Caseware Extractly

Caseware Extractly does a narrow job that every audit team recognises: taking a sample, finding the invoice or bank statement behind each item, keying the figures and ticking them off. It is repetitive, error-prone work that junior staff dislike, and it is exactly where OCR plus matching earns its keep.

What it does well. Living inside Excel is the key design choice, because auditors do not have to learn a new environment or move schedules elsewhere. Extraction handles both fields and tables, saved layouts help with recurring document types, and automated vouching matches each sample line to a document and surfaces variances. Every value links back to its source, and the export bundles comments and exceptions into one linked PDF, which reviewers and file inspectors will appreciate. Backing from Caseware, a long-established audit software house with a strong UK presence, makes it a safer bet than a standalone startup.

Where it falls short. Pricing is not published, so firms must go through Caseware sales or a reseller. OCR on poor scans and handwriting is never perfect, so review time does not disappear, and the time savings quoted are Caseware's own. The product page says little about Extractly-specific security or how the cloud storage ties into Caseware Cloud, so firms should ask where documents are held and for how long. It is a vouching tool, not an analytics or risk engine.

Best for. Audit teams of any size, in the UK or elsewhere, that still hand-key vouching schedules, bank tie-outs and invoice testing in Excel. A fair trial is to take one completed engagement's vouching file, rerun it through Extractly and time both extraction and review, paying attention to scanned and handwritten documents. DataSnipper is the obvious rival and is the more mature Excel-based option with a broader feature set, so firms should trial both side by side before buying.

Verdict. One of the more practical uses of AI in audit today: focused, inside the tool auditors already use and backed by a serious vendor. Worth testing against DataSnipper on real files.