Sourcing a Vanishing Skill: APL Programming Talent for a Big Four Legacy Model Conversion

Objective

A Big Four professional services firm engaged Graeme Group to source specialist contractors for a legacy actuarial systems conversion it is delivering for a US life insurance client, rewriting a suite of long-standing actuarial tools written in APL into a modern Python codebase.

APL is the constraint. A terse, symbol-dense array language widely used for actuarial and quantitative work from the 1970s onward, it has since been almost entirely displaced. The conversion needs more than Python engineers, who are readily available. It needs people who can read decades-old APL, reconstruct the intent behind it, and explain that intent to a development team that has never seen the language. That combination is scarce, and the program could not proceed at pace without it.

Graeme Group’s remit was talent supply, not delivery: identify, qualify and place contractors who could hold that bridging role, and keep the seat filled as the program grew. The conversion work itself is performed by the client firm’s own team.

Scope of Work

The mandate was to build a supply of contractors able to work directly with legacy APL, at close to full-time availability and within a defined commercial range, and to sustain it through the life of the program, including replacing a resource mid-engagement without a gap in coverage.

Candidates had to clear a technical qualification gate, not a keyword match: which models they had built or maintained, which APL dialect and how recently, and whether they could explain legacy model logic to Python developers. Each also had to clear the client firm’s independence, background screening and onboarding requirements.

Services Provided

Establishing the Scarcity of the Skill

The starting point was our own bench. Graeme Group maintains a specialist actuarial talent database of more than 550 contractors and candidates. Searching it for genuine APL capability returned roughly one percent of the bench, and one of even that handful was a false positive, because in actuarial CVs “APL” frequently means automatic premium loan, an entirely unrelated policy feature. Any automated scoring for this search needed an explicit guard against that collision before it could be trusted.

Structured Market Sweep

With the internal bench largely exhausted, we ran a paid deep review of the external market: several hundred profiles acquired in full, scored against a purpose-built rubric weighting APL dialect evidence, life and annuity exposure and actuarial credentials, then re-ranked to favor genuine modeling background and narrowed to a written shortlist. Each shortlist entry included its supporting evidence, the gap against the brief, the recommended approach channel, and whether the contact details had been verified as deliverable.

The sweep reached beyond conventional recruitment sources into the surviving APL community itself, language vendor networks, user group and conference lineages, historical practitioner directories, because the people who still write APL are largely not discoverable through standard channels.

Qualification and Submission

Shortlisted candidates were screened by a credentialed actuary rather than a recruiter, on recorded and transcribed calls, so claimed experience could be probed against the actual mathematics needing conversion. Those who cleared were submitted as blinded CVs with a written rationale, the candidate’s limitations stated plainly, and availability confirmed in advance. Where a candidate could evidence ability directly, in one case with working code demonstrating the exact translation the program required, that went to the client with the submission.

An Auditable Search Process

The search ran on a tracker keeping candidate status, the outcome of an individual approach, and source coverage deliberately separate. The tracker distinguished a failed contact attempt from an exhausted lead, and researched profiles from completed outreach. Every sweep was reconciled and graded, so the client could be told not only who had been found but how much of the market had been looked at.

Deliverables

• Blinded, role-titled CVs in Graeme Group format, each paired with a written evidence rationale and a candid statement of the candidate’s limitations.
• Scored and ranked candidate datasets with per-candidate evidence extracts, plus written shortlists with reach, verification and gap notes.
• Recorded and transcribed technical qualification interviews, and a reconciled search tracker with a graded coverage audit per source.
• Full contracting and onboarding administration: contractor agreements, background screening, independence acknowledgements, and ongoing timesheet and invoice management.
• A reusable specialist search methodology, generalized beyond this mandate for future scarce-skill searches.

Outcome

• The Seat Was Filled, Held and Grown: The engagement began with a single named contractor and now runs three concurrently active resources, each approved by name by the end client. When the original resource came off, a qualified replacement was already identified and contracted, and delivery continued without interruption.
• A Requirement the Market Forced Open: Part way through the search the client relaxed its own specification, accepting deep APL capability from analogous quantitative domains rather than insisting on life-actuarial background. Both subsequent placements came through that widened gate. A buyer widening its requirement rather than raising its price is a clear measure of how thin the market is.
• An Evidenced View of a Disappearing Market: The search also produced a durable asset: a documented map of who can still do this work. The recurring barrier was rarely capability. It was availability, retirement, decades of recency decay, or practitioners who would rather keep writing APL than help retire it.
• Accurate Channel Attribution: The paid sweep made that map possible; the placements themselves came through referral chains and warm network paths, and we report the distinction.