Is your BI ready for AI?

The self-assessment for BI teams: How well is your data prepared for AI queries? 12 questions, about 10 minutes.

Go straight to the checklist

How reliably an AI answers questions about metrics depends on two things: the approach the solution takes and the groundwork in your company.

The approach is about how the number in the answer is produced. If the software lets the AI write the query itself and recalculate the metric for every question, the results may sound plausible, but they are not reliably correct. Answers only become reliable when they return a verified metric whose calculation was defined in advance. The AI is then really only needed to understand the question correctly and, if necessary, ask follow-up questions until the question is unambiguous.

Verified metrics, in turn, require solid groundwork. This check shows you which starting path suits your data, how well your data and metrics are prepared, and what you can do next. The checklist also includes questions you should ask every vendor of an AI solution for BI.

The check assumes that you start with one business unit and then expand unit by unit. Answer each question as things stand today. If in doubt, choose the more cautious answer; your next steps will then fit better.

To begin: your reason

What is the main reason you are looking into AI for metrics?

One answer only. Choose your main reason.

Your answer does not count towards the result. It determines which arguments for your conversation with senior management you get at the end.

A. Where your data is stored

Where is the data of the business unit that will be the first to use a new AI solution?

One answer only.

B. State of your data

One answer per item: Applies · Partly applies · Does not apply

1.The data of the business unit that will be the first to use a new AI solution is fully contained in these systems.
2.Incorrect or incomplete data is caught before anyone works with the numbers.
3.Terms such as region, customer group or product line are defined consistently and maintained consistently in the data.

Test: Is a customer who appears in two systems assigned to the same region in both?

4.It is documented who is allowed to see which numbers.

C. Your metrics

These items do not have to be done before you start. Metrics are defined and verified business unit by business unit during the rollout. Whatever is already in place speeds up the start.

One answer per item: Applies · Partly applies · Does not apply

5.The same metric shows the same value in different reports.

Test: Compare last month’s revenue in three reports.

6.Your ten most important metrics are defined in writing, with source, calculation and filters.
7.For each of these metrics, it is clear who answers questions about its definition.

D. Questions for every vendor

These questions do not count towards the result. Take them into every conversation with a vendor.

  1. 1.

    Do the numbers come from verified metrics, or does the language model recalculate them for every question?

    Why this matters: Only if a verified metric is returned as the answer, with no calculation by the LLM, does the same question lead to the same result every time.

  2. 2.

    What happens if there is no verified metric for a question?

    Why this matters: A reliable solution says openly when there is no verified metric and only recalculates, in a traceable way, after you have agreed.

  3. 3.

    Can employees see at a glance whether each number is verified, and can they open its definition, source and filters directly from the answer?

    Why this matters: Only then can a number be backed up in a meeting without asking the BI team.

  4. 4.

    Can access rights be set per role, so that each person only sees the numbers they are allowed to see?

    Why this matters: Otherwise the chat shows numbers that this person is not authorized to see.

  5. 5.

    Can the solution run where your data is allowed to be, that is in the cloud, hosted in Germany or on-premises?

    Why this matters: Where data may be processed is often decided by data protection requirements, not by IT.

E. Your project

These three questions are optional and do not affect the result. They make your next steps more specific: you get a suggested scope for the start and notes on where to run the solution.

One answer per question.

1.Which business unit should be the first to work with AI queries?
2.How many people there should be asking questions at the start?
3.Where may an AI solution for your metrics run?

Please answer all questions in parts A to C first.

Book a demoContact