AI for Recruitment Assessment: Off-the-Shelf Tools vs Custom Hiring Workflows
Off-the-shelf AI recruitment tools start cheap, but you inherit the vendor's model and bias controls. A custom hiring workflow fits your roles and your compliance line. Here is how UK mid-market teams choose.
By Ivan Pylypchuk, CEO of SoftBlues. Has led Claude and AI implementations for finance, legal and healthcare teams across the UK and Ireland.
Off-the-shelf AI recruitment tools screen and score candidates quickly and cost little to start, but you inherit the vendor's model, bias controls and data policy. A custom hiring workflow costs more up front and is shaped around your roles, your compliance line and the systems you already run. Most UK mid-market firms should start with an off-the-shelf tool and build custom only where the assessment is core to how they hire.
At SoftBlues, an AI implementation firm working with regulated mid-market companies across the UK and Ireland, we help teams decide which of the two to buy before they spend a penny on either. This guide is the version of that conversation we have most often.
Key facts
Who this is for, and who it isn't
This is for a 50–500-person UK or Ireland firm that is hiring at enough volume to feel the strain, wants AI to help sift and assess, and cares about getting the compliance right the first time. Finance, legal, healthcare and professional-services teams are the typical reader.
It is not for a solo founder wanting a quick CV filter for one role, and it is not a ranked list of products. We name categories and price bands, not a "best tool of 2026".
What does "AI recruitment assessment" actually mean?
AI recruitment assessment is any software that uses a model to evaluate candidates rather than just store them. In practice that covers CV and application screening, skills and cognitive tests scored by a model, structured-interview scoring, and ranking or shortlisting.
It is worth separating three things people lump together. Sourcing tools write job ads and find candidates. Assessment tools judge them. Workflow tools move an application through your stages. This guide is about the middle one, because that is where the accuracy, the bias risk and the legal exposure sit.
The reason so many teams are looking now is volume. With applications per role more than doubling since 2022, a human-only sift is slower and less consistent than it used to be, which is exactly the pressure that pushes firms toward automation before they have thought through the controls.
Off-the-shelf or custom: which should you choose?
The honest answer for most mid-market firms is: start off-the-shelf, and only build custom where the assessment itself is a competitive advantage or the off-the-shelf model cannot be made compliant for your roles.
Off-the-shelf means a product like an applicant-tracking system with built-in scoring, or a dedicated assessment platform. You get a working tool in days, a fixed monthly cost and a vendor who maintains the model. The trade-off is that the model was trained on someone else's data, its scoring logic is largely a black box to you, and you configure rather than control it.
Custom means a workflow built around your competencies, your scoring rubric and your data, usually assembled from a foundation model like Claude plus your own assessment criteria. You control what the model sees, how it scores and how a human signs off. The trade-off is cost, a longer build and the need to own testing and monitoring yourself.
| Off-the-shelf tool | Custom hiring workflow | |
|---|---|---|
| Time to live | Days to weeks | 6–12 weeks (our data) |
| Cost | £8–£30 per user/month (market) | £15,000–£60,000 build (our data) |
| Fit to your roles | Generic, configurable | Built to your competencies |
| Control of scoring | Limited, vendor's model | Full, your rubric |
| Bias testing | Vendor's, ask for evidence | Yours to design and run |
| Best for | Standard roles, high volume, fast start | Core roles where assessment is a differentiator |
| Avoid if | You cannot see or audit the scoring | You need something live next week |
What do AI recruitment tools cost in the UK?
Pricing splits along the same line as the build decision.
| Option | Typical cost | Provenance |
|---|---|---|
| ATS with built-in scoring | £8–£15 per user/month | Market, July 2026 |
| Dedicated assessment platform | £15–£30 per user/month, or per-assessment | Market, July 2026 |
| Custom assessment workflow (build) | £15,000–£60,000 one-off | Our data, indicative |
| Custom workflow (run + monitor) | £500–£2,000/month | Our data, indicative |
Market figures are indicative ranges from published vendor pricing as of July 2026 and will vary by seat count and features. The build figures are from our own UK engagements and depend on how many roles and how much integration you need. Treat both as starting points for a quote, not a benchmark.
Where off-the-shelf tools fall short
Three gaps show up again and again once a tool is live.
The first is fit. A generic screening model scores against patterns in its training data, not against what actually predicts success in your roles. For a specialist finance or clinical position, that mismatch is where good candidates get filtered out.
The second is opacity. When a candidate or a tribunal asks why someone was rejected, "the software scored them low" is not an answer you can defend. If you cannot see the scoring logic, you cannot explain the decision, and under UK GDPR you may have to.
The third is bias you did not introduce but still own. A model trained on historic hiring data can reproduce the very patterns you are trying to move away from. The ICO audited AI recruitment providers in 2024 and issued nearly 300 recommendations to improve compliance, which tells you how common these gaps are (ICO).
None of this rules out off-the-shelf tools. It means you buy one with your eyes open, ask for the bias-testing evidence, and keep a human in the loop.
What a custom hiring workflow looks like
A custom workflow is not a science project. It is a defined sequence with a model doing the heavy lifting and a person owning the decision.
A typical shape: applications land in your ATS; the model sifts them against a rubric you defined, not a generic score; shortlisted candidates get a structured assessment mapped to the role's competencies; a hiring manager reviews the model's reasoning and can override it; and every step is logged so the decision can be explained later.
Worked example. A mid-market professional-services firm was receiving several hundred applications per role and sifting them by hand. We built a workflow that scored applications against the firm's own competency framework and produced a short, plain-English rationale for each shortlist decision, with a hiring manager signing off before any rejection. First-sift time dropped from days to hours, and every decision came with a record the firm could stand behind. Our own assessment build for an HR platform, SofiaHR, shows how far this can go when candidate assessment is the core product.
The compliance line you cannot cross
Both routes sit under the same UK rules. The Equality Act 2010 means you must not discriminate, directly or indirectly, and an AI tool that disadvantages a protected group does exactly that even if no one intended it. UK GDPR means processing must be fair and transparent, and Article 22 gives candidates rights around decisions made solely by automated means. The ICO expects bias testing, transparency and a genuine human review, not a rubber stamp.
If you hire in the EU, add the EU AI Act, which treats recruitment and selection AI as high-risk with obligations from 2 August 2026, including risk assessment, human oversight and the right to an explanation.
We keep this to a full checklist in a companion piece, the AI recruitment assessment compliance checklist, which is worth reading before you sign anything. The short version: the vendor supplies the tool, but the legal duty is yours.
Red flags when choosing a vendor
Questions to ask on the call, and what a good answer sounds like
1. How was the model tested for bias, and can I see the evidence? A good answer names the protected characteristics tested, the method, and hands over a report. Silence here is the red flag.
2. Can you explain, in plain English, why a candidate was scored the way they were? A good answer shows a per-candidate rationale, not just a number.
3. Where does our candidate data go, and for how long? A good answer gives the hosting location, the retention period and the sub-processors.
4. How does a human override a decision? A good answer describes a real step in the workflow, not a theoretical option.
5. Who is liable if the tool discriminates? A good answer is honest that the employer carries the duty and explains how the vendor helps you meet it.
Frequently asked questions
Is it legal to use AI to screen candidates in the UK?
Yes, if you use it lawfully. You must not discriminate under the Equality Act 2010, you must process data fairly and transparently under UK GDPR, and candidates have rights around solely automated decisions. The tool being legal to sell does not make every way of using it compliant.Should a small team buy off-the-shelf or build custom?
Almost always off-the-shelf to start. A custom build only pays back when the assessment is core to how you hire, you have the volume to justify it, or no off-the-shelf model can be made compliant for your roles.Do AI recruitment tools actually reduce bias?
They can, and they can also amplify it. A tool trained on biased historic data will reproduce that bias. Reduction only happens when the model is tested, monitored and paired with human review.How long does a custom hiring workflow take to build?
In our experience, 6 to 12 weeks for a working workflow, depending on how many roles it covers and how much it integrates with your ATS (our data).Can I keep my existing ATS and add AI assessment?
Usually yes. Most custom workflows sit alongside an existing applicant-tracking system rather than replacing it, using it as the system of record.What is the single biggest mistake teams make?
Letting a tool auto-reject candidates with no meaningful human review. It is the fastest route to both a bad hire and a compliance problem.SoftBlues is a registered Anthropic Partner Network member and a Google Cloud Partner. We are practitioners, not slideware consultants: we build the workflow, test it, and hand you something you can defend. If you are weighing an off-the-shelf tool against a custom build, we can tell you which one fits your roles and your compliance line before you commit. You can see how we approach business process automation, and if AI assessment sits inside a wider hiring or onboarding change, our guide to employee onboarding process automation covers the next stage.
See it in production
Systems we have built and run for clients, with the numbers that came out of them.
Related Articles
AI Recruitment Assessment Compliance Checklist: Bias, Explainability, Consent, Human Review
Your Biggest AI Risk Isn't Technology. It's Your Team.
