

The audit team accesses evidence through their normal workflow. The difference is that the evidence is already there, organised, and linked to the relevant workpaper, rather than arriving in batches over several days and requiring manual filing.

Quality and documentation

Every retrieved document is logged with:

• Source system and retrieval method
• Timestamp of retrieval
• Audit area classification and assertion mapping
• The workpaper it supports

This audit trail satisfies ISA 230 documentation requirements. An engagement quality control reviewer (or an external inspector) can trace any audit conclusion back through the workpaper to the evidence to the source system.

For recurring audits, the system retains the prior year evidence file structure. Year-on-year changes in the client’s systems or document formats are flagged so the team can assess whether the evidence gathering approach needs updating.

Capacity impact per engagement

For a standard SME audit:

• Traditional evidence gathering: 2-4 days of team time
• With AI automation: 4-8 hours (primarily reviewing non-automated items and client-provided documents)
• Time recovered per engagement: 1-3 days

For a practice conducting 100 audits per year, that is 100-300 days of recovered capacity, enough to handle 20-40 additional engagements without hiring, or to reallocate the time to higher-value advisory work.

The system connects to Xero, Sage, QuickBooks, IRIS, and specialist audit platforms including CaseWare, Inflo, and Pentana. Data stays on UK-hosted infrastructure with encryption and access controls meeting ICAEW, ACCA, and ICAS requirements.

Typical timeline: 6-8 weeks. Typical investment: £18-30k / $23-40k.