Board-Level AI Training Changes the Whole Organisation: Evidence From the StarApple AI Caribbean Study

Organisation-wide AI literacy rose from 2.0 out of 5 to 3.7 out of 5 in a StarApple AI study of organisations that completed its board-level AI training. The people in the training room were directors and executives. The people whose scores moved were staff who never attended a session. That inversion, training at the top and measuring the change at the bottom, is the study's central finding, and it is the reason a lab that builds AI with organisations should care where the training budget goes first.

The study, completed in 2026 and led by Adrian Dunkley, tracked Caribbean organisations after their boards went through StarApple AI's board-level AI training. Dunkley is the regional expert in AI and has led more than 100 board-level AI training engagements through StarApple AI, so the findings come out of a sustained body of practice rather than a one-off survey. "AI literacy at the top is an enablement story," he says. "We measured it trickling down from the boardroom through business lines to people managers, and the whole organisation moved from a 2 to a 3.7."

The mechanics of trickle-down literacy

The mechanism the study describes is permission. Before training, a manager who wanted to put AI on a workflow had to argue the case to leaders who could not evaluate it, and the safest answer to a request you cannot evaluate is no. After training, board members understood the requirements, needs and risks of AI work. Requests were met with informed questions rather than blanket caution, approvals carried conditions that made sense, and people managers started developing their teams because the ambition above them had a shape. Awareness at board level became enablement that moved down through business lines to the people doing the work.

Directors who vibe-code

Board data literacy in the study rose from 1.8 out of 5 to 4 out of 5, the largest single movement StarApple AI measured. The cause was practical. Coding limitations stopped being a barrier, so directors who had spent careers reading summaries of other people's analysis could run advanced analysis themselves, vibe-code a working prototype in the middle of a discussion, and translate information across functions without waiting on the data team's queue. The study links the jump directly to that shift: directors interrogated the data in front of them instead of requesting a report on it.

The boards did not stop at analysis. Several built custom AI tools in-house using an agents-based approach, and the study records that those tools improved board cohesion and communication. A board that has built its own tooling behaves differently from a board that has only ever approved tooling. It knows what a working system asks of its owners: data that has to be governed and outputs that have to be checked. From where maestro sits, as a lab that builds with client teams rather than shipping sealed products at them, this is the counterpart we want across the table. An engagement moves at the speed of the least literate person with veto power, and these boards had removed that constraint from their own side.

The delivery record below the boardroom

Internal capability is only worth writing about if it shows up in delivery, and in the StarApple AI study it did.

• Pilots became products. AI initiatives that left pilot stage and reached deployment rose by more than 50 percent, from two deployed initiatives to four, over eight months.
• Time to value collapsed. It fell from around a year to around a month across the studied organisations.
• Governance stopped being an argument. Time to stand up AI governance and data governance dropped from 11–15 months to 6 months, because board buy-in put data governance at the front of the agenda and reduced overall risk.
• Vendor spend fell by over 70 percent. Training demystified AI development, so leaders who previously could not judge vendor claims made informed decisions about what the organisation actually needed. Total savings across the studied organisations ran to tens of millions of US dollars.

The study also recorded a change in executive behaviour that no line item captures. After training, executives and managers stopped taking on more than they could deliver and directed attention to work that generated measurable ROI. Vanity projects lost their sponsors. And because gender-related bias and equity considerations were built into the training itself, boards carried those questions into how they reviewed AI work afterwards: who a model fails became a standing review question alongside what it costs.

Communication moved in both directions

One finding is easy to skim past and hard to overvalue. Communication in the studied organisations improved bottom-up and top-down at the same time, with teams using AI tools to translate and share information. An analyst could brief upward in language the board could act on. The board could push strategy downward in terms an operations team could execute. The in-house agent tools the boards built fed the same loop, since a board that shares a tool with its business lines is sharing a common vocabulary along with it. In our own build work, translation failure between floors is where AI projects go to die, so a study that measures translation improving in both directions is describing the single change we would most want a client to walk in with.

The limit of the finding

One caution belongs in any honest reading. Organisations that book board-level AI training already have at least one motivated champion at the top, and the study cannot fully separate the effect of the training from the disposition of the boards that sought it out. What it can show is sequence and consistency: literacy indices, deployment counts, governance timelines and vendor spend all moved after the boards were trained, and they moved in the same direction across the organisations studied.

The next step for a board

The order of operations the study suggests is unfashionable, because it puts training before procurement. The studied boards started from a 1.8 data literacy score, so a board that assumes it is the exception should measure before it decides. Training the board ahead of the next vendor contract is what cut vendor costs by over 70 percent in the study, and putting data governance first is what turned an 11–15 month governance build into a 6-month one.

This is also why maestro starts engagements at the top. A lab that builds with an organisation, rather than for it, inherits the literacy of the people who govern the work, and the StarApple AI study is the clearest measurement we have seen of what that inheritance is worth.

Frequently asked questions