Recruitment agency software in Australia: JobAdder, Bullhorn and AI shortlisting

Should an agency replace its ATS or build around it?

Build around it. An agency’s ATS and CRM holds candidates, contacts, jobs, placements and the job board feeds, and replacing that is rarely worth the cost or the disruption. The useful custom work is in what the ATS does badly for your particular desks: shortlisting at volume, onboarding temps, matching timesheets to payroll and reporting margin.

Your situation	Usually the right move
Permanent desks, under 20 consultants, standard workflow	Configure JobAdder, Bullhorn, Vincere or Loxo well; use the marketplace integrations
Heavy temp or contract volume	Keep the ATS, add a temp onboarding and timesheet app connected to payroll
High application volume per role	Add an AI shortlisting assistant over ATS data, with an audit log
Owners can’t see margin by client, desk or consultant	Add a reporting layer that joins ATS placements with payroll and invoicing
You are building a recruitment product to sell	Build, and expect to become the vendor with everything that implies

Check API access before any of this. JobAdder’s partner guidance describes a single developer account per technology partner, OAuth2 authorisation by a client admin, and webhooks and partner action buttons that must be removed when an integration is disconnected. Bullhorn publishes developer and API documentation for its REST API. If the endpoints you need are missing or rate limited, that changes the design, so confirm them in the Discovery stage. Our API development service and the API integration cost guide cover how these projects are priced.

Where does AI shortlisting go wrong?

AI goes wrong in recruitment when it decides instead of assists, or when nobody can explain why a candidate scored low. The technology is good at reading 400 CVs against a brief, pulling out years of experience, licences and location, and producing a ranked list with reasons. It is bad at knowing that a career gap was parental leave, or that a candidate’s unusual title means the same thing as yours.

Anti discrimination law applies to the outcome whether a person or a model produced it. Commonwealth laws cover discrimination on grounds such as sex, race, disability and age, and each state has its own Act. A model trained on your past placements can learn the patterns of your past choices, including the ones you would not defend.

The Australian Human Rights Commission, with LexisNexis, publishes an AI and recruitment compliance checklist built around four areas: privacy and data protection, digital robustness and safety, fairness and transparency, and accountability and sustainability. Its framing is that humans stay in control of decisions and candidates can challenge unfair outcomes. In software terms, that becomes a short design brief:

1. Score against the brief, not against history. Criteria come from the job’s stated requirements, written by the consultant, not inferred from who you placed before.
2. Strip what shouldn’t count. Names, photos, dates of birth and addresses beyond the commute radius are hidden from the scoring step.
3. Show the reasons. Every score comes with the evidence quoted from the CV, so the consultant can check it in seconds.
4. A person decides every rejection. The assistant suggests; a named consultant moves candidates forward or out.
5. Log everything. Keep the brief, the model version, each score, its reasons and the consultant’s decision, so you can answer a complaint months later.
6. Test before go live. Run the assistant over past applications and look for unequal outcomes between groups before it touches a live role.

Our AI agent development work follows this pattern, and our guide to AI hallucinations explains why quoted evidence matters.

What does 10 December 2026 change for recruiters?

From 10 December 2026, Australian Privacy Principles 1.7 to 1.9 require an APP entity’s privacy policy to describe automated decisions that could reasonably be expected to significantly affect an individual’s rights or interests. The duty covers decisions a computer program makes, and decisions where it substantially and directly contributes. The policy must describe the kinds of personal information used and the kinds of decisions made.

For an agency, a knockout question that silently rejects applicants is the clearest case to assess. A ranked list a consultant reviews is less clear, which is exactly why you want the record of who decided. Many agencies will be APP entities because they exceed the small business turnover threshold or handle certain kinds of information; confirm your position with your adviser.

A practical inventory looks like this:

Question for each tool in your stack	Why it matters
Does it reject, rank, or filter candidates without a person looking?	That is the decision most likely to be in scope
What personal information feeds it?	The policy must describe the kinds used
Who owns the configuration, and when did it last change?	Your description must stay accurate as rules change
Can you show a candidate how a decision was reached?	Complaints and access requests will ask

Our Privacy Act automated decision making guide covers the rule in detail.

How should right to work checks fit onboarding?

Make the work rights check a step the system won’t let anyone skip, with the evidence stored against the placement. Home Affairs says registered organisations can use VEVO, through an ImmiAccount, to see the visa conditions of visa holders who have given them permission. VEVO can’t confirm Australian citizenship, so citizens need a different document, such as a passport or citizenship certificate.

A good onboarding flow for temps and contractors captures, in order: identity, work rights with the permission recorded, the VEVO result and its date, visa expiry and conditions such as work limits, licences and tickets, super and tax details, and the induction for the host site. The system then reminds the team before a visa or licence expires and stops the worker being booked if one lapses.

What about temps, timesheets and payroll?

For temp desks the money is made or lost between the timesheet and the pay run. Payroll tools built for recruitment and labour hire, such as Astute Payroll, let you configure award and enterprise agreement rules with timesheets and invoicing, and general platforms such as Employment Hero handle pay for direct staff. The custom work is usually the gap: a mobile timesheet that the host supervisor approves on the spot, rates that follow the right award or agreement for that client and role, and an exceptions queue for anything unusual.

If you supply workers into larger hosts, check whether a Fair Work Commission regulated labour hire arrangement order covers any of them. Those orders can require labour hire employees to be paid at least the host’s enterprise agreement rate. Our labour hire software page covers licensing and Same Job Same Pay in detail.

How clean is your talent pool?

Most agency databases hold years of candidates who never agreed to stay that long. APP 11 requires reasonable steps to destroy or de identify personal information an entity no longer needs, and stale data is also breach exposure. The RCSA Code of Professional Conduct sets professional standards for its members on top of the law.

A hygiene job usually means three things: retention rules by record type, a consent refresh that asks dormant candidates whether they want to stay, and an automated purge or de identification of everyone else. Doing this before you point an AI assistant at the database also improves its results.

Which reports do agency owners actually want?

Margin by client, by desk and by consultant, and the pipeline that will produce next quarter’s margin. ATS dashboards show activity; payroll shows cost; invoicing shows revenue. Owners want them joined:

• Gross margin per placement and per hour for temps, by client and consultant.
• Time to fill and submittal to interview ratio by desk.
• Revenue at risk from contracts ending in the next 60 days.
• Consultant activity against outcomes, so coaching is based on evidence.

This is a data analytics project rather than an AI one, and it is often the quickest return.

What does it cost?

Indicative Australian market ranges, AUD ex GST, for builds around an existing ATS. These assume agency rates in the ranges in our developer rates guide; roughly $1,000 to $1,600 a day for a mixed team means a $40,000 build is about 25 to 40 days of work.

Build	Indicative range
AI shortlisting assistant over ATS data, with audit log and bias testing	$30k to $90k
Temp onboarding and timesheet app connected to payroll	$50k to $150k
Margin and performance reporting across ATS, payroll and accounting	$25k to $70k
Talent pool retention and consent workflow	$15k to $40k

Running costs include hosting, model usage for the AI parts and maintenance, usually budgeted at 15 to 20% of build cost a year. These are ranges, not quotes.

How AWLabs approaches recruitment software

AWLabs hasn’t yet delivered a project for a recruitment agency, so we don’t claim sector experience we don’t have. We bring integration, data and AI engineering, and your employment lawyer and privacy adviser own the legal interpretation. The first step is a Development Consult. If custom work makes sense, our paid Discovery Audit maps your ATS, payroll and reporting and confirms the API access, with the fee credited in full against the build. You then receive a fixed price Schedule of Work with three options: The Minimum, The Recommended and The Complete. Senior engineers lead every build, updates arrive every Tuesday and Friday, Sydney time, and the code and IP are assigned to you as they are paid for.