Caseware’s Canadian AI hub presents its capability as Caseware Verity — agents that work inside the engagement file across prepare, plan, evaluate and report. Check availability status before planning a busy season around any of them.
Key takeaways
If AiDA is the name you know, look at what Caseware documents today before assuming a feature set. The company’s Canadian AI hub presents its AI capability as Caseware Verity, a set of purpose-built agents that execute time-intensive work across the engagement lifecycle, organised around four phases and sitting inside the engagement workflow rather than beside it. Product names in this category change quickly, and a two-year-old blog post about an assistant is a poor basis for a purchase decision. The vendor’s own current page is the only reliable description.
Disclosure Checklist Agent. Caseware describes it as automating disclosure checklist review against financial statements, generating citation-backed suggestions auditors can review, refine, accept or override directly within the engagement workflow, surfacing evidence-backed yes, no and not-applicable suggestions that reviewers can validate from the citations.
Risk Suggestion Agent. Listed as coming soon, described as surfacing engagement-specific risk suggestions by synthesising financial data, prior-year risks, controls, board minutes and supporting engagement context.
Document Intelligence Agent. Also listed as coming soon, described as automating review and extraction of source documents directly into workpapers to reduce manual evidence preparation.
Alongside these the Canadian page lists related products — Extractly for automated data extraction and matching in Excel, Validate for financial statement validation and review, ReviewComp for review and compilation engagements, and Sherlock for practice and engagement intelligence. When you are comparing options, compare against that list, and check availability status: "coming soon" is not a feature you can plan a busy season around.
The word is doing real work here, and it is worth being precise. A chatbot answers a question in a window. An agent inside an engagement file reads structured content from that file, performs a defined task and writes a proposed result back into the file, where the normal review and sign-off chain picks it up.
That architectural difference is why an embedded tool is more defensible than a general-purpose assistant for assurance work. Output can be tied to a workpaper, a citation can point at the document that produced it, and the firm has something to show when the review is reviewed.
Buying the tool is the easy part. The Canadian Public Accountability Board treats technology in the audit as an inspection focus and says its inspections include evaluating the system of quality management practices associated with obtaining or developing, approving, implementing and maintaining automated tools and techniques at both the firm level and the engagement level. CPAB has published The use of artificial intelligence in the audit — balancing innovation and risk and The evolving use of technology in auditing, and leads IFIAR’s Technology Task Force.
In practice that means a firm needs four things written down: who approved the tool and on what basis; how staff are trained to review its output; what the tool is not used for; and how a version change is assessed. None of that comes in the box.
The engagement standards themselves do not move. The Canadian Auditing Standards and quality management standards issued by the Auditing and Assurance Standards Board remain the authority for what the work has to be, whoever or whatever prepared the first version of it.
A firm with nine assurance staff runs a two-engagement pilot on disclosure checking only. It records three numbers per engagement: how many checklist items the agent proposed, how many the reviewer changed, and how many changes were substantive rather than wording.
The decision rule is set before the pilot starts, not after. If substantive changes exceed the firm’s threshold, the tool is used as a second pass after a manual review rather than as the first pass. If they fall below it, the manual first pass is retired for that engagement type only.
Both outcomes are useful. What is not useful is a pilot with no threshold, because the answer then depends on who is asked.
Which agents are generally available today in Canada, and which are announced? Where is engagement data processed and retained, and is it used to train models beyond our own engagements? Can every suggestion be traced to the source document? What is logged, and can we export that log for an inspection? How are model changes released, and can we decline one mid-engagement? Data residency and retention deserve particular attention, because an assurance file contains personal information about your client’s people — the Privacy Commissioner’s guidance on privacy and AI is the reference point for that conversation.
Is this only for large firms?
The pricing question is one for the vendor, but the governance question is not size-dependent. A smaller firm needs the same four documents; it just needs shorter ones.
Does an embedded agent replace a reviewer?
No, and the vendor’s own language does not claim it does — the design assumes a reviewer accepts, refines or overrides each suggestion.
What if we already use a different engagement platform?
The evaluation questions above are platform-neutral. Ask them of whoever you buy from, and hold the answers in writing.
A 30-minute call is enough to tell you whether AI pays for itself here.
