Data-driven decisions for SMEs: where to start

For an SME, making data-driven decisions means answering a concrete business question with data the company already produces – in the ERP, at the till, on machines, in time tracking. A new analytics tool is rarely the first step. What comes first is a clear question, connected data sources and checked data quality. Only then do numbers give a better basis than gut feeling.

What data does an SME already have?

Many small and mid-sized businesses have more data than they use. The problem is rarely a lack of data; it is that the data sits in separate systems that were never built to talk to each other. A typical business works with several sources, each describing one part of the operation.

• ERP or inventory system: orders, stock, purchasing, invoices, customer and supplier master data
• Point-of-sale system: sales by time, branch, item and payment method
• Machines and equipment: run times, downtime, unit counts, fault messages – often available via controllers or sensors
• Time tracking: attendance, hours per job or cost centre, overtime
• Accounting: costs, revenue and cash position, usually lagging behind day-to-day operations
• Spreadsheets and emails: planning lists, quote overviews and notes that exist nowhere else

The last category is often underestimated. In many businesses the real steering knowledge lives in a spreadsheet maintained by a single person. Anyone introducing data-driven decisions has to take these lists as seriously as the ERP system.

Why isn't exporting the data enough?

An export from every system into one big spreadsheet looks like a quick fix. In practice it fails for three reasons: the exports go stale quickly, matching records between systems is manual and error-prone, and later nobody can trace where a number came from. A decision based on a spreadsheet only one person understands is hardly more data-driven than a decision based on instinct.

Data sources should therefore be connected through interfaces so that data flows into one place automatically and regularly. That can be a small database, a data warehouse or a well-built integration layer between the ERP and reporting. The size of the solution is secondary. The path of the data must be traceable and repeatable.

How do you combine data from ERP, till and machines?

Combining data starts with a shared key. An order in the ERP, the hours booked against it in time tracking and the run time on the machine must match unambiguously through an order number, item number or cost centre. Without that shared key, even good software cannot create a reliable link.

1. 01Define one concrete question, for example: which orders actually make a contribution margin once real effort is counted?
2. 02Identify the sources that answer this question – usually two or three systems, not all of them.
3. 03Check which key links the data and name the gaps in how it is recorded.
4. 04Set up interfaces that move the data to one place automatically and regularly.
5. 05Cross-check the first results with the people who know the process every day.
6. 06Only once the numbers hold up, make the report permanent and move on to the next question.

This order prevents a business from launching a large data project that produces many figures but does not improve a single decision. One answered question is worth more than many half-finished reports.

Which traps distort data-driven decisions?

The wrong metric

The most common trap is a metric that is easy to measure but does not answer the real question. Revenue per customer says nothing about profit if some customers cause far more effort than others. Machine utilisation says nothing about delivery performance if the machine is producing the wrong parts. A good metric is tied directly to a decision: when the number changes, it is clear what someone would do differently. How to turn such metrics into a dashboard that people actually use day to day is covered in a separate article.

Poor data quality

In SMEs, data quality is rarely a technical problem; it is a capture problem. Hours are filled in afterwards from memory, items are booked under a catch-all number, machine stops have no reason attached. Every report inherits these inaccuracies. Before a number carries a decision, someone should check how it is produced and where it has gaps.

Same word, different meaning

An “order” in the ERP is not automatically the same as an “order” in time tracking, and “revenue” at the till means something different from revenue in accounting. Anyone combining data has to define such terms once, clearly and in writing. Otherwise management ends up discussing figures that different departments understand differently.

Numbers without context from the floor

Data shows what happened but rarely why. A drop in output can be down to a machine, missing material or a shift that was short-staffed. The people who live the process every day therefore belong in the analysis. Data-driven decision-making does not replace experience; it makes experience testable.

Does an SME need AI for this?

Most data-driven decisions in an SME do not need artificial intelligence; they need connected systems and clean data. AI can help later, for example in spotting patterns in fault messages or answering questions about internal documents. But it relies on the same foundation: findable, consistent data. Without that foundation, AI starts at the wrong end. Businesses that want AI assistants can also run them locally on their own network, which makes it easier to keep track of GDPR data protection and the requirements of the EU AI Act.

Who in the business should be responsible for the data?

Data-driven decisions need clear ownership. For every important data source it should be clear who ensures that data is captured correctly – usually the department that works with the system every day, not IT. IT or an external partner makes sure the connections between systems keep running and the data is stored securely.

In addition, someone needs to maintain the definitions of the most important terms and metrics and coordinate changes. In smaller businesses that is often management itself or the controller. Without this role, every report quickly grows its own variants, and in the end there are several versions of the truth again.

• Business department: responsible for complete and correct data capture
• IT or IT partner: responsible for interfaces, storage, access rights and backup
• Management or controlling: responsible for terms, metrics and which questions are answered

How NDVDL takes this on for your business

NDVDL does not start with a tool but with an appointment at your premises. We look at how orders, material and hours actually flow, which systems generate data along the way and which decisions are hard for you to make today.

• On-site appointment: which questions should the data answer, and where is the data captured today?
• Written proposal: which sources to connect, which metrics make sense and what is needed for this
• Implementation: interfaces between ERP, till, machines and time tracking, and a suitable report where needed
• Operation and maintenance: we keep the connections running, even when systems change, with one fixed contact person

The decisions stay with you, but the work of bringing data together reliably does not. When a standard system changes or a new machine is added, NDVDL adapts the connection instead of letting the report quietly go out of date.