16 Great Chapel St
London W1F 8FL
United Kingdom
We have watched the same pattern play out in too many client use cases. A leadership team agrees a platform. The contract is signed. Training decks appear. Six months later the usage graph is flat. People have found ways to work around the shiny thing. The conclusion is predictable. Buy a different tool. The real problem sits elsewhere.
Technology does not adopt itself. Culture does. The firms that win with AI do not have the newest stack. They have habits that let people change how work actually gets done. They treat AI as a shift in operating practice rather than a shopping trip. That shift begins with what people believe about quality, ownership, risk and pace. It shows up in how decisions are made when the pressure is on.
What is really shifting
AI is not a bolt-on. It changes the rhythm of routine work. It compresses cycle times. It exposes messy processes that used to hide behind manual effort. It increases the blast radius of small mistakes. It creates new judgement calls that no policy has named yet. In that environment the cultural defaults of an organisation either protect progress or poison it.
If your culture rewards heroics over clarity, AI will amplify chaos. If your teams are afraid to surface broken steps, automation will lock in those breaks. If data ownership is fuzzy, every pilot will stall on permissions or quality. The inverse is also true. Where teams share facts openly and fix the roots before they scale, AI compounds value quietly.
Why culture beats technology
A tool can be perfect on paper yet fail on contact with the real day. Culture is the glue that holds a change together once the consultants leave. I look for four cultural foundations before I expect any AI investment to pay back.
Clear purpose: People need to know which business outcomes matter most. Save three hours per case. Cut error rates on invoices. Reduce time to complete onboarding. Without a crisp outcome, AI becomes theatre.
Ownership with teeth: Every critical process needs a named owner. Not a committee. Someone with authority to simplify steps, to set data standards, to decide what stays human. If ownership is vague, decisions drift.
Safety to speak plain truth: Teams must be able to say this process is broken without fear. They must be able to ask basic questions about risk. They must be able to show the ugly logs in daylight. Silence kills adoption.
Incentives that back the change: If performance reviews still reward the old path, the old path will win. Tie recognition to use of the new workflow. Retire the old metrics. People follow the scoreboard.
When those foundations are weak, more technology only increases friction.
Two stories I keep returning to
A service operation tried to roll out a copilot for case handling. The model worked in tests. In live use it tripped over inconsistent categories and missing notes. The team had learned to keep their own shadow tags to survive a cluttered system. The rollout ignored that lived workaround. Adoption flatlined. We paused the build. Standardised the categories with the team. Retired old fields. Wrote a crisp rule for note quality with examples. The same model then delivered measurable gains because the culture allowed the work to get tidier first.
A compliance unit feared that AI would create unmanageable risk. They were right to be concerned. We gave them a better job. Co-design the controls so safe experiments could start quickly. Define how to log decisions. Set thresholds for human review. Agree a simple model of harm and detect it early. Their posture shifted from gatekeeper to co-author. Adoption accelerated because risk had a voice at the table from day one.
Practical moves leaders can make this quarter
1. Set one narrative
State the business goal in thirteen words or fewer. Repeat it until people can say it back. Tie every initiative to that goal. Kill work that does not move it.
2. Run adoption as a product
Assign a product owner for the workflow, not just for the tool. Measure daily usage. Track where people drop out. Interview users who avoid the path. Fix friction weekly. Treat adoption work as delivery work.
3. Write the rules of the road with risk
Create a short AI use policy that answers three questions. What is allowed. What is not. How to get to yes. Add logging, review, and escalation steps that busy teams can follow. If a policy cannot be used in a rush, it will be ignored.
4. Train on live work
Use the real cases people handle. Pair training with floor support in the first fortnight. Add job aids inside the workflow rather than in a shared drive. End each session with two asks. What slowed you down today. What would help tomorrow. Act on the answers.
5. Change the incentives
Move at least one KPI to reflect the new world. Reward improvements in cycle time with quality held steady. Reward clean data submission. Reward removal of redundant steps. People are experts at reading what leadership values.
6. Protect time to tidy
Create a regular slot for process hygiene. One hour a week. No slides. Walk the path together. Remove one small source of friction each time. Tiny cuts heal faster than surgery twice a year.
7. Put pilots on production rails
From day one use the same logging, access patterns, and support routes you expect at scale. Pilots should prove more than model quality. They should prove you can run the thing in daylight.
The mindset shifts that matter
From secrecy to daylight. Share the messy reality early. Show the metrics, the errors, the audit findings. People trust what they can see. Trust drives use.
From heroics to systems. Stop relying on champions to push adoption. Build a backbone that makes the right path the easy path. Champions burn out. Systems endure.
From suspicion to stewardship. Move the conversation with risk from will you let us to how do we run this safely together. That language change does more than any memo.
From knowledge hoarding to coaching. Reward people who teach others. Give experienced staff time to coach. Their influence matters more than any vendor handbook.
Red flags you can spot in a week
- AI is discussed as brand rather than operations
- The only metrics on the report are pilots completed or licences bought
- No one can name the owner of data quality for the target process
- Training is scheduled before workflow changes are agreed
- Risk and compliance are briefed in the last week before go live
- Teams keep private spreadsheets to survive a broken path
Each flag signals culture misalignment. Fix the misalignment before you scale.
How to rebuild culture while you deliver
Culture shifts when people experience a different way of working. Not when they read a manifesto. Use your first AI use cases as culture practice.
- Make the team that runs the work the design authority
- Publish a one page charter for the use case that lists outcome, controls, and owners
- Hold a weekly open review where anyone can see progress and failure
- Remove one legacy report or form for each new feature you add
- Celebrate the first person who raises an ugly truth that saves time later
These moves look small. They build trust at a pace that technology alone cannot. They show people that leadership will support change when it gets uncomfortable.
The risks we should name out loud
Change fatigue is real. If every month brings a new tool with no visible relief, people will tune out. Fight this by killing work that the new system replaces. Put the removal on the roadmap. Shadow usage will appear if controls are unclear. People will paste sensitive data into public tools if you do not provide a safe route. Give them a clear path. Make it fast enough to beat temptation. Automation can fossilise bad practice. If you apply AI to a broken process, you will scale the break. Slow down to simplify first. Speed returns once the path is clean. The narrative can drift into job fear. Be honest about what will change. Name the tasks that will reduce. Name the judgement work that will grow. Support people to move across that line.
Where this is heading
Over the next two years the technical options will converge. Most firms will have access to similar models and services. The gap will widen on culture. The winners will be the organisations that made it safe to tell the truth about their processes. They will have owners who act. They will have risk partners who design. They will have leaders who protect time to tidy. Their people will move at a steady pace without drama because the system around them makes progress normal.
AI is not a parade of tools. It is a choice about how you want your people to work together. Culture sets that choice long before procurement signs anything. Choose clarity. Choose daylight. Choose ownership with teeth. The technology will still matter. Culture will decide whether it matters in your organisation.
