Skip to content

Data & Analytics

Business intelligence: numbers that can explain themselves

A number you can't trace is an opinion with a decimal point. Business intelligence starts from the decisions you make every month, defines the metrics behind them, and lets anyone walk a figure back to the records it came from.
Discuss your systemFollow a 19% drop to its cause ↓
Cost per booked job · March$142Definitionad spend ÷ jobs booked from adsAd spend, by monthJobs, by first sourceAd accounts · 3CRM · bookingsBy channelSearch$118Social$164Directory$231

Start from the decision, not the data

Most BI projects begin with “connect everything and see what we learn”. They end with forty charts and the same arguments as before. The useful ones begin with a decision you make on a cycle: where to spend on marketing next month, which location needs help, which services to push.

Each decision needs a few metrics, each metric needs a definition, and each definition needs an owner. Every metric is a stand-in for something you can't measure directly, so agreeing what it stands for matters more than the chart it ends up in.

Booked jobs fell 19%. Did they?

A March number that would start a difficult meeting, and the questions that explain it before anyone argues about it.

Step 1 of 4: All locations

  • All locationsFebruary 212March 171

Booked jobs: February 212, March 171. Down 19%.

The model knew which loads were incomplete, so the drill ended at a data gap instead of a staffing review. What's left, fewer web-form bookings at one location, is a real question with a smaller number attached.

Five questions worth answering every month

Most service businesses come back to these. Each one needs a specific metric, defined once.

  1. 01

    Which sources produce customers?

    Booked jobs and revenue by first source

    Not leads by source. A channel that sends many inquiries and few jobs looks best on the wrong chart.

    Watch for: sources credited for leads they didn't create.

  2. 02

    Where does work wait?

    Time between stages, by stage and team

    Inquiry to first reply, quote to decision, booked to completed. The slowest handoff is usually the cheapest fix.

    Watch for: averages that hide a long tail.

  3. 03

    Why is this month different?

    Change by location, service and source, with freshness

    A change is only worth explaining once the data behind it is complete.

    Watch for: comparing a full month with a partial one.

  4. 04

    Which customers come back?

    Repeat rate by first service and location

    Which first jobs lead to a second, and where.

    Watch for: counting a customer as new because their name was spelled differently.

  5. 05

    Which work makes money?

    Margin by service and job size

    Revenue says what's busy; margin says what's worth it.

    Watch for: costs that live in a different system from prices.

Calculated, then explained

AI is useful for the paragraph under the chart: summarising what moved and where. It isn't the calculation. The numbers are computed the same way every time from governed definitions; the explanation is generated from those numbers, marked as generated, and can be wrong in ways the calculation can't.

Calculated

Booked jobs, March
196
Change vs February
−8%
Definition
booked jobs v3 · data complete

Explanation · generated · check before acting

Most of the change is at Riverside, where web-form bookings fell from 22 to 15. Riverside call data for 4–21 March was missing and has been reloaded.

The figures are computed from the model; the explanation is written from them and labelled as such.

What business intelligence hands off

In

Decisions
what you decide, how often
Modeled data
definitions, grain, freshness
Access
who sees which numbers
Business intelligence
question → metric → view

Out

Metrics
defined, owned, versioned
Views
breakdowns that answer the question
Trace
from any number to its sources

Worth building when the same decisions keep coming back

Good fit

  • Recurring decisions depend on data spread across several systems.
  • Teams spend time reconciling reports instead of acting on them.
  • You can name the decisions the numbers are for.

Not the right fit

  • The underlying data and definitions aren't reliable yet. Build ingestion, validation and the model first.
  • The questions are one-off; an analyst with a spreadsheet will answer them faster.

Where business intelligence does its work

Next to business intelligence

Questions about business intelligence

Isn't business intelligence just dashboards?

Dashboards are one way to show it. The work is underneath: deciding which questions matter, defining the metrics, and making every number traceable. A dashboard on undefined data is a chart of an argument.

Which BI tool should we use?

For many businesses a tool like Metabase on a PostgreSQL model is enough; larger setups may justify a warehouse. We pick from your existing tools and constraints. The model and definitions carry over if the tool changes.

How do we know a number is up to date?

Every view shows when its data last loaded and whether any source was incomplete for the period. A comparison against a partial period is flagged before anyone acts on it.

Can we ask questions in plain English?

Yes, once the metrics are defined. Natural-language questions are answered from the same governed definitions, and the answer shows which metric and filters it used. That's the AI Analytics solution.