Candidate Matching Software: How Scores Are Calculated
Discover how candidate matching software scores CVs against Australian Government ICT roles, what match scores mean and how to judge a tool before you buy.
It's Monday morning. You've got forty CVs to assess against a BuyICT notice for a cloud engineer with NV1 clearance, and the supplier wants a shortlist today. Several candidates mention Azure, but which people have the delivery experience, clearance and availability that make them worth calling?
That's the useful test for candidate matching software in Australian Government ICT: not how many profiles it returns, but whether you can defend the shortlist it helps you build.
Table of Contents
- What Candidate Matching Software Does for Recruiters
- How a Match Score Is Built and What the Numbers Mean
- Why Keyword Search Misses Good ICT Candidates
- Government Fields Your Matcher Must Read
- Where to Find Live Government ICT Roles to Match
- How to Evaluate Candidate Matching Software
- Consent, Governance and Defensible Shortlists
- How to Roll Out Matching and Choose a Tool
What Candidate Matching Software Does for Recruiters
Candidate matching software reads CVs and role requirements, extracts information such as skills, clearance and location, then scores each candidate against each role with supporting evidence. It helps you decide where to focus your assessment, rather than deciding who gets submitted.
For the cloud engineer vacancy, the workflow starts before you open a candidate profile. Once you upload a PDF or DOCX, our parser processes it automatically in the background, turning the resume into information the matcher can compare with a role.
From uploaded CV to reviewable recommendation
The process has three practical moves:
- Extract the evidence. Identify skills, project history, clearance statements and other relevant details from the CV.
- Standardise the language. Map different descriptions of related capabilities into a consistent ICT vocabulary.
- Show the reasons. Display the candidate-to-role score alongside the evidence that supports it.
A keyword search finds a mention of Azure. Matching asks whether that mention describes recent engineering work, a training course or an incidental reference in a project summary. You still need to inspect the answer, but you're starting with a more useful question.
Australian adoption remains early. A 2024 Queensland University of Technology survey of Australian organisations found AI-enabled screening and shortlisting in 3.4% of cases, while 20.9% reported using AI partially or to some extent for those activities.
Those findings aren't a benchmark for product accuracy. They establish why a human-led workflow matters: recruiters need recommendations they can interrogate, not an opaque ranking they're expected to accept.
Practical rule: The software ranks and explains. The recruiter decides who is submitted.
For a panel supplier, that boundary is especially important. A technically promising candidate isn't automatically ready for the particular government contract behind an advertisement.
How a Match Score Is Built and What the Numbers Mean
erecruit builds each match score in four passes. Each pass asks a narrower question about whether the candidate can do the work described in the role.
- Skill coverage checks which required capabilities appear with evidence in the CV. A skills list starts the case; it does not prove delivery.
- Role match compares title, seniority and delivery context with the work required. An architect's responsibilities are not interchangeable with an administrator's.
- Semantic similarity recognises related descriptions. Container orchestration experience can support a Kubernetes match, but it does not establish Kubernetes expertise without supporting detail.
- The final language-model pass confirms and refines the strongest candidates, weighing the combined evidence rather than counting terms.
Together, those passes produce a score and a set of reasons. The reasons matter more than the decimal because they show where the evidence is strong or thin.
The visual maps the information a recruiter should inspect, including clearance fit. Treat those details as review prompts that sit alongside the score.
Match score bands and what to do with each
- Score band — Score range — What it indicates — Typical recruiter action
- High — 0.70 and above — Skills, role and evidence align strongly — Check the evidence panel and consent status before shortlisting
- Medium — 0.50 to 0.69 — Partial fit, with one or more signals missing — Read the gap in the evidence before deciding
- Low — Below 0.50 — Weak fit or thin evidence — Usually filtered from digests; review only for hard-to-fill roles
The 0.70 and 0.50 thresholds are erecruit's own High, Medium and Low bands. They are not an external standard or government eligibility criteria. They guide review, and they do not certify suitability.
In a hypothetical comparison, one candidate scores 0.74, supported by recent cloud delivery and a current clearance statement. Another scores 0.48, with cloud keywords appearing only in an unrelated project from 2012. The higher score gives you a reason to investigate first. It does not prove that the clearance statement remains accurate, that the candidate can start on time, or that the supplier has permission to represent them.
For a broader explanation of evaluating recommendations against evidence, the Synopsix guide to better talent decisions provides useful context.
Check the evidence panel for every High match before adding it to a shortlist. A match score is a ranking signal. It is not a probability of placement, and it does not replace eligibility checks.
Why Keyword Search Misses Good ICT Candidates
More applicants won't necessarily fix a technical-fit problem. Jobs and Skills Australia's analysis of changing IT recruitment trends reports that in 2025–26, 32% of IT employers recruited, compared with 49% across all industries, while 34% received at least 20 applicants per vacancy.
The recruiter's task is to distinguish deployable capability from superficial similarity. That requires reading beyond the job title.
Compare the work, not the keyword count
Consider these hypothetical CVs for the same government cloud engineering role.
- Evidence — Candidate A — Candidate B
- Headline — Cloud engineer — Infrastructure platform specialist
- Technical description — Repeated Azure and Kubernetes keywords — Describes Azure platform automation and managed cluster operations
- Delivery context — Training exercises without production responsibilities — Production deployment, monitoring and controlled releases
- Constraints — Clearance and availability unstated — Clearance statement and availability provided
- Recruiter's next action — Investigate substantial gaps — Verify the stated evidence
A title-based search may favour Candidate A. An evidence-led assessment gives Candidate B a stronger case for a call, subject to verification.
Our CV matching uses a canonical Australian Government ICT vocabulary covering cloud, data, security and project delivery. This helps recognise equivalent terminology without requiring candidates to write identical resumes.
Equivalence needs limits. Java isn't JavaScript, Microsoft 365 administration isn't Azure cloud architecture, and container experience doesn't automatically demonstrate production Kubernetes operations. The principles in our skills-based hiring guide help keep the assessment focused on demonstrated capability.
Use this distinction when reviewing demonstrations of automated matching:
Measure false positives per shortlist, meaning the High matches a recruiter rejects after checking the evidence. Record the reason for each rejection, such as an unsupported platform skill or unsuitable contract timing, rather than treating faster search as proof of better matching.
Government Fields Your Matcher Must Read
A security clearance works like a building pass, not a qualification badge. It determines which doors a person may open, and only while that pass remains current. If a role sits behind a door the candidate cannot enter, the strongest technical CV in the pile changes nothing.
The AGSVA Security Clearance Applicant Guide Book dated March 2025 sets these access ceilings for Baseline, NV1 and NV2.
Security clearance levels and what a matcher should verify
- Clearance level — Access ceiling — What the matcher should verify
- Baseline — Classified resources up to and including PROTECTED — Current status, expiry date and that the role does not require a higher level
- NV1 — Classified resources up to and including SECRET — Current status, expiry date and the role's NV1 requirement
- NV2 — Classified resources up to and including TOP SECRET — Current status, expiry date and the role's NV2 requirement
Ceilings describe the highest classification a level can reach. They do not confer automatic access to every system at that classification, because agencies still apply need-to-know rules and their own system controls. Read the role's stated requirement and verify the candidate's circumstances rather than inferring readiness from a CV.
Our matching extracts skills and security clearance from a resume and scores it against BuyICT opportunities. That extraction is a reading of the document, not a clearance check. “Willing to obtain NV1” is not the same as “holds NV1”, and a matcher that blurs the two produces shortlists you cannot defend in front of a panel.
Read the procurement context
BuyICT is a procurement portal organised into marketplaces, including ICT Labour Hire. Match the person to the opportunity behind the advertisement, not merely to its title.
Confirm these fields before shortlisting:
- Buyer. Identify the government agency purchasing the work, separately from the recruitment firm advertising it.
- Delivery state and working arrangement. Check location, onsite requirements and any stated flexibility.
- Closing time. Record the notice's deadline and timezone, not just the public advertisement's date.
- Invitation posture. Establish whether only invited sellers can respond and whether the relevant supplier can participate.
- Clearance evidence. Verify current status, recorded expiry or review information, and any uncertainty against the role requirement through the appropriate authorised process.
A missing field should stay visibly unconfirmed. Don't let a strong technical score conceal an unanswered procurement question, because the panel will ask about it and a blank is harder to explain than a “no”. Each field you verify becomes a line in the record behind the shortlist.
A candidate can be an excellent cloud engineer and still be unsuitable for this particular submission.
Check current clearance status with AGSVA and current marketplace and invitation details in BuyICT. Note the date you verified each one, then re-check whenever the notice changes or the candidate's circumstances shift.
Where to Find Live Government ICT Roles to Match
A candidate pool is only useful when it is tested against roles you can still act on. A shortlist built on a closed notice, or on an advertisement whose contract has already been awarded elsewhere, wastes the same hours as no shortlist at all. erecruit links live BuyICT opportunities to the job ads tender-panel agencies are running, and prunes matches for closed opportunities or stale ads so the feed only shows roles that can still be bid for.
Begin with the role family. A generic government jobs search returns a flat list where an APS payroll analyst and a security-cleared cloud engineer sit side by side, and the ranking tells you nothing about which one a panel agency is actively trying to fill. Our guide to government jobs in Australia explains how that market is organised and is a reasonable place to start.
Choose a destination that matches the work
Search destinations fall into three groups:
- Employment roles. APS ICT jobs, government software developer and engineer jobs, government DevOps jobs and government cloud jobs.
- Specialist capabilities. Government cybersecurity jobs, government data and AI jobs, and government technical architect jobs.
- Contract work. ICT contractor jobs in Australia, Azure Government jobs in Australia and security-cleared technology contract opportunities.
The employment side of that market is measurable. The official APS workforce tables recorded 7,556 ongoing employees in ICT and Digital Solutions at 30 June 2024, up from 6,757 a year earlier, representing 4.4% of the ongoing APS workforce, alongside 5,506 employees in Data and Research.
Contracting needs its own search path. The 2025-30 Data, Digital and Cyber Workforce Plan reports that more than 40% of external labour expenditure identified in an ANAO review was attributed to ICT and Digital Solutions, equivalent to more than 6,000 full-time-equivalent employees, excluding Defence.
Then run the match. Pick a role family, read the live requirement in full, and only after that test your candidate pool against it. For APS employment, open APS ICT jobs; for contracting, start with ICT contractor jobs and BuyICT opportunities. That order matters. Source first and read second, and you end up with a stack of plausible CVs that look interchangeable and none of which you can defend to a buyer or a candidate.
How to Evaluate Candidate Matching Software
Use a vendor demonstration as a test of shortlist defensibility, not interface polish. Bring a role you understand, candidates you've assessed and examples where a title hides an important mismatch.
Evaluation criteria for candidate matching software
- Criterion — Question to ask the vendor — What good looks like — Red flag
- Matching accuracy — How is a score built, and can you show results on past shortlists? — Layered scoring with stated thresholds and measured false positives — A single percentage with no published method
- Evidence fields — Which evidence sits beside each score? — Role, clearance, location and timing, each traceable to its source — A score with no reasons attached
- Contract-to-ad linking — How is a job ad tied to the contract behind it? — An evidence-backed link to a navigable source — Links based only on title similarity
- Re-scoring — What happens when a notice or ad changes? — Automatic re-score with a change alert — Manual re-upload to refresh results
- Consent workflow — Is representation consent recorded before submission? — A gate that blocks submission until consent is approved — Consent kept as a free-text note
- Candidate data — How long are CVs and cached candidates kept, and who can delete them? — Documented retention, deletion and access controls — Retention terms that are unclear or missing
- Integrations — Is there an API for matches and reporting, and which filters does it support? — A matches endpoint with the score breakdown and documented filters — Export only, or no documented API
Weight these criteria by what they protect. Accuracy and evidence protect shortlist quality; consent and retention determine whether your team can use candidate information appropriately; integrations determine whether the workflow fits your CRM and reporting stack.
A good integration answer can include a limitation. Our Agency plan includes everything in Recruiter plus API access for opportunities and matches, but the API supports only limit and offset parameters, with no filtering or search. Your own system must handle filtering after retrieval.
Compare the usable workflow
Our Recruiter plan includes 10 CV uploads and a full match reveal: which agency is chasing each contract, the match score and confidence, supporting evidence, and a direct link to every job ad. Compare those inclusions with the size of your library and the work your recruiters need to perform.
Don't assume match score and confidence mean the same thing. Have the vendor demonstrate what each field represents and what happens when evidence is missing.
Pilot test: Use a hypothetical evaluation set of 20 known candidates against three live roles, including clear fits and clear mismatches.
Count which High matches survive recruiter review. Also inspect rejected or low-ranked candidates who should have surfaced, so a tidy shortlist doesn't hide missed talent.
Consent, Governance and Defensible Shortlists
Governance belongs inside matching, before a recommendation becomes a submission. The Australian Government's 2025–30 Data, Digital and Cyber Workforce Plan identifies lengthy recruitment processes as a critical challenge and continuing critical skill shortages that put digital projects at risk.
That creates pressure to automate, but a quicker workflow can also move an unauthorised or incorrectly matched candidate further before anyone notices. Neither a high score nor possession of a CV establishes permission to represent someone for a particular opportunity.
Put controls in submission order
- Record consent before submission. Capture the candidate's representation approval for the relevant opportunity and supplier.
- Preserve CV provenance. Keep the origin and version of the resume used for assessment.
- Confirm role-to-contract evidence. Make sure the advertisement belongs to the intended procurement opportunity.
- Require human review. Assign a recruiter to approve High matches before they progress.
- Save score explanations. Preserve the evidence supporting the decision, not just the final number.
- Log re-scores. Retain what changed and why the candidate required another review.
- Set retention and deletion rules. Define ownership, access and removal responsibilities for stored candidate information.
- Provide a correction route. Let authorised users resolve inaccurate fields and reassess affected matches.
Our pipeline lets users source candidates, track them from unqualified to placed and gate submissions on consent. That gate makes approval a workflow condition rather than a note someone might overlook.
Make explanations useful to reviewers
A reviewer should be able to distinguish information extracted from a CV, information stated in a notice and information confirmed directly with the candidate. Our guide to matching evidence helps readers interpret the reasons attached to a recommendation.
Governance also improves operational quality. Checking the intended supplier, location and start date can prevent duplicate representation and incorrect contract assumptions before they become submission problems.
These controls don't establish legal compliance by themselves. Have the responsible privacy or compliance lead assess the complete workflow, including access permissions and retention, rather than treating a consent checkbox as the whole answer.
How to Roll Out Matching and Choose a Tool
Start with a bounded pilot, not your entire historical database. The technical setup is straightforward: sign up, upload a CV and let the matcher process it; operational trust requires deliberate review.
Keep the recruiter in control
- Choose the library. In Resumes, use Personal for your own CV or Organisation for a team-accessible pool, so ownership and reuse are clear.
- Check extraction. Upload PDF or DOCX files, allow background parsing to finish and, as a practical pilot exercise, compare ten records with their source CVs before relying on scores.
- Set digest thresholds. Start with High matches only and expect one best match per opportunity ad, widening the threshold only after reviewing relevance.
- Validate the opportunity. Link each live BuyICT opportunity to its advertisement and confirm buyer, delivery location and closing time so recruiters target actionable work.
- Run the candidate test. Apply the proposed 20-candidate pilot to three live roles and record which High matches are genuine fits, giving your team a reviewable baseline.
- Activate submission controls. Enable the consent gate and assign responsibility for changed matches, because our system re-scores CVs whenever a BuyICT opportunity or its live advertisement changes.
Keep a permissioned candidate cache under your documented retention policy. Someone unsuitable for an immediate start may fit a later contract, and the 2025 APS Digital Workforce Insights report projects a potential digital talent shortfall of more than 8,000 people over the next five years.
Choose for your operating model
A solo recruiter should favour affordable seats and clear digests. A labour-hire panel agency needs a shared Organisation library and a workable CRM reporting connection; a panel-supplier consultancy should prioritise consent controls, auditability and documented retention.
The Australian Government's digital procurement performance reporting states that government sourced more than AUD 6.4 billion of digital products and services through its marketplaces and agreements in 2023–24. That total isn't all labour hire, but it illustrates why procurement-aware matching matters.
Run the pilot against live opportunities and linked job ads, compare the results with the score bands and evaluation criteria, and have your compliance lead approve the consent workflow before the first submission.
eRecruit connects Australian Government ICT opportunities with live panel-supplier job ads and provides CV scores with supporting evidence. Visit erecruit to explore live opportunities, upload a permitted CV and start testing whether your next shortlist stands up to recruiter review.
