Role
Data analyst
Who answers when the board asks for a number that is in no report? Usually one person - and usually they first have to assemble it from five systems, each of which calls the same thing something different.
Intro call
Four questions we would start with
Who answers questions about data in your company?
If it is one person, they are the bottleneck of the whole organisation - and every holiday they take is a week without answers to questions that look simple.
How many times a month do you answer the same question?
If it is often, it means the answers are one-offs. They are produced, they go into an email and they vanish, and next time you assemble them again from the same sources.
How long does it take before a new question gets an answer?
At three days, the decision is either made without data or it waits. Both are expensive, and both look like a problem with the analyst although they are a problem with access to data.
What do you not do, even though you know it has to be done?
Usually: describing the terms, the documentation, putting the glossary in order. The tasks nobody orders, and without which every further answer costs as much as the first.
We take over the assembling, you keep the hard questions
Pulling data from the sources, watching its quality and handling repeat questions can be taken off the desk. The questions with no ready answer stay with the analyst - and that is the point.
- One term has one meaning. - Margin, client, order - each with a recorded definition visible next to the number. Two departments then stop arguing about who counts correctly.
- A repeat question stops being a question to a person. - It gets a ready summary that whoever asks reaches for themselves. The analyst goes back to the questions that genuinely require thinking.
- Missing data is shown as missing. - Not as a zero and not as an approximation. It is the only honest answer, and with financial decisions also the safest.
Scope
What can be taken off an analyst's desk
Twenty example tasks. Each one has a human approval point. Below is a selection, and the scope we fit to your process.
Getting data from the sources
- Regularly pulling data from the systems into one place
- A completeness check - detecting missing periods and sources
- An alert when data stops arriving or the structure of a source changes
- A history of changes in the configuration of the source systems
- Marking periods disturbed by a migration or a rollout
The glossary of terms
- A register of definitions with who owns each term
- Pointing out the places where the same term is calculated differently (settled by: the analyst)
- Documenting how a definition changed over time
- A summary of which reports and decisions rest on which definition
- Detecting reports using a withdrawn definition
Self-service reporting
- Summaries available to the people who used to ask the analyst
- A recurring report in a fixed form (conclusions: the analyst)
- A comparison of changes against the previous period
- A description of what a given report shows and what it does not
- Monitoring which reports are actually opened
One-off questions and data quality
- A register of questions asked and answers given
- Pointing out recurring questions, the candidates for a permanent summary
- Detecting outliers and likely errors in the data
- A list of records failing the quality rules (decision to correct: the data owner)
- A list of questions there was no way to answer, as material for what to start measuring
A day's work
Today
- The day is broken up by ad hoc questions arriving by email and chat
- Each one starts with pulling data from several systems
- When figures disagree you work out again which source is the right one
- You assemble the recurring reports by hand, always on the same days of the month
- You find out a source is broken while analysing an odd result
- The documentation exists in your head and stays there
With AI assistance
- Repeat questions have ready summaries that people reach for themselves
- The data is pulled before anyone asks
- The definition stands next to the number, so the argument about it never starts
- The recurring report is waiting, ready, you add the conclusions
- Data that stops arriving reports itself, before it spoils somebody's decision
- The glossary is a document, not your memory
- Numbers when they are needed
decisions are made on data rather than on a hunch
an answer stops requiring a queue to one person
- Independent readers
fewer questions to the analyst, more answers found directly
repeat summaries are available without them
- One meaning per term
two departments talk about the same thing in the same way
the definition is recorded and has an owner
- Resilience to source changes
a rollout in one system stops breaking reports quietly
a change in a source structure is detected rather than discovered
We start with the questions, not with building a warehouse.
- Step 1
We collect the questions for a month
Who asks, about what and how often. That list is the whole justification for the project - and it usually shows that a dozen or so questions account for most of the interruptions.
- Step 2
A glossary for the terms in those questions
Only the ones that actually come up. A glossary built complete from day one never gets finished.
- Step 3
The first self-service summaries
For the most frequent questions. We measure whether people use them - if they do not, it means they were asking about something other than we thought.
- Step 4
Data quality and source monitoring
Last, because it needs knowing which data is actually used. Watching the quality of everything cannot be sustained.
Common questions
Questions that come up at the first meeting
Does the analyst not become redundant once the data falls into the report by itself?
Preparing data is not analysis, it is the precondition for starting it - and today it eats most of the week. When it stops, the analyst starts answering questions instead of preparing to answer them. Experience says the number of questions put to them then grows, because people stop fearing that the answer will come in two weeks.
Does this fall under the AI Act?
Internal reporting is not a high-risk use. The line appears where the data concerns employees and serves to assess their work - separate rules apply then, and that is a conversation to have before such a summary is built, not after.
Still have a question that is not here? Book a call
Next step
How many decisions in your company are made without data because the answer would come too late?
We will show you what an analyst's work looks like when repeat questions have their own answers and terms have one meaning.