Somerset Bridge protected the hour because the licence was not the problem

An afternoon that is already spoken for will beat a skill that has no appointment. The deadline keeps its chair. The new tool waits in a tab, and at the end of the week the organisation can still say it has rolled something out.

aibl's account of Somerset Bridge Group, dated 30 October 2026, starts from the other end of that week. The motor insurer's first conversations about AI were about buying hardware, tools and licences. Then it surveyed more than 220 colleagues, to get a baseline, and found very different starting points across the business. Katherine Hawkins-Jones, chief people officer, and Ceri Sharples, learning and development manager, speaking at aiblLIVE London, named five barriers: confidence, capability, time, leadership and governance.

Time was the largest. Ordinary delivery got in the way of experimenting, and learning had to compete with the work that already had a due date. That is a plainer diagnosis than a speech about culture. If the day is full, a new skill is whatever is left, and whatever is left is not a programme.

The page names five barriers and then spends itself on one of them. Confidence, capability, leadership and governance stay on the list. They are not each given a programme in this account. Time is. That emphasis is worth keeping, because a plan that answers all five with the same session has already left the diagnosis. A person who does not yet trust the tool, and a person who has no hour in which to try it, are not the same obstacle. Hawkins-Jones's ranking says which one was stopping the work.

Hawkins-Jones's line is the one the rest of the account hangs on. "It was clear that we didn't have a software tool or development problem, we had an adoption problem." The page calls the consequence an organisational change, not a purchase. It does not describe a product that failed in a pilot. It describes a company that had been about to solve the wrong thing. A team that answers "adoption" with another licence has not yet used her sentence.

She also names the habit they had to stop calling normal. Too much control from the top, she said, can suppress ideas and slow the path to market, and that was typically how Somerset Bridge operated. The response was to run both directions at once, from the top and from the work, rather than to pick the one that felt orderly. A leaders-only briefing, or a volunteer channel with no leader in it, is each a half of what she says they refused.

At the top, 16 executives went through an AI for Leaders programme so they would be seen learning, experimenting and failing. The page gives the need a face. One executive kept asking who this new employee, Claude, was, and why everybody was talking about him. That is not a joke the account leaves hanging. It is the exhibit for why a leader who will not be seen struggling cannot set the terms for anyone else. The page does not say the 16 then passed an assessment. Being seen is the standard it states. A completion certificate would be a different standard, and this account does not offer one.

What those leaders were then for is narrower than a sponsorship logo. They were there to set guardrails, and, as the page puts the emphasis, to give people permission to step away from the day and experiment. An AI steering group sets the direction and helps decide which opportunities to pursue. Permission without a place a pursuit can be refused is enthusiasm. The steering group is that place, as far as the account describes it. The criteria it uses are not published here. A team that copies the group and invents the criteria has left the page.

Across the business, the AI Academy gives people protected time to experiment during the working week, mainly with Copilot, and the page is explicit that this is not an add-on after the job. Somerset Bridge also runs learning sessions every fortnight, and it sets aside time to experiment once a quarter. Those are three different decisions about the calendar, not one slogan repeated. Ambassadors and communities are how practical needs and credible uses get named. The page does not count the ambassadors, and it does not say how a person becomes one. A title without the hour is the add-on they said they were done with.

An AI forum connects the two directions. The promptathons, in Bristol and Newcastle, are where the account says the learning finally met the work. Earlier learning activity had not been tied to the problems people actually had. In the rooms, cross-functional teams were asked what drove them mad, and they worked on genuine problems, live, with AI champions beside them. The page says this surfaced problems the company would never have thought to put into a change programme. Two events produced 20 ideas. An idea count is a count of ideas. It is not a list of changes anyone is running on Monday, and the account does not pretend otherwise.

The only hours figure has a fence around it, and the fence is the useful part. At Bristol, participants estimated that nine of the ideas could save about 5,000 hours. The page says these were self-assessed estimates, made on the day, and that the actual benefit was still to be established. Nine is not twenty. Newcastle is not given an hours number at all. A later retelling that turns 5,000 into a saving the business has taken is changing the tense. The account keeps it as an estimate from the room.

The two directions meet when someone decides what is allowed to continue. Somerset Bridge has recently introduced what it calls kill or keep it, looking at ideas from a commercial point of view and not only from the attachment people have formed. Hawkins-Jones says the hard part out loud. "It's really hard to convince people to kill something that they've become emotionally attached to." Recently introduced is not a track record. The page does not say how many ideas have been stopped, or what commercial test is on the page in front of the group. The existence of the phrase is the fact. A graveyard of named bets would be a further fact, and it is not in this account.

She names the step that has not happened yet, and she names it as a shift in words. The next move is from "our metrics to our measurements": from what they can count now, ideas generated and hours estimated, to what the business actually banks. Metrics, on that wording, are the activity. Measurements are the consequence. A company that stops at the promptathon has the first and is still waiting on the second. That is not a criticism the page hides. It is the sentence she uses for what comes next.

Banked, in that sentence, is doing quiet work. It is not another workshop score, and it is not the 20. It is whatever the business can show it kept after the estimate left the room: a cost that moved, a queue that shortened, a decision a customer felt. The account does not give an example of one. Naming the absence is the discipline. An L&D lead who reports the promptathon as the measurement has used her word for the thing she says they have not reached.

The same discipline applies to the people who are not in the story. The survey is more than 220 colleagues. The executives are 16. The ideas are 20, and nine of them carry the hours estimate. Nowhere does the page say how large the workforce is, so none of those counts can be turned into a share of the company. A percentage added later would be a calculation this account refuses to make. The steering group, the forum, the ambassadors and the champions are roles, not headcounts. Copying the chart without the hour, and without a commercial refusal, reproduces the names and leaves out the two moves that make the names matter.

What an L&D lead can lift, without borrowing a confidence score the account does not print, is the order. They baselined more than 220 people before they argued about the tool. They put three kinds of time inside the week and the quarter, and they said the time was not an add-on. They pointed a cross-functional room at an irritation the business already had. They then built a commercial refusal, and they said the hours on the wall were not yet money. A dashboard of licences, read against that order, answers a question they had already retired.

1. Put the spread on the table before the purchase gets the credit. More than 220 colleagues were surveyed, and the starting points were not the same. A single licence report cannot show that, and the account begins by refusing to let it.

2. Make the time three decisions, not a slogan. Protected time in the working week, mainly on Copilot, and not an add-on to the day job. A session every fortnight. A block once a quarter. Any one of those, stacked on a day that is already full, is the add-on the page says they left behind.

3. Start from an irritation the team already has. Bristol and Newcastle asked what drove people mad, because earlier learning had not been tied to those problems. Twenty ideas is the yield. It is not a release list, and the account does not supply one.

4. Leave the 5,000 hours in the tense they were spoken in. Nine ideas at Bristol, estimated on the day, by the people in the room, with the benefit still to be established. Kill or keep it is the commercial look she says is hard precisely because people get attached. What the business banks is the measurement she says they do not have yet.

A productivity session is a place to take one irritation a team already has, put the time for it inside the week, and write down what would count as banked before an hours estimate is treated as a result.