A mid-tier firm's pivot from AI committee to delivery cell

This case study is drawn from a single ongoing Shepherd Thomas engagement, abstracted to preserve client anonymity. Sector is stated generically, no individuals are identified, specific tools and dates are not named, and the write-up captures the engagement at a point rather than a final outcome. The case is useful because the firm’s pivot from a deliberative committee to a small delivery cell — driven by sponsor patience expiring rather than by planned restructure — surfaces patterns about committee-led AI adoption that the usual success stories do not.

The firm and the engagement

Mid-tier Australian professional services firm in a growth phase. Around 80 staff, an internally-developed strategic framework, several industry verticals, a stated growth target. An earlier external coaching arrangement of approximately a year had focused on AI literacy and committee formation; a frontier-model platform had been progressively rolled out to staff. An AI usage policy had been issued. The firm contracted a more substantive engagement covering discovery, strategy, fortnightly committee facilitation, support for the firm’s projects analyst, and monthly briefings to the executive sponsor.

The committee that did not move the dial

The retainer commenced but the committee did not reconvene for a quarter. BAU pressure on the firm’s side prevented activation. The advisor invoiced approximately half the contracted amount, and sponsor frustration accumulated silently.

When the committee did reconvene, the round-table format produced rich discussion but little delivery. Members reported on personal AI use, traded prompting tips, made occasional substantive contributions. They also displayed the displacement form of Unvoiced staff resistance is the primary failure mode of AI initiatives: tasked with a tooling problem that had a behavioural component, energy gravitated to the behavioural side, and discussions concluded that the real problem was behavioural and therefore out of scope for AI. The pattern was hard to interrupt without sounding dismissive of staff who were, on the surface, engaging seriously with a real problem.

The pivot

The sponsor’s patience expired. An exchange — advisor characterising the committee’s pace as insufficient to move the dial, sponsor confirming the assessment was not new to her — surfaced the position openly. The sponsor then moved unilaterally on staffing, hiring an external AI engineer from outside professional services on a three-month contract, briefed against creating unmaintainable technical debt.

The committee’s role was reconceived as training and communications. Delivery shifted to a small cell of advisor, projects analyst and new hire, working weekly. The lesson is not that committees are wrong but that a deliberative committee in a busy mid-tier firm is asked to do two different things — build literacy across the firm, and deliver tools — and is reliably good at the first while reliably failing at the second. Separating those functions, even after the fact, is what allowed delivery to start moving.

What made the rollout tractable

Once the delivery cell took over, three architectural choices unblocked progress. The platform decision went to a single frontier-model platform firm-wide rather than tiering by seniority — the staff most likely to do new things with AI are not always the most senior, and tiering would systematically miss them (Hire for durable AI judgement, not transient AI mechanics). The knowledge base went into a markdown wiki, deliberately not connected to the firm’s existing document management system: Word-formatted documents carry too much overhead to retrieve efficiently (Structure documents for AI consumption, not just human reading). The rollout went out in two waves with junior and operational staff first, and AI-mediated policy enforcement substituted for conventional training — staff are directed to ask the platform itself about policy questions, which both teaches the platform’s capability and operationalises the policy in one move.

What the case study illustrates

The engagement instances The mid-tier AI adoption threshold, Measure adoption, not just implementation, Useful AI is a context problem and Architect AI around principles, not vendors alongside the unvoiced-resistance pattern named in section three. The distinctive contribution beyond those notes is the committee-to-delivery-cell pivot itself: a structural recognition that deliberation and delivery are different functions, and that fitting both into one body produces neither, even when the body is staffed by capable and willing people. The engagement is ongoing; this write-up captures the pattern at a point rather than a final outcome — see also An ongoing AI advisory engagement with a growing firm for the LLM-as-mentor-shaped variant of the same advisory pattern in a different sector.