How to Automate CaseWare IDEA

Testing every transaction beats testing forty of them

IDEA is the data analysis tool auditors reach for when a sample will not do. Caseware makes it, and the company writes its own name as one word now, so the firm that bought CaseWare IDEA years ago is running Caseware IDEA today.

The premise is simple and it is the whole reason the tool exists. Rather than pulling forty invoices out of two hundred thousand and hoping they represent the rest, you test all two hundred thousand. Every journal entry posted after midnight. Every payment to a vendor whose bank details changed the week before. Duplicate invoice numbers, round-sum expenses, gaps in a cheque sequence, the same address on two unrelated suppliers.

External auditors use it, internal audit teams use it, and so do corporations and government bodies testing their own books. In forensic work it is the thing that turns a drive full of exports into an exhibit somebody can defend.

The analysis is a morning; getting the data took a month

A forensic engagement opens with a list of what you need and no way to make anyone produce it.

The bank has been served and has not responded. The client's IT will export the general ledger once a ticket clears. Opposing counsel put eleven thousand pages on a shared site at six on a Friday. A custodian's laptop image is still with the vendor. Somebody has to know which of those has actually landed, and the only way to know is to open each one and look.

Underneath runs a calendar that does not care. A discovery deadline in eight days. An expert report due before it. A deposition to prepare, with exhibits that have to tie back to the schedules. An engagement letter that was never returned signed, on a matter with three hundred hours already in it.

Every one of those is a status held on somebody else's system, and a senior associate is effectively paid to keep asking.

CaseWare IDEA

A production log written as things arrive What landed on the shared site overnight, checked against what was actually requested, with the difference named rather than discovered later.

Deadlines that move without telling you A docket entry that shifts a discovery date is the one change nobody hears about in time. It can be read the morning it appears.

Nothing leaves without the expert Statuses are gathered and lists are drafted. Anything going to a court, to opposing counsel or to a client waits for the person who signs the report.

IDEA tests the population it has been handed

Caseware built IDEA to take the drudgery out of the analysis, and it does exactly that.

It imports from Excel, CSV, PDFs, more than fifty accounting systems and Working Papers, and it does not alter the source file on the way in, which is the property that lets a finding survive being questioned. SmartAnalyzer and custom routines run a firm's standard tests across the entire population rather than a sample, so exceptions surface instead of being hunted. Scripts can be scheduled to run overnight against refreshed data, with a success or failure alert when they finish. A run history records each step in order, which is what makes review and sign-off possible months later when nobody remembers the sequence.

That is a serious answer to the analysis problem.

All of it starts, though, from a file that is already in IDEA. The bank that has not responded, the production that appeared on a shared site on Friday night, the docket entry that quietly moved a date, the ledger export still behind a ticket: none of those are data yet. They are statuses on other people's websites, and each one gets checked by a person or not at all.

The import queue can fill itself overnight

Data reaches IDEA because somebody went and fetched it, and the fetching is the half that can be handed over.

The shared site checked each morning for whatever landed overnight, with what arrived logged against what was requested and the gap named. Subpoena and third-party responses followed on the portals that hold them, so a response gets worked the week it lands rather than the week somebody happens to look. The client's accounting system watched for the export that was promised, and the ticket behind it followed until it closes.

Then the calendar wrapped around the analysis. Court dockets read for the entry that moved a deadline. Deposition and report dates counted down with the schedules they depend on listed beside them. Chain of custody entries logged as each item is received, so the log gets written when the thing arrives instead of reconstructed the night before a hearing.

None of that is analysis. It is the queue of things that have to be true before analysis can start.

The expert takes the stand, not the spreadsheet

Between the subpoena and the schedule sits a fortnight of asking people for files. Asking is not expert work, and the expert is who spends the fortnight doing it.

WebRun is an agent that works a real Chrome browser, signed in as you. It opens the production site, the bank's portal, the docket and the client's system, reads what has landed and what has not, and hands the analyst one page of what is now ready to import.

It gathers unattended, because reading a status changes nothing. Anything that reaches a court, opposing counsel or a client waits for the person who will be cross-examined on it, and a run can be watched live and stopped.

The workflows below are already built, and each one names what it opens.