How Berkeley Group lifted apprentice retention from 62% to 85%.

Berkeley Group attracted strong interest in its apprenticeship programmes. The early-careers team had no shortage of applications, no obvious problem at the top of the funnel. The problem was further down: too many apprentices were entering the role without a full understanding of what construction work actually involved, then leaving in the first six to twelve months.

Two things were driving it. The first was CV screening. For entry-level roles where applicants typically have limited prior experience, a CV doesn’t tell you much about whether someone will succeed on site. It mostly tells you who has already had similar opportunities, which selects against the candidates most likely to widen the talent pool. The second was the expectation gap. Apprentices were applying for what they imagined construction to be, not what it actually is — and discovering the difference in their first months on the job.

The brief to ThriveMap was specific. Improve early retention, especially in the critical first year. Make sure candidates know what they’re signing up for before they say yes. And do it without leaning on prior experience as a proxy for ability, which was excluding the candidates Berkeley most wanted to attract.

The core reframe was simple. The hiring process went from asking “can you get the job?” to asking “is this the right job for you?” That sentence sounds soft, but it has hard structural consequences. It moves the candidate from being evaluated to being informed; it moves the employer from selecting against a CV to selecting for demonstrated behaviours; and it gives candidates who realise the work isn’t for them a clean way to step out before either side has invested further.

ThriveMap built Berkeley a Realistic Job Assessment as a day-in-the-life simulation of an apprenticeship. Candidates worked through the actual realities of the role: early starts, physical demands, on-site working, the kind of communication, judgement, and stress management that construction sites require. Rather than testing abstract reasoning or general knowledge, the assessment scored candidates on the practical behaviours that predict success on site. And critically, candidates were given the option to self-select out at any point if the role wasn’t right for them.

Berkeley measured retention at two windows: six months in (when most expectation-gap leavers go), and twelve months in (where the first-year picture settles). Across cohorts that went through the Realistic Job Assessment, both windows improved meaningfully.

The bigger move is at twelve months. A 23-point lift in first-year retention is meaningfully more impressive than the headline six-month figure, because the twelve-month window catches the leavers who pass the early adjustment but then drop out when the realities of the work catch up with them. That second wave is exactly what a realistic preview prevents.

The unexpected result was who Berkeley ended up hiring. Across the cohorts that went through the new process, 32% of apprentices hired were women — nearly three times the construction industry average of around 11%. This wasn’t the result of a targeted diversity initiative. There were no quotas, no demographic targets, no positive action programmes layered on top.

It came from removing CV screening. CVs in construction reliably select for candidates who already look like the existing workforce — typically because the proxies they screen on (prior site experience, relevant trade qualifications, network references) are themselves the product of a workforce already skewed toward one demographic. When Berkeley moved to competency-led assessment, those proxies stopped doing the screening, and a different group of applicants made it through to interview. The women who applied for these roles weren’t a new population. They’d been applying all along. The hiring process had simply been quietly filtering them out.

The diversity gain wasn’t a bonus. For an industry under sustained pressure to widen its talent pool, it was arguably the most strategically important result of the work.