How to connect Dext to Claude: what accounting firms are building

No native integration exists between Dext and Claude. Dext has its own OCR engine that extracts data from receipts, invoices, and bills, and its own categorisation suggestions. Formulaic builds the pipeline that takes Dext’s extracted data further: Claude validates categorisation accuracy, flags unusual expenses, generates narrative reports, and cross-references receipt data against the client’s actual chart of accounts in Xero or Sage. This is for accounting practices that use Dext across their client base and want AI-powered quality control and reporting on top of Dext’s native extraction.

Short answer: No native integration exists between Dext and Claude. Formulaic builds custom pipelines that take Dext’s extracted receipt data and route it through Claude for categorisation validation, unusual expense detection, and narrative expense report generation.

What this integration does

Dext solves the data entry problem: photographs of receipts become structured data fields (date, amount, supplier, description, suggested category). That’s valuable, but it’s only the first step. The accounting practice still needs to verify categorisation, spot unusual items, and produce reporting from the expense data.

Categorisation accuracy validation. Dext suggests a category based on supplier name and transaction patterns. But “Amazon” could be office supplies, stock, or personal expenses. “Uber” could be client entertaining, staff travel, or a disallowable benefit. Claude reads the receipt description, cross-references against the client’s chart of accounts and historical coding in Xero or Sage, and flags items where Dext’s suggested category doesn’t match the pattern. The bookkeeper reviews flagged items rather than checking every transaction.

Unusual expense flagging. Beyond miscategorisation, Claude identifies expenses that warrant attention: amounts significantly higher than the client’s average for that category, transactions from new suppliers the client hasn’t used before, expenses submitted outside normal patterns (weekend transactions for a Monday-to-Friday business), and items that may represent disallowable expenditure for tax purposes. These flags prevent problems from reaching the accounts.

Narrative expense reports. For clients who need regular expense reporting (particularly those with expense policies or board reporting requirements), the pipeline generates narrative reports from Dext data: total spend by category for the period, comparison to the prior period and budget, top suppliers by spend, flagged items requiring review, and a summary paragraph suitable for board or management consumption.

Cross-referencing against accounting data. The pipeline connects Dext’s receipt data with the client’s Xero or Sage records. This catches: receipts that don’t match corresponding bank transactions (different amounts, missing entries), duplicate receipts and transactions, expenses posted to Dext but missing from the accounting system, and bank transactions without supporting receipts.

How the data flows

Dext’s data is already structured, which makes this pipeline more straightforward than raw accounting integrations.

Extraction from Dext. Formulaic connects to Dext’s API to pull extracted receipt data: date, amount, currency, supplier name, description, suggested category, VAT amount, and the client reference. For practices using Dext Prepare, the data includes the approval workflow status.

Enrichment from Xero/Sage. The pipeline pulls the client’s chart of accounts, recent transaction history, and category patterns from Xero or Sage. This provides the context Claude needs to validate Dext’s categorisation suggestions against actual coding practice.

Processing. Claude receives: the Dext receipt data, the client’s chart of accounts and historical coding patterns, category validation rules (your practice’s specific coding policies), and any expense policy rules (spending limits, allowable categories, approval thresholds). Claude returns: confirmed categories, flagged items with reasons, and narrative report sections.

Delivery. Validated categories feed back to Dext (for approval workflows) or directly to the bookkeeper’s review queue. Flagged items appear as an exception list. Narrative reports export as formatted documents or emails. Cross-reference discrepancies generate task items for investigation.

Processing logs record every receipt analysed, every flag raised, and every report generated. This supports the practice’s quality assurance documentation.

Use cases we build

Multi-client Dext quality control. A practice processes Dext receipts for 60 clients. Each client submits 30-100 items monthly. The pipeline runs daily, scanning new Dext submissions across all clients. It validates categories against each client’s specific chart of accounts in Xero, flags items outside normal patterns, and generates a consolidated exception report for the bookkeeping team. One practice reduced categorisation errors reaching the accounts by 70%, catching an average of 8 corrections per client per month at the Dext stage rather than during reconciliation.

Expense policy enforcement. A professional services firm has expense policies: maximum per-head for client entertaining, approved hotel rate bands, approved airlines for travel. The pipeline checks every Dext submission against these policy rules before the expense enters the accounting system. Policy violations are flagged for manager approval rather than processed automatically. The accounts team stopped manually checking every expense claim against the policy PDF.

Monthly expense analysis for advisory clients. For clients receiving advisory services, the pipeline generates monthly expense analysis from Dext data: spend by category as a percentage of revenue, trend lines over 6 months, comparison to industry benchmarks (where provided), and specific recommendations (e.g., “Travel spend increased 40% quarter-on-quarter — worth reviewing the travel policy”). These reports position the accountant as an advisor, not just a bookkeeper.

VAT reclaim optimisation. The pipeline analyses Dext receipt data for VAT recovery opportunities: fuel receipts that qualify for VAT reclaim, business entertainment where input VAT is being incorrectly reclaimed, and mixed-use items where partial reclaim applies. Claude identifies receipts where the VAT treatment may be incorrect and generates a summary for the practice’s VAT specialist to review.

What about Dext’s own AI improvements?

Dext continuously improves its OCR and categorisation engine. Their extraction accuracy for standard receipts is high and getting better. The Formulaic build doesn’t replace Dext’s core function. It adds a quality control and reporting layer on top.

Dext extracts data from images. Claude analyses that data in context. These are different tasks. Dext looks at a receipt and reads “Tesco, £47.32, groceries.” Claude looks at that data, checks the client’s chart of accounts, sees the client is a limited company, and flags that a grocery purchase needs reviewing because it may be a disallowable personal expense.

As Dext’s categorisation improves, fewer items will need correction. But the analysis layer (policy checking, anomaly detection, narrative reporting) remains valuable regardless of how good the categorisation becomes.

Timeline and investment

Every build is scoped and priced based on complexity. Simpler data syncs take less time than multi-system orchestration with compliance requirements. Start with an audit to get a clear proposal.

Take the AI Opportunity Scorecard to see which integrations would deliver the highest ROI for your firm.