Five hard AI questions most boards are ignoring

AI transformation is a human capital discipline. To convert AI spend into lasting market differentiation, Adamantia Velonis of Legora argues that boards must address the supervisory, cultural and cognitive dynamics of knowledge work.

Over the past five years, executive committees have spent countless hours on vendor contracts, security and compliance, and pilot results. Yet despite substantial outlay, a troubling pattern recurs: after launch, a few user groups race ahead, while the wider firm lags, creating a performance gap that dilutes overall ROI.

AI, as a technology, fundamentally disrupts how professional services work is performed. It creates friction across supervision, talent development, commercial models, and culture. However, too often AI deployments are treated as another software rollout, ignoring the reality that firms are human capital ecosystems.

These are the five questions boards need to consider.

1. The Apprenticeship Dilemma

If AI automates first-pass drafting and analysis, how are we structurally ensuring that the next generation develops the judgement required to become future partners?

The apprenticeship model relies on learning through manual execution and repetition. Uncritical automation risks stripping away the "productive friction" necessary for junior skill formation. As a result, if the opportunities for juniors to perform unassisted foundational work erode, they may not develop the experience needed to spot nuanced errors in AI-generated outputs (particularly as senior practitioners "self-serve," and such tasks stop reaching juniors at all).

Boards should focus on two fronts. First, equipping leadership to spot where working patterns are shifting and building in deliberate touch points, such as reverse mentoring and point-in-time feedback, that ensure that the output is reviewed and that the associate understands it.

Second, ensuring that AI tools can be configured to include cognitive checks, such as Socratic prompting, and that verification gates are calibrated to seniority and task risk. For example, a senior partner may reasonably use a tool that delivers a finished product, whereas it may be preferable for a junior to be walked through the same workflow step by step, testing assumptions and justifying judgement calls at the design stage, not only at review.

This “productive friction” must be explicitly preserved on live matters (rather than simply during training) to guard against over-reliance; raising the question of how junior training should be funded in the age of AI, as clients become more sensitive to cost.

2. The AI Disclosure Trap

Is our culture enabling transparent, supervised AI use, or are our people using tools covertly for fear of a performance penalty?

Effective supervision of AI outputs depends on appropriate disclosure of tool use. Yet recent research reveals a stark "disclosure trust penalty": professionals who admit to using AI are systematically rated as less competent and trustworthy by evaluators. (This is distinct from the covert use of unapproved AI tools; it concerns failure to disclose the use of tools the firm has already sanctioned). If associates believe disclosure will damage their performance reviews, they will not stop using AI but will avoid disclosing its use.

Underground use creates regulatory exposure under supervisory rules (such as the SRA’s guidance for solicitors), breaking the chain of professional accountability, and risking reputational harm if AI-generated outputs are presented without testing their underlying reasoning. The fix requires a culture of psychological safety, clear guidance, and systemic visibility, spanning three audiences: clients, supervisors, and the internal institutional record of AI involvement in the work itself. On culture, boards should ensure that their AI disclosure framework explicitly formalises partner behaviour; requiring partners to signal legitimate use of AI solutions at the point of delegation is one of the most effective ways to normalise transparent use.

3. The Productivity Gap

When AI frees up hours across the firm, where do they go, and are our commercial models designed to capture that value?

In a billable-hour environment, unmanaged efficiency gains either compress revenue or simply evaporate. Firms need an explicit strategy for reallocating freed capacity. High performers redirect saved hours into higher-margin fixed-fee structures, expanded advisory work, or business development. The right approach depends on client price sensitivity. For example, in bet-the-company litigation, clients rarely want a smaller bill. Instead, they want freed time reinvested into winning the case.

Impact clearly depends on the legal market and client base served. To inform strategy, boards should insist on tracking profit per matter and turnaround speed. This is tied to compensation models, below.

4. (Dis)incentive Structures

Do our partner compensation models disincentivise the behaviours needed to drive transformation?

Many firms still compensate partners on revenue generated rather than profit delivered. Under hourly billing, revenue was a reasonable proxy for profit, but AI severs that relationship: under alternative fee arrangements, a partner using AI can deliver a matter more cheaply while lifting the firm’s margin substantially, yet a revenue-based system doesn’t reward this outcome. Firms with discretionary compensation structures hold an advantage, since they can directly reward margin expansion. Boards can act now by using year-end partner questionnaires to probe AI use and pricing conversations; the friction partners report maps onto where transformation is stalling.

A further challenge concerns intellectual property: as firms encode partner expertise into reusable AI workflows, they convert personal, portable knowledge into firm-owned assets. Boards should consider how partners are compensated for this and factor it into lateral hiring evaluations.

Partner behaviour is the strongest predictor of technology adoption in professional services. Yet if partners are judged solely on short-term billable utilisation, they have no incentive to redesign workflows or champion new ways of working, because the compensation model suppresses the very behaviour adoption depends on.

5. DE&I Exposure

Is our AI implementation unintentionally penalising specific cohorts, or are we using it to democratise mentorship?

Female professionals face more than double the competence penalty of men when disclosing AI use, 13% versus 6 per cent; contributing to a documented 25 per cent global gap in AI adoption among women, driven by rational anticipation of heightened scrutiny of their AI use.

Female professionals often lack senior role models and report that competing demands inside and outside the workplace hinder upskilling. AI implementations that are not sensitive to these nuances risk unwittingly creating a talent divide.

Concerns about AI disrupting associate training assume traditional mentorship already works well, when in practice it falls unevenly across cohorts, especially across genders. AI has infinite time and patience: rather than an unexplained redline returned late at night, it can walk every associate through legal reasoning and drafting technique step by step. For boards, monitoring for performance evaluation bias post-AI implementation is critical, alongside mandating the use of AI to deliberately democratise mentorship.

The Strategic Imperative

Ultimately, these five considerations point to one decisive question: is your Chief People Officer actively shaping your AI transformation program? Too often, treating AI purely as a technology rollout leaves critical leaders out of the room. A resilient AI strategy requires a CPO who not only has a seat at the table, but who is deeply embedding in the program and equipped to evaluate how the features, design, and limitations of the technology itself impact your workforce.

The competitive dividing line in professional services will be drawn by firms that master the human architecture of AI adoption. Boards that look beyond high-level usage metrics to confront these five human capital questions will protect their talent, meet their regulatory obligations, and convert technological efficiency into lasting strategic advantage.

Adamantia Velonis is director of AI transformation and applied research at Legora, leading global change and adoption advisory for strategic law firms and corporate clients. Ada advises organisations on how to move from AI adoption to “hybrid intelligence,” designing workplaces where AI complements and augments human cognitive performance.