AI is only as good as the context you give it
By Sascha Cutura

Most conversations about AI focus on models, tools and features. Faster models. Better prompts. New releases every few weeks. In practice, this is rarely where things break.
When AI disappoints in a real organisation, it is almost never because the model cannot do the task. The weaker point is context. AI does not understand your business by default. It does not know your history with customers, how your teams actually work, or why certain numbers look the way they do. Without that context, even a very capable system produces generic answers that sound confident and miss the point.
What context actually includes
Context is everything that sits around the data.
It includes how metrics are defined, how decisions were made in the past, what changed last quarter, and which exceptions matter. It includes the assumptions that never made it into a dashboard and the operational realities that sit behind the numbers.
Two companies can look at the same revenue chart and tell very different stories. One sees growth. The other knows a single contract skewed the figures. AI can only tell the right story if that context is available to it.
The same is true of day-to-day work. An assistant that drafts a customer reply without knowing the account history, or a forecast that cannot see a one-off contract, will sound sure of itself. The people who have to use the output will not be.
The hidden data problem
Most organisations already have plenty of data. The issue is fragmentation.
Finance works from one set of numbers. Sales tracks pipeline in a different system. Operations has its own view of delivery and capacity. Each area makes sense on its own, and together they do not form a single, coherent picture.
When AI is layered on top of this, it has no stable foundation. It tries to connect dots that were never aligned in the first place. That is why outputs often feel vague, overly cautious, or just slightly wrong.
Gartner has found that 63% of organisations either do not have, or are unsure whether they have, the right data management practices for AI. It predicts that through 2026, organisations will abandon 60% of AI projects unsupported by AI-ready data.
At operational level, AI-ready data is unglamorous. One customer record in place of three. An agreed definition of an active client. Clear rules on which data a tool may access. When sales and finance report different revenue figures, an AI summary will pick one and present it with complete confidence, and trust in the tool tends to go within weeks. Whoever owns AI has to own this problem as well, or make sure someone with the authority to fix it does.
Structure first, then the tool does more useful work
Good outcomes start with good structure: clear ownership of data, shared definitions across teams, and connections between systems that reflect how the business actually runs. It also means keeping history over time, rather than relying on static snapshots in reports.
Once that foundation exists, the role of AI changes. It can explain why something happened, rather than only describing what happened. It can support decisions instead of simply responding to questions.
Models are becoming easier to obtain. Everyone has access to similar technology. What is much harder to copy is your internal context: how you collect data, how you structure it, and how much institutional knowledge is embedded in it.
This work pays off even before a new tool is introduced. Teams that agree what a customer is, what "active" means, and which system is the source of truth already make better decisions. AI then has something solid to read.
If you are planning a first implementation, this is part of the readiness work, alongside process understanding and a short skills plan. A model placed on top of conflicting numbers will not settle the argument. It will inherit it.



