The audit tasks worth automating first share four properties: they are high volume, rule-based with a single correct answer, performed the same way on every engagement, and currently done by hand on a sample because doing them completely is too expensive. Matching, reconciliation, recomputation, duplicate detection and document chasing all qualify. Risk assessment, materiality, estimates and conclusions do not, because they have no single correct answer and the value is in the judgement rather than the throughput.

The four properties

What makes a task a good automation candidate
Property	Why it matters	Counter-example
High volume	Return scales with the number of items. Automating something done twice a year saves nothing.	Going concern assessment: once per engagement, and the work is the thinking.
Deterministic	There is one right answer, checkable by anyone. Correctness does not depend on who ran it.	Whether an exception is material: depends on context, and two competent auditors may differ legitimately.
Identical across engagements	The rule written once applies everywhere, so the investment is amortised across the client base.	Related-party identification: depends on the entity's specific structure and relationships.
Currently sampled because complete coverage is unaffordable	Automation changes the coverage as well as the cost, which is a quality gain and not only an efficiency one.	Bank reconciliation on a single account: already done completely because it is small.

What makes a task a good automation candidate

The fourth property is the one most often overlooked and is the most valuable. A task automated that was already being done completely saves hours. A task automated that was being sampled changes what the audit covers, and that is a different kind of benefit.

Common tasks, ranked by return

Audit tasks scored on the four properties
Task	Volume	Deterministic	Same everywhere	Currently sampled	Verdict
Matching the purchase register to GSTR-2B	Very high	Yes	Yes	Usually sampled or done crudely	Automate first. Recurring, monthly, and the manual version is genuinely painful.
Three-way matching of purchases	Very high	Yes	Largely	Always sampled	Automate early. The coverage gain is the real benefit.
Duplicate invoice and payment detection	High	Yes	Yes	Rarely done at all	Automate early. Sampling cannot find duplicates reliably, so this is close to pure gain.
Cut-off date comparisons	Moderate	Yes	Yes	Always sampled	Automate. The rule is simple; the judgement about delivery terms is not and stays with the auditor.
Document collection and chasing	High	Partly	Yes	Not applicable	Automate the coordination. This is the largest source of elapsed time in most engagements.
Extracting fields from invoices and statements	Very high	No, it is a reading task with errors	Yes	Done by hand	Automate with review. Extraction is useful only if every extracted figure remains linked to the page it came from.
Vendor and bank reconciliation	High	Yes	Yes	Sampled	Automate the matching; the categorisation of differences is partly judgement.
Working paper assembly	High	Partly	Yes	Not applicable	Automate the structure and the evidence linking. Never the conclusion.
Risk assessment	Low	No	No	Not applicable	Do not automate. A generated risk assessment is last year's risk assessment with a new name on it.
Materiality determination	Low	The arithmetic yes, the choice no	No	Not applicable	Automate the calculation once the benchmark is chosen. The choice is the work.
Writing conclusions	Moderate	No	No	Not applicable	Do not automate. The conclusion is what the signature is attached to.

Audit tasks scored on the four properties

Sequencing for a firm of 5 to 50 staff

01Start with the recurring compliance work, not the auditInput credit reconciliation happens monthly across the whole client base. It is the highest-volume, most deterministic, most identical task a practice has, and improving it produces a return every month rather than once a year. It is also the safest place to learn what automation does to a team's habits.
02Then the transaction matching in the audit itselfPurchase and payables testing is where the audit hours concentrate. Full-population matching changes both the cost and the coverage.
03Then document collectionThis is a coordination problem rather than a computation one, and it is worth treating separately. The gain is elapsed time, which does not show up in chargeable hours but determines whether the season is survivable.
04Then evidence linking and working paper assemblyOnce testing is automated, the papers can be assembled from it rather than written afterwards. This is where documentation quality improves as a side effect.
05Stop before judgementThere is a natural boundary and it is worth stopping at it deliberately rather than drifting across it because a tool offers to help.

Automating a process that is not written down automates whatever the last person happened to do. Before automating anything, write down how it is currently performed. Firms routinely discover at this point that three teams do the same reconciliation three different ways, which is itself worth knowing.

What changes in the file, and what has to be checked

Automation changes the audit evidence, and the documentation has to keep up with it. Four things need explicit attention.

The population becomes the whole population, so the sample basis in the working paper has to change. Carrying forward a sampling rationale on a procedure that now tests everything is a review point waiting to happen.
The procedure description has to describe what the software did, specifically enough that another auditor could understand and evaluate it. “System-matched” is not a procedure description.
The reliability of the tool itself becomes relevant to the evidence. If a routine produces the results, the auditor needs some basis for believing the routine works, which usually means testing it on data with a known answer.
The exceptions produced still need an auditor's disposition. A tool that raises 57 potential exceptions has not found 57 findings, and the file must not read as though it has.

Regulatory requirement

Using software does not change the auditor's responsibilities. The requirements in SA 230 for documentation, in SA 500 for sufficient appropriate evidence, and in SA 220 and SQC 1 for direction, supervision and review apply identically whether a procedure was performed by hand or by a tool. Automation changes how the requirement is met, not whether it applies.

What not to expect

Two expectations lead firms to disappointment, and both are worth setting straight before any purchase.

Headcount does not fall. What changes is what the existing team spends its time on, and how many engagements they can carry in the peak. A firm that buys automation to reduce staff usually ends up with the same staff and worse morale.
The first season is slower. Data has to be shaped, rules configured, and the team has to learn where to stop trusting the output. The return arrives in the second year, and a firm that expects it in the first will abandon the effort halfway.

How we think about this

The line Lekhya draws is between work with one right answer and work that requires a view. Matching, recomputation and reconciliation are the first kind and should be computed by rule-based software, not generated. Reading a document to extract a field is a model's job, and every extracted figure stays linked to the page it came from so it can be checked. Deciding what any of it means is the auditor's, and the product is built so that it cannot quietly become anything else.

This is practical guidance written for audit teams, not an authoritative statement of any standard or statute. Standards, forms and timelines change. Check the current text issued by the ICAI and the relevant authority before relying on anything here in an audit file.

Free tools, no signup

Each one reads a file entirely in your browser — or, for two of them, just answers a few questions — and nothing is ever uploaded.

Audit automation for CA firms: what to automate first