Most organisations know something is blocking progress with data and AI. The hard part is figuring out what.
The symptoms are usually obvious: reports that nobody trusts, pilots that never scale, tools that sit unused, or governance conversations that go in circles. But the cause is often unclear. Is the problem data quality? Workflow design? Tooling? Adoption? Governance? Or some combination that is hard to untangle?
That uncertainty is where most teams stay stuck. This diagnostic is a way to move from stuck to specific.
The friction map
Start by mapping where friction actually shows up, not where you assume it lives.
Ask each team that interacts with data or AI to answer three questions:
- What is the last thing you tried to do with data or AI that did not work as expected?
- Where did you stop or switch to a manual workaround?
- What information were you missing when you made the decision?
The answers reveal patterns. If most teams stop at the same point, you have a structural bottleneck. If they stop at different points, you have a coordination problem. Both are fixable, but the fix is different.
Common friction points include:
- analysts waiting on data access or provisioning
- teams building dashboards on metrics that nobody has formally owned
- AI prototypes that cannot connect to production data
- governance reviews that happen too late to influence the actual design
- tooling decisions made before the workflow is understood
The friction map does not solve these problems. It makes them visible, which is the prerequisite for solving any of them.
The system risk view
Once you can see where friction concentrates, assess the risk.
Not every friction point is equally urgent. Some are inconveniences. Some are structural risks that will compound over time. The system risk view separates the two.
Look for three signals:
Single points of failure. If one team, one pipeline, or one data source blocks multiple downstream workflows, that is a system risk. It may work today, but the blast radius of a failure is large.
Unseen dependencies. If teams are making decisions based on data they do not understand the lineage of, that is a risk. The numbers may look right, but the confidence is borrowed, not earned.
Adoption gaps. If tools or platforms are technically available but not being used, that is not a tooling problem. It is usually a workflow problem. The tool does not fit the way people actually work, or it was introduced before the workflow was designed.
These risks do not resolve on their own. They tend to get worse under pressure, which is exactly when you do not want to discover them.
The opportunity list
The diagnostic is not only about problems. It also surfaces opportunities that are hiding in plain sight.
After mapping friction and risk, ask:
- Which workflows are already working well that could be extended or automated further?
- Where is manual work being repeated that a model or pipeline could handle?
- Which teams have strong data practices that could serve as templates for others?
The best opportunities tend to be in the overlap between low friction and high value. They are not the most glamorous projects. They are the ones where the groundwork is already in place and the impact is measurable.
Examples include:
- extending an existing trusted pipeline to cover a new use case
- automating a manual classification step that already follows clear rules
- standardising metric definitions that multiple teams already use informally
- connecting an AI drafting step into a review workflow that already has human oversight
These are not moonshots. They are practical moves that compound.
The recommended next step
The diagnostic should produce one clear next step, not a full roadmap.
If friction is concentrated in data access and quality, the next step is to fix the foundation: data cataloguing, metric ownership, and pipeline reliability before adding more tools or models on top.
If friction is concentrated in workflow design, the next step is to map the actual operating process before selecting or deploying technology. The tool should follow the workflow, not the other way around.
If friction is concentrated in governance, the next step is to simplify the governance model so it can operate at the speed the business needs. Governance that blocks everything is not governance. It is a bottleneck wearing a policy hat.
If friction is concentrated in adoption, the next step is to embed the capability into the workflow people already use, rather than asking them to adopt a separate AI experience.
The diagnostic does not tell you which answer is correct. It tells you which question to answer first.
Why this matters
Most data and AI programmes do not fail because the technology is wrong. They fail because the problem was misdiagnosed.
A data quality investment will not fix a workflow design problem. A governance framework will not fix an adoption problem. A new tool will not fix a coordination problem.
The clarity diagnostic is about asking better questions before committing to bigger bets. Once you know where the real blockage is, the next step is usually obvious. The hard part is getting honest about what is actually in the way.