What to define before buying BI

In 1 minute

A business intelligence (BI) tool can apply an agreed definition; it does not decide for you which one it is. If Finance and Sales calculate margin differently, the new tool will show both margins, better presented and faster. Before choosing one, it pays to write down four things for each metric that matters: the formula, where each piece of data comes from, what it is reconciled against and who approves it. With that, the purchase becomes a simpler decision. Without it, the implementation will have to sort it out along the way, with the licenses already paid.

Imagine a services chain with fifteen branches. It is fictitious, but it looks like many. Management has been asking for the same thing for a year: "I want to see margin by branch without waiting until the 10th."

The team evaluates three BI tools. The demos are flawless: maps, filters, charts that move. One is chosen, the licenses are bought and in six weeks there are dashboards.

At the first meeting, a regional director asks why her margin on the dashboard is four points lower than in the Finance report. Nobody can answer on the spot. A week later the answer turns up: the dashboard spreads the head-office rent across the branches; the Finance report does not. Both figures were "right". They measured different things under the same name.

This article is for whoever is about to approve the purchase of a BI tool, or has already bought one and does not believe it. It makes one argument: first you write the definition of each metric; then you choose what to show it with. If you already know who answers for each piece of data, this is the next step.

The problem is not in the chart

When a company grows, each area builds its reports with what it has at hand. Over the years, the same word —"sale", "active customer", "margin"— ends up with several formulas. Nobody did it on purpose: each definition made sense to whoever wrote it.

A BI tool speeds up what already exists. If there is a single definition, it spreads it across the whole company. If there are three, it spreads all three, and now each one has a chart behind it.

The industry has been working on this problem for decades in the data warehouse world. Ralph Kimball called conformed dimensions those that are defined once, together with the business's data governance representatives, and reused across the processes that share them; according to the Kimball Group, they deliver "analytic consistency" and reduce future cost, because the wheel is not re-created each time [1]. Newer tools say the same in other words: the documentation of MetricFlow, from dbt, proposes defining the metric once and having every query use it the same way, under the principle of not repeating yourself [2].

The lesson does not depend on the tool: a metric shared by several areas is defined in one place, with one current version, and everyone consumes it the same way. An area's own metrics can live alongside it if they carry another name and are not presented as the company's figure.

The four things you write for each metric

You do not need to do it for two hundred metrics. Start with the ones that change a decision today or generate the most discussion in the meeting.

1. The formula, in words

"Gross margin = net sales − cost of goods sold", and then what really matters: do net sales include last month's returns? Does the cost include freight? Is the head-office rent spread across branches? If the answer changes depending on who writes it, the metric is not defined yet.

Also include how it is aggregated. Not all figures can be added up: Microsoft's guidance for Power BI gives the example of a unit price, which can be summarized with certain functions but must not be summed [3]. An amount is summed; a percentage is recalculated with its numerator and its denominator (the company's margin is not the sum of the branches' margins); a balance is summed across accounts at the same date, but not across dates. Write the rule for each metric and each cut that matters.

2. Where each piece of data comes from

For each term in the formula: from which system, from which field and with what date cut-off. "Net sales: invoicing system, stamped invoices minus credit notes, by issue date." This line makes visible, before the meeting, when two areas start from different sources or rules.

3. What it is reconciled against

A metric without a check is an opinion with decimals. Write which official figure it is compared with —the month's invoiced total, the accounting balance, the closing inventory— and what difference is accepted. If the figure does not add up, it should be noticed before the meeting, not in it.

4. Who approves it

A person with a name approves the definition and any change. When Sales wants to measure margin "before special discounts", the existing formula is not changed: another metric is created, with another name and its own owner.

The matrix that orders the purchase

Kimball proposes a simple tool for planning: the matrix of processes and dimensions (the bus matrix). The rows are the business processes (invoicing, collecting, buying, storing); the columns, the dimensions they are analyzed by (customer, product, branch, date). A mark in a cell says that process is analyzed by that dimension. According to the Kimball Group, the matrix also serves to prioritize projects together with management [1].

For a committee about to buy BI, a simple version of this matrix answers three questions before seeing a single demo:

• Which processes have to be visible first, and with which metrics.
• Which dimensions they share and therefore have to be defined the same for everyone (the Sales branch has to be the same branch as in Collections).
• Which systems they come from, and which ones do not have their data ready yet.

With the matrix in hand, the demos are evaluated differently: the question is no longer "how good does it look?", but "does it show this metric, with this definition, from this source?"

When it pays to do it first and when it does not

You can evaluate or try first if:

• Your reports come from a single system that already brings its definitions, and what you want is to see them better.
• The tool is for a single team that already agrees on how it measures.
• It is a limited trial, with authorized data, with no money decisions riding on it and with someone to validate the result.

Before connecting it to operations or using it for commissions, write down the priority metrics and reconciliations.

It pays to define first if:

• Two areas already arrive at the meeting with different figures for the same thing.
• The metric combines data from two or more systems.
• The dashboard will be used to pay commissions, assign budget or report to a board.
• You already bought a tool and nobody believes its numbers.

What we ruled out, and why

• "The tool comes with the industry's metrics." It comes with templates. Your company's formulas —what a sale is, when a customer stops being active— belong to your company, and someone has to write them.
• "Let each area define its metrics in its own dashboard." That is what produces three margins. Each area can have its own metrics, with their own name; shared ones are defined once.
• "Dashboards first, we reconcile later." A dashboard with a wrong number can lose the meeting's trust in the first week, and recovering it usually costs more than having reconciled it first.
• "Let's pick the most complete tool, just in case." More features do not resolve different definitions. With the matrix written down, a simpler tool often suffices, or the one your company already pays for.

Five questions for your next board meeting

1. Which are the five most argued-about metrics, and how many different definitions does each have today?
2. Is the formula of each one written down, with what it includes and what it does not, and who approved it?
3. Which system does each piece of data come from, and against which official figure is it reconciled?
4. Which processes and dimensions have to be visible first, and which share definitions?
5. Do we evaluate the tool against our written metrics, or against its demo?

If the answer to the first is "more than one" for several metrics, the purchase can wait a few weeks; the definitions cannot.

Where to start this week

Pick the most argued-about metric and write its sheet on one page:

• Name and, in one sentence, which decision it serves.
• Formula in words, with what it includes and what it does not, and how it is summed.
• Source of each term: system, field, date cut-off.
• Reconciliation: against which official figure and with what accepted difference.
• Owner who approves it.

Take the sheet to the next meeting and ask each area to say whether that is how it calculates it. The differences that appear are the pending work, and it pays to resolve them before the purchase.

What Hábil does

At Hábil we build custom systems and integrations that move your company's data, and we operate the platform where that data comes together.

If what you are looking for is a single version of your numbers, the proposal is in Hábil Lake: a single version of your numbers, reconciled against the system that issues the invoices, with the detail of how each figure was calculated. We choose together the three to five questions that cost you the most today. For example: margin by product line, sales by season, inventory that doesn't move or receivables that fall behind. For each one we put in writing which system each piece of data comes from, how it is calculated and what it is reconciled against. When we finish, you have your first dashboards working and the evidence that your numbers match the ones in your system, in your Excel or your dashboard.

The conversation can start with the sheet of your most argued-about metric, even if it is half done. With it we review together where the figures part ways and where it makes sense to start. Write to us on WhatsApp; to protect your information, do not share data about your customers or your systems through that channel: we will agree in the conversation on the channel for the case.

Request your first conversation, at no cost.

Consulted on 13 November 2026. Each reference carries its confidence: high (primary source read).