One of the most common things we hear from organisations is this:
our data is not yet in good enough shape.
The concern is completely understandable.
There is a lot of information, but it is scattered. Some of it is outdated, some overlapping, some difficult to find. At the same time, organisations collect and store a great deal of data that has little practical relevance to the work itself.
The information that actually matters, meanwhile, easily gets lost in the mass. Or it exists, but people do not know how to use it at the right moment.
So it is worth stating one thing clearly:
in practice, no organisation has perfect data.
Perfect data is therefore a poor starting criterion for building AI agents.
A better starting point is this:
we have identified the challenge, we understand what information is needed to solve it, and we are able to evaluate the solution against the goal.
This is where role-based AI agents have particular value.
They do not require an organisation's entire data estate to be brought into perfect shape. They force you to identify what information is essential for a specific task and focus on that.
Not all data needs fixing. Only the essential information needs to be brought into working order.
Not all data needs fixing
One of the most common misconceptions is that an AI agent initiative first requires a large-scale data harmonisation effort.
In reality, that is usually not the case.
A well-designed agent initiative does not begin with reviewing the entire mass of data, but with a much more practical question:
what is the challenge we want to solve?
Only then does it make sense to ask:
- what information is required to perform this task successfully
- where that information exists today
- what is reliable and usable
- what is missing
- what information is critical for this specific use case
This framing matters because it shifts the focus from the volume of data to its relevance.
Not all data is valuable. And not all valuable information is equally visible inside an organisation.
For example, we have a case where ten years of WhatsApp conversations have proven to be a highly valuable source of information. Not because they were structured or curated data, but because they contained task-relevant practical knowledge, decision logic, and tacit understanding that could not be found as vividly in any formal document.
It is a useful reminder that the most valuable source of information does not always look like good data.
A role-based agent makes the essential information visible
When an agent is built for a specific role or task, it becomes necessary to make visible what information actually affects successful performance.
This is where a great deal of value is created.
An AI agent project is not only about building a new interface to information. It is also about identifying and defining the knowledge base that allows that task to be carried out better.
This is often a very practical process.
What does a maintenance expert need in a problem situation? What does a manager need to support decision-making? What does HR need in order to guide people correctly? What information helps sales act consistently?
When you start breaking this down, information quality issues also become visible very quickly.
Is the same instruction available in several different versions? Is a critical step missing? Does the information exist, but in such a fragmented form that it cannot be used reliably? Is there a clear owner for that information?
In this way, the agent does not only use information.
It reveals the actual condition of the information that matters for the task.
The right starting point is not perfect data, but a clear goal
So the most important question is usually not whether the data is perfect.
A more important question is this:
have we identified the challenge precisely enough?
If the challenge has been identified well, it becomes possible to define what information is needed to solve it.
And once that information is clearly scoped, the solution can be evaluated against the goal.
This matters because the value of an AI agent should not be assessed by how much data it can process.
Its value comes from whether it helps solve an identified problem in a measurable way.
Does the work get done faster? Are errors reduced? Does information become easier to find? Does the solution support better decisions? Is onboarding time shortened? Does dependency on individuals' tacit knowledge decrease?
When the solution is interpreted against the goal, conversations about data quality also become more useful.
At that point, the question is no longer simply whether the data is generally in good shape.
The question becomes whether the information that is critical for the task is in good enough shape for the desired impact to be achieved.
Using the agent can also keep information in shape
One of the most underrated benefits of role-based agents is that they do not only use information, they can also help keep it relevant.
When an agent is used in everyday work, it quickly becomes visible where information needs correction, clarification, or updating. Use exposes gaps in a way that would otherwise take much longer to notice.
At best, this means that the information that matters for the task stays current, relevant, and more accurate without requiring a heavy separate maintenance model.
In other words, the agent does not only use the knowledge base.
It can also support its continuous upkeep with minimal effort.
This matters because, over time, it is the maintenance of relevant information that determines whether the solution remains genuinely useful.
Waiting for perfection often slows things down
That is why I do not believe organisations should wait for perfect data before they begin.
What matters far more is identifying the right challenge, scoping the essential information, and evaluating impact against a clear goal.
In many cases, this is exactly where the real work of getting data into shape begins.
Not as a massive clean-up project, but as the process of identifying, defining, and maintaining the information that matters for work and decision-making.
So no, perfect data is usually not required.
But a good AI agent project can be exactly the practical mechanism that helps an organisation separate the essential from the non-essential and gradually bring critical information into better shape.
In the next newsletter, we will look at one of the most practical AI leadership decisions: buy or build?
