Every recruitment business knows the pitch.
The AI tool will cut time-to-hire by 40%, remove bias from candidate screening and halve recruiter workload. The dashboard looks slick. The case studies are impressive. The projected return makes the decision feel obvious.
Senior leadership gets excited.
Then, six months after implementation, reality hits.
Candidate matching is producing obvious poor fits. Time-to-hire has barely moved. Your biggest billers have returned to their old ways of working. Data problems have been amplified rather than solved. You may also have taken on compliance risks nobody identified during procurement.
And getting out of the contract is rarely as easy as getting into it.
The problem is not that AI recruitment tools don’t work. It is that businesses often buy them without being clear about the problem they need to solve, the evidence behind the vendor’s claims or the responsibilities they are taking on.
Start with the problem
The pressure to adopt AI in recruitment is intense. New tools appear constantly and standing still can feel like falling behind.
So when a credible contact recommends a tool, or a polished demonstration captures the attention of senior leaders, investigating it feels both justified, even urgent.
But the first question you need to ask should have nothing to do with the tool.
What problem are you trying to solve?
It might be slow screening, inconsistent shortlisting, poor CRM data or too much time spent on administration. Some of these problems may have more to do with process or management than technology.
Without a clear problem and a baseline for improvement, any tool can look useful. You also have no reliable way to tell whether it worked.
The demo that proves nothing
Vendor demos show the tool working at its best. The data is clean, the examples are carefully chosen and the workflows run smoothly.
None of that tells you how it will perform on your data, in your market, with your team.
Before committing, test the tool using representative data and real recruitment scenarios. Compare it with your current process. Look at the poor matches as closely as the successful ones. Find out how much human intervention is needed to achieve the results shown in the demo.
Remember, a compelling demonstration shows what a product can do under favourable conditions. It does not prove what it will do for your business.
Bad data, amplified
AI does not repair weak data foundations simply by being more sophisticated.
AI systems learn from the data they're trained on. If that data contains inconsistent decisions, skewed outcomes, missing information or embedded bias, an AI system can reproduce those problems more quickly and perpetuate those patterns at scale.
Amazon’s experimental recruitment tool remains a useful example. In 2014, the organisation introduced automated screening. It abandoned the system after discovering that it had learned from, and was perpetuating, historical hiring patterns in a male-dominated technology workforce that penalised women.
The system had simply learned this from the data it was given.
Yet many recruitment businesses will still procure AI before considering the quality of their own data. They hope the technology will compensate for poor records, inconsistent processes or years of subjective decision-making.
It will not.
Before implementation, or even thinking about buying any AI tools, you need to thoroughly assess whether your data is complete, consistent and accurate. Examine whether historical outcomes reflect sound business logic or patterns you do not want repeated. Decide which data should be cleaned, excluded or treated with caution.
This is the foundational work covered by our Data Edge Framework. It may be less exciting than choosing a new platform, but it often determines whether the platform succeeds.
When data foundations are solid and clean, AI tools work as intended. When they're not, no amount of algorithmic sophistication compensates.
The responsibility stays with you
Buying technology from a third party does not hand over responsibility for how your business uses it.
As Tanvi Mehta Krensel, Partner in Data Privacy and Cybersecurity at Squire Patton Boggs, explains in our recently published paper, AI Tools Due Diligence Guide for the Recruitment Sector:
“The legal and regulatory risk in AI procurement sits primarily with you as the buyer, not the vendor. Your contracts won't protect you if the underlying tool creates discriminatory outcomes or breaches data legislation. That's why understanding what you're buying and what you're responsible for before implementation isn't optional; it's essential.”
If a screening tool disadvantages a protected group, it is your recruitment process under scrutiny. If candidate data is processed unlawfully, pointing to the vendor will not remove your obligations.
Due diligence needs to assess more than performance. It must establish what the tool does, what decisions it influences, what risks it introduces and what controls you need around it.
Can you explain the decision?
Candidate screening is not a low-consequence use of AI. These decisions affect people’s access to work and can expose recruitment businesses to legal, regulatory and reputational risk.
Explainability is essential.
Sophisticated AI systems, particularly those using deep learning or complex neural networks, often operate as "black boxes" where it's difficult to explain exactly why they made a particular decision. This creates an accountability problem when the AI tool is making decisions that affect people's lives and livelihoods.
Could you tell a candidate why the system scored or rejected them? Could your consultants understand and challenge its recommendation? Could you explain the decision-making process to a client or regulator?
The answer does not need to be a technical account of every calculation. It does need to be clear enough to show which information shaped the outcome and how human oversight operates.
If you cannot explain or defend a decision, you should not automate it.
The hidden compliance burden
AI tools create compliance obligations that are easily overlooked during procurement. GDPR, Australia’s Privacy Act and the EU AI Act can all affect how automated decision-making and candidate data are handled.
Recruitment businesses may need to tell candidates when AI is being used in decisions that significantly affect them, provide ways to access relevant information or challenge an outcome, and assess the privacy risks created by new data flows.
The vendor relationship adds another layer. You need to know where candidate data is stored, who can access it, how long it is retained and whether it crosses borders. If the tool collects data from social media or other public platforms, you also need to consider data protection law and the platforms’ terms of service.
These obligations remain with you regardless of how confidently a vendor describes its own compliance. By the time a candidate complains or a regulator asks questions, the underlying process is already in place and much harder to change.
Put the claims to the test
Rigorous due diligence does not require every recruitment leader to become an AI expert. It requires a structured way to test the claims being made.
Before buying an AI recruitment tool, establish:
The problem: What specific business problem should the tool solve?
The evidence: Has it been tested on representative data and real scenarios?
The comparison: Does it perform better than the current process?
The data: Is the underlying information accurate, complete and suitable?
The decisions: Can its outputs be understood, challenged and overridden?
The risks: Could it create discriminatory, privacy or security concerns?
The obligations: What must candidates be told and what rights can they exercise?
The ownership: Who is accountable for monitoring outcomes after launch?
The exit: Can you retrieve your data and leave the vendor without unreasonable cost or disruption?
A tool that cannot withstand those questions is not ready for implementation, no matter how convincing the demonstration.
Discipline is the advantage
Moving quickly on AI can feel like a competitive necessity, but buying quickly is not the same as implementing well.
The recruitment businesses most likely to gain an advantage will be those willing to test vendor claims, examine their own data and walk away when a product cannot demonstrate better results than the process it is meant to replace.
That discipline may slow procurement, but it also reduces the risk of poor adoption, wasted investment, candidate harm and compliance problems that surface long after the contract is signed.
Our AI Tools Due Diligence Guide for the Recruitment Sector provides practical questions and assessment criteria for evaluating AI tools before implementation.
Download the AI Due Diligence Guide and find out what you are really buying before you commit.
