One dashboard number, defined once and agreed by everyone

Agree the definitions before building anything

An AI business dashboard is one operating picture assembled automatically from the systems you already run. Sales, orders, enquiries and cash pulled into a single view, with every number defined once so two reports cannot disagree in a meeting. The AI part summarises what changed and why it might have changed. The definitions are the part that decides whether anyone trusts the dashboard six months in.

Ask three people in the same business what counts as a lead and you will often get three answers. One counts every form submission. One counts the ones sales accepted. One counts the ones that had a phone number attached. Every dashboard built on top of that disagreement inherits it.

Writing the definitions down is the least popular part of the work and the part that determines whether anyone trusts the output. A metric with one written definition and one place it is calculated is worth more than ten charts.

• One written definition per metric, agreed by the people who use it
• One place each number is calculated, so two reports cannot disagree
• A stated refresh time, so nobody argues about whether the data is current
• A named owner per metric, because unowned numbers rot

What the daily picture should contain

A useful operating dashboard is short. It answers what happened, whether that is normal, and what needs attention today. Anything else is analysis and belongs somewhere a person goes deliberately rather than somewhere they glance.

The test is whether someone can look at it for fifteen seconds and know whether to change their plan. A dashboard that requires interpretation is a report, and reports get read once.

• Volume against the same day last week, since week-on-week beats month-to-date for spotting change
• The one or two numbers that lead the rest, usually enquiries and response time
• Anything currently failing, stated as a failure rather than as a dip in a line
• What is stuck, meaning work that entered a stage and did not leave it

Where AI adds something and where a chart is enough

AI is useful for reading unstructured material at volume. Summarising what enquiries actually said this week, clustering support conversations into themes, and writing the paragraph of commentary that a human would otherwise write every Monday morning.

It is not useful for arithmetic that a query already does correctly. A dashboard that routes a simple count through a language model has added cost, latency and a new way to be wrong. The arithmetic should be deterministic and the language should be the part that is generated.

Generate the sentence, not the number. A number that is sometimes wrong is worse than no dashboard, because people act on it.

Delivery beats destination

Most dashboards are visited for two weeks and then forgotten. The ones that survive arrive where people already are: a message at the start of the day, a weekly summary, an alert when something crosses a threshold that was agreed in advance.

The dashboard remains, for the moment somebody wants to look deeper. The daily habit is carried by the message, not by the link.

How reporting work gets quoted

Reporting is a project first and an operation afterwards. The project ends when the definitions are agreed and the first dashboard is producing numbers people trust. The operation is what keeps it true, because a source system changes a field and a dashboard nobody maintains carries on displaying a number that stopped being correct weeks ago.

What moves the figure: how many source systems the numbers come from and whether any of them need a manual export; whether the definitions are already agreed or the engagement has to broker that agreement, which is a meeting problem rather than a technical one and takes longer than the build; and how often the numbers have to refresh, since hourly and daily are different pieces of engineering.

The thing to establish before a quote is whether two departments currently produce two different figures for the same measure. If they do, the first phase is definition work. A provider who quotes a dashboard before finding that out has priced the drawing rather than the plumbing.

• How many source systems, and how many of them need a manual export
• Whether the definition of each measure is agreed, or still disputed
• How fresh the numbers have to be, since refresh frequency drives the build
• Whether the output is a dashboard, a scheduled message, or both

When another dashboard is not the answer

If the underlying data is not captured reliably, a dashboard makes the gaps look like facts. Fixing the capture is the first project, and building a dashboard on top of incomplete data produces confident wrong answers faster than a spreadsheet would.

If nobody has said what decision the dashboard is meant to support, it will become a wall of charts that everyone agrees is impressive and nobody uses. The decision comes first, then the number, then the chart.

How long before a number appears, who checks it, and where it lives

The first week is definitions rather than charts. What counts as a lead, when revenue is recognised, whether a refund reverses the original month or lands in the current one. That week is uncomfortable and it is the whole value, because two departments producing two revenue figures is a definition problem wearing a visualisation costume. A first dashboard on agreed definitions is live inside a fortnight.

Numbers that will be acted on get a person in front of them. Anomaly detection runs unattended and flags a figure that has moved outside its normal range, then waits for a person to say whether that is a real change or a broken connector, because a dashboard quietly reporting zero after a feed has died is worse than no dashboard. Nothing reaches a board pack without human approval.

The warehouse, the queries and the dashboard definitions sit in your own accounts. You own the query layer, which is the part that took the thinking, and it survives a change of visualisation tool later. Moiz Khan, the co-founder who owns automation architecture, takes the definitions work himself, having come to systems through chartered accountancy. Wobble is answerable for the definitions it wrote and works from Karachi across 25 engagements in six countries.

Doing this in-house is often the right call. If you employ an analyst who knows the business, buying the definitions workshop and the pipeline and letting your own team build the views is better value than buying finished dashboards nobody trusts. Wobble has published its reporting work on the work page, including three custom ERP systems for RM Gulistan Engineers covering accounts, human resources and inventory.

Common questions

Why do two reports in my business show different numbers?

Almost always because the metric has more than one definition and more than one place it is calculated. One report counts every form submission as a lead, another counts only those sales accepted. Agreeing one written definition and one calculation point fixes it; building a better chart does not.

What should a daily operations dashboard show?

Enough to decide whether to change today's plan in about fifteen seconds. Volume against the same day last week, the one or two leading numbers, anything currently failing stated as a failure, and any work that entered a stage and has not left it. Deeper analysis belongs somewhere people go deliberately.

Should AI generate the numbers in a dashboard?

No. Arithmetic should be deterministic, because a number that is occasionally wrong is worse than no dashboard once people start acting on it. AI is well suited to the commentary, to summarising unstructured material such as what enquiries actually said, and to clustering conversations into themes.

How do you stop a dashboard from being abandoned?

By delivering the important part rather than waiting for a visit. A short message at the start of the day, a weekly summary and threshold alerts agreed in advance carry the habit. The dashboard stays available for the moment someone wants to look deeper.

What data sources can be combined into one view?

Typically the CRM, the website analytics, the ad platforms, the messaging channels, the billing or accounting system, and any operational tool the business runs on. The constraint is rarely technical access; it is whether the same entity, usually a customer, can be matched reliably across them.

Do we own the reporting system?

Yes. The pipelines, the metric definitions and the dashboard itself run on infrastructure the client controls and are documented for handover. Metric definitions in particular belong to the business, since they are the agreement everyone else's reporting depends on.