AI can speed up hiring, but it can also inadvertently screen out great candidates. We explore how biased AI filters in FMCG manufacturing miss real skills, and share some practical steps to audit your process, reduce false negatives, and build a fairer, skills-first pipeline.
AI can make hiring better—or shrink your pipeline. If your models mirror yesterday’s criteria, they’ll screen out today’s great people.
The bias problem nobody budgets for
Automation is brilliant at taming volume; it is terrible at reading context. Early filters—JD parsers, CV rankers, one‑way video screens—inherit yesterday’s rules and quietly delete tomorrow’s hires. As Fast Company points out, many tools still over‑index on keyword matches and formatting quirks rather than true capability. And Harvard Business School’s Hidden Workers research shows how rigid screens push capable, motivated people to the edges—just because their path isn’t linear.
“If you want to test your AI, run a quick false‑negative audit. Pull a random set of auto‑rejected applications from last quarter; mask identity; have hiring managers re‑score them against outcomes, not pedigree. If you find even a handful who should have advanced, time to review.
AI should find more great people, not hide them. That means human‑in‑the‑loop decisions, fresher training data, and a skills‑first mindset—not checkbox compliance.”
LIBBY NOBLE, Business Manager for Supply Chain, Operations & Engineering
FMCG manufacturing: high volume, low quality? What’s getting lost
If you’re hiring in FMCG manufacturing right now, you’re not alone in feeling frustrated. Application volume is high—but the quality often isn’t. To cope, many teams lean on AI‑powered ATS filters. They help with speed, but too often optimise for keywords over capability. Result: great people get screened out before a human sees them.
Real‑world examples we’re seeing:
- A factory candidate’s CV isn’t polished—but they’ve done the job for 15 years. Filtered out.
- A Supply Chain Analyst and a Forklift Driver assessed by the same logic. Mis‑scored.
- A migrant worker uses non‑local terminology or imperfect grammar. Deprioritised.
- Context is lost; nuanced experience and transferable skills get binned. Missed.
- Creative or non‑standard CV designs trigger parsing errors. Candidate lost.
- A seasoned FLM writes “batch runs” instead of “production cycles.” Ignored.
- A maintenance tech writes “fixed downtime issues” vs. “equipment optimisation.” Overlooked.
None of this is malicious. It’s what happens when one set of rules tries to govern a messy, human labour market.
What smart FMCG employers are doing instead
The smartest employers are teaching tools what “good” looks like with skills, outcomes, and context.
Start by rewriting what success actually means in the role. Not “3–5 years in X” or “degree in Y”. Spell out the competencies and outcomes: “can run a line changeover in under 20 minutes”, “can model inventory scenarios under variable demand”, “has reduced scrap by X%”. Bring hiring managers into that conversation early—especially for frontline and technical roles. They know where the real value sits.
Then recalibrate your screening. Weight for adjacent skills, not just exact keyword matches. If your star Line Lead says “batch runs”, teach the system that it’s equivalent to “production cycles”. If a great maintainer writes “fixed downtime issues”, treat it as evidence for “equipment reliability”. You’re not dumbing down standards; you’re translating real-world language into recognised capability.
And remember, no automated system should have the final say on who gets seen. Build human review into any step that rejects candidates. When volume is heavy, review a rotating sample of near-miss applications each week and feed that insight back into the tool. Over time, your false negatives drop and your shortlists get sharper.
In a nutshell:
- Audit your filters. Sit with HR and the vendor; learn exactly what the AI is screening out—and why.
- Differentiate workflows. Don’t apply the same rules to frontline roles and corporate roles.
- Get hiring managers involved early. They know the work; capture their judgment as rules/training data.
- Emphasise context. Manually review a diverse sample each week and feed those nuances back into the system.
- Leverage human judgment. Use AI to support—not replace—human review for skilled trade or high‑impact roles.
- Build for the long game. Tag promising candidates, re‑engage past applicants, and segment pools (shift patterns, line types, plant locations)
Compliance and trust without the headaches
The bar for automated decision-making in hiring is rising. The safest posture is also the most candidate-friendly: be open about where automation is used, keep records of how decisions are made, give people a way to contest outcomes, and make sure a human can intervene. Build that into your flow and you won’t slow down—you’ll clean up your data, simplify audits, and strengthen your employer brand.
A simple 90-day reset (without turning your process upside down)
Weeks 1–2: Map your hiring flow. Anywhere the system auto-rejects, switch to “review required”. Pull 50 recent near-miss applications across a few roles—frontline and corporate—and ask hiring managers which they’d resurrect and why. Capture their language.
Weeks 3–4: Redefine success by outcomes. Refresh the success profiles for priority roles and translate local phrasing into your skills framework. Update screening so equivalent terms carry equal weight.
Weeks 5–6: Check fairness. Look at pass-through rates by stage for the last 6–12 months. Where are certain groups dropping off? Remove sharp edges: rigid keyword rules, unnecessary degree requirements, opaque knockout questions.
Weeks 7–8: Hold vendors to account. Get a plain-English summary of how their models are evaluated and how you can tune them. You want logging, oversight, and a clear way to reduce errors over time.
Weeks 9–12: Pilot, learn, scale. Run the new flow on one site or function. Track the quality of shortlists (skills breadth), interview-to-offer ratio, time-to-shortlist, and candidate feedback. If quality rises and speed holds, roll it out.
Useful resources: ICO recruitment guidance and AI risk toolkit.
In short, the goal isn’t to rip out your tools—it’s to help them make better judgements. Over time, a living talent database will do more for quality and speed than another burst of job-board spend.
Talk to Denholm
At Denholm, we help ambitious companies hire at the level they need to grow. Whether it’s rare skillsets, tight timelines, or complex hiring challenges, we know how to find the people who raise the bar and stay the course.
We don’t just connect you with exceptional talent. Contact us on 03303 359 818 for advice and support. We’re ready to help.
