A KPI dashboard brings together the indicators linked to your business objectives, along with the reporting period and the comparisons needed to interpret them. For a small or medium-sized business, it should help you make a specific decision: which quotes to follow up, which jobs to review or which overdue payments to chase. Choosing the charts comes later.
You can start with a shared spreadsheet if you have a small amount of data and someone checks it. Connecting every business application to a new dashboard straight away risks automating calculations that nobody has agreed on yet. First, try using the indicators in an actual meeting and see which information is missing when you need to make a decision.
Which business KPIs to choose for management
A KPI is a key performance indicator: it measures your progress towards an objective. The number of quotes sent can be a useful metric; it becomes a KPI if you link it, for example, to the need to generate future sales. If your objective is to reduce unpaid invoices, the number of quotes tells you little.
The Microsoft guide to KPIs recommends starting with objectives and assigning each indicator an owner and a monitoring frequency. In practice, ask the people who will use the dashboard what they would do differently if a value fell outside the agreed threshold. If nobody can answer, I would leave that figure in the detailed report.
Sales: distinguish activity from results
A distributor might need the value of orders won and the quote win rate. The first shows the work coming in; the second helps identify which quotes led to an order. The number of phone calls alone does not tell you whether sales performance is improving.
Define the rate before building the chart. One possible convention is to divide accepted quotes by all quotes closed, whether won or lost, during the same period. Quotes that are still open are excluded. This measures outcomes during the period, not the likelihood of success for every quote issued that month. For that analysis, you need to follow the same group of quotes through to their outcomes.
If nobody marks lost quotes as closed in the CRM, the indicator will be distorted. You need a rule for updating the sales process before you need a dashboard. The article on CRM and the sales process explores this connection between people's work and the available data.
Operations: understand where work is building up
An installation contractor might track visits that are overdue against the agreed date and completed service reports that have not yet been invoiced. These are different indicators: the first prompts a review of scheduling, while the second calls for a check of documents and administrative steps.
Definitions matter here too. A job on hold because a spare part is missing should be distinguished from one that nobody has assigned. You can include both in the overdue total, but you must be able to separate them when investigating the cause. Hiding them all under “in progress” makes it difficult to establish who needs to act.
Finance: separate sales, margins and cash receipts
Revenue is not the same as available cash. A management view can show outstanding overdue invoice balances alongside a measure of job profitability. The finance team needs to agree which entries go into the calculations and the date up to which they are current.
If you show revenue minus direct costs allocated to a job, label it accordingly and specify which costs are excluded. That margin is not the company's net profit. If some hours or supplier invoices have not yet been recorded, flag the result as provisional. Otherwise, a green status could suggest profitability that you have not verified.
A definition sheet to complete before building the dashboard
The simplest way to avoid arguments about the figures is to write a definition sheet for each indicator. A shared page can be enough, provided both the people preparing the data and those using it approve it.
This is an example sheet, not a recommended threshold for every business:
| Field | Example: outstanding overdue customer balances |
|---|---|
| Decision to support | Choose which accounts to review and follow up for payment |
| Definition | Sum of unpaid balances on invoices with a due date before the observation date |
| Source | Receivables schedule in the business management system, updated with recorded receipts |
| Scope | Selected company, amounts in euros; disputed balances shown separately |
| Observation date | End of the previous day |
| Expected refresh | Every morning, after importing payment receipts |
| Owner | Finance contact designated by the company |
| Comparison | Previous week, using the same scope |
| Action | Review the accounts and assign the next follow-up |
Add rules for credit notes and partial payments, agreeing them with the finance team. Avoid adding the original amount of an invoice again when it has already been partly paid. Disputed balances need to be interpreted differently from those that require a routine payment reminder.
The sheet must also explain what happens when data is missing. An absent amount is not zero: zero indicates that a measurement has been made, while an absence means you cannot draw a conclusion yet. Show a readable warning on the dashboard and retain the last valid value with its date if you choose to display it.
Where business dashboard data comes from
Typically, the CRM describes quotes and sales opportunities, the business management system holds orders and invoices, and the e-commerce platform records purchases and returns. These sources need to be reconciled, rather than treated as three totals to add together. An online order imported into the business management system must appear only once in the order total.
You therefore need an identifier that lets you recognise the same document across systems. Customer records also need to be linked using a stable rule: the company name alone can change or be written differently. The article on scattered data and systems integration explains the operational consequences of these gaps.
Before importing everything, check which fields your chosen indicators actually need. A monthly view of orders may require only a few data fields; calculating a job margin also means reconstructing the costs allocated to it. I would not promise that margin while technicians' hours remain in notes that nobody enters into the system.
Another point to make visible is data freshness. In the Power BI documentation on data refresh, Microsoft distinguishes imported data from its sources and explains that refreshing involves several stages. Labelling a dashboard “automatic” does not mean it represents what is happening right now.
Show both the period covered by the documents and the time of the last successful refresh. If the connection fails, leave a warning next to the affected data. A manager must be able to distinguish a day with no orders from an import that did not work.
Example: an installation contractor's dashboard
Consider a hypothetical example, not a result achieved for a Selli client. The owner wants to understand why completed jobs are not reaching invoicing. The technician closes the job in the app, the office checks the service report and the finance team prepares the invoice.
An initial view could show completed jobs without an invoice, separating those with incomplete documentation from those ready for the finance team. Selecting the total lets the office see the relevant jobs and the step still outstanding. This allows the owner to assign a check without asking everyone to reconstruct the list.
To avoid a misleading total, jobs that the contract schedules for later invoicing must be excluded or shown separately. A newly completed job is not necessarily overdue. The planned invoice date and contract type help identify which cases need attention.
Acceptance testing means taking some actual jobs and following the process through to the administrative document. If the dashboard cannot explain why a job appears on the list, adding another chart will not solve the problem.
How to make the dashboard readable
The opening page should provide a summary; the documents belong in the detail view. The Microsoft dashboard design recommendations suggest considering your audience, keeping an overview readable on a single screen and using consistent comparison periods.
For management, I would choose a view with a small number of indicators and visible filters. Every value should show its unit of measurement and period. Next to it, display a meaningful comparison: comparing the current month with the whole previous month can make performance look negative simply because fewer days have elapsed.
Use colour to flag an agreed condition, accompanied by a description. A high overdue balance may need attention; an increase in orders may be positive only if the company can fulfil them. Discuss thresholds with the people who know the process, rather than choosing them to fill the screen with traffic lights.
Excel, a BI tool or custom development
If you review a small amount of data each month and one person can prepare and check it, starting with Excel is reasonable. It lets you test the formulas and understand which indicators you will actually use before investing in integrations.
A business intelligence tool is suitable when you need to consult multiple sources and analyse the same data by department or period. Check the available connections, however, as well as who will maintain the calculation rules and how you will manage access. The work involved does not end with the first chart.
Custom development makes sense when the dashboard needs to become part of the process: opening a job, assigning a payment follow-up or collecting confirmation with specific permissions. For a screen that only shows totals already available in your business management system, I would start with the existing reports. Integrations between applications may be enough without replacing the software you use.
Acceptance testing before release
Ask for the dashboard to be checked against the source documents, together with the person who knows them. Useful acceptance testing includes these checks:
- Compare a closed period with the report from the business management system, using the same filters and exclusions.
- Change a test document, for example by cancelling an order, and check how the indicator changes after the refresh.
- Check that an order present in two systems is counted only once.
- Simulate an interrupted refresh: the dashboard must flag that the data is not up to date.
- Sign in with different roles and check the detail views and exports too, not just the opening page.
Then have people use the dashboard during a meeting. Note which decisions were made and which questions required a separate spreadsheet. Those questions will tell you what is missing from the next version; indicators that nobody consults can wait.
Automated reports and AI: after checking the calculations
AI can help prepare a written summary or find information in documents. For adding up invoices and calculating percentages, I would use deterministic, verifiable rules instead. If a model suggests an explanation for falling sales, check that explanation against the data: a correlation does not establish the cause.
In its work on business automation and reporting, Selli connects data and documents to existing software, retaining human review at important steps. The starting point remains the indicator agreed with the people who need to use it. A well-written summary cannot fix an incomplete receivables schedule.
Bring your current report to a call
If you want to build a KPI dashboard, bring a report you currently use and identify a decision that requires too many manual checks today. In a call with Selli, we can assess whether improving what you have is enough, whether to connect the sources or whether to develop a dedicated view. If the data for your chosen indicator is missing, it is better to establish that before choosing a tool.