Running an agency

The Australian recruitment agency systems checklist

Most system evaluations compare feature lists, which is why most system changes disappoint. These are the questions that predict whether a product will still be the right one in three years.

The Recruited team Reviewed 4 September 2026 9 min read Running an agency

Tenancy and data boundaries

Ask where the boundary between your data and another customer’s is enforced. "In the application" is a weaker answer than "on the server, in one place, with a test". Ask what happens when a user account exists but has no workspace assigned. The correct answer is that it sees nothing, and a surprising number of systems answer that it sees everything.

Ask whether the vendor operates a recruitment business of its own, and if so, what technical separation exists. This is not a rude question. It is the single most important one if the answer is yes.

Ask whether there is any pooled candidate database across customers, in any form, for any purpose, including "matching" and "market insights". A pool that exists for analytics is still a pool.

Portability, before you need it

Ask to see the export. Not to be told it exists: to see what it produces, on a plan you would buy, for a workspace with real volume. The gap between "we support export" and "here is a complete, structured file" is where agencies get trapped.

Ask which categories the export covers: candidates, clients, contacts, opportunities, candidacies, submissions, placements, notes, tags, owners, communications and attachments. Ask specifically about attachments and full email bodies, because those are the two most often missing.

Ask whether export is available on every plan or only on the expensive one. Portability behind a paywall is not portability.

Search that works on your data

Ask how skills and occupations are classified. Free-form tagging as the primary method degrades within eighteen months into six spellings of the same word, and no amount of AI on top recovers what was never structured.

For any AI search, ask to see the filters it produced before it ran. If you cannot inspect and edit the interpretation, you cannot defend the shortlist to a client, and you will not trust it by the third search.

Ask what happens to a candidate with a missing field. Being silently ranked lower is worse than being shown with the gap named, because you cannot act on what you cannot see.

Pipeline discipline

Ask whether every open candidacy is required to have an owner and a next action, and what the system does when one has neither. "Nothing" is the common answer and the expensive one.

Ask whether a stage transition can require a reason. Rejection and withdrawal are the two you will be asked about months later and the two nobody remembers.

Ask whether stage history is append-only or a field that gets overwritten. The second makes disputes unanswerable.

Ask what the board can do with a keyboard alone. A pipeline only operable by mouse excludes people from their own job.

Communications

Ask what a mailbox connection captures. "Everything, and we sort it out" is a different product from "what relates to a record you hold", and only one of them is a reasonable thing to ask your team to agree to.

Ask whether an individual can exclude addresses, domains and threads, and whether an administrator can connect a mailbox on someone’s behalf. The second should be no.

For AI summaries, ask what each line cites. A summary you cannot check is a summary you will eventually be embarrassed by in front of a client.

Advertising and applications

Ask which job boards are connected today, with a signed agreement. Then ask what happens to a publish that fails: a silent failure is how a role sits unadvertised for a week.

Ask whether applications arrive as records with a source attached, whether they are deduplicated against existing candidates, and what happens to a CV the parser cannot read. Quarantined is right; discarded is not.

Ask whether closed roles leave the sitemap immediately. A live page for a filled role is a candidate experience problem and a search problem at once.

Commercials and the exit

Ask what is metered and what is merely counted. Storage of candidate rows is rarely a real marginal cost; crawling, AI extraction, document parsing and message delivery are. A vendor that meters the first is pricing on leverage rather than cost.

Ask what happens when you hit a limit mid-placement. Being blocked during an offer, a privacy request or an unsubscribe is not an acceptable answer at any price.

Ask what the notice period is, what happens to your data at the end of it, and how long you have to export. Then ask for that in the contract rather than in an email.