A data-driven company makes decisions based on the data it produces every day (orders, invoices, stock, production, customer service) instead of relying only on the instinct of the people running it. In an SME, data-driven decision making does not require a team of data scientists. What you need is data that is connected, a business dashboard with a few indicators that update automatically and, only later, a few artificial intelligence models for forecasting.
The owner's experience still counts, and a lot. The problem is that beyond a certain size there are too many questions to answer from memory: which customers are ordering less than usual, which product family really makes money, which machine stops most often and why. This guide explains what it means to base decisions on data, where to start and when AI is actually useful.
What data-driven means in a company
Data-driven means "guided by data". In a company it comes down to three habits.
- Important questions have a reference number, for example margin per customer, average delivery time or the share of customers who reorder within six months.
- Those numbers are up to date and reliable. If someone rebuilds them by hand at the end of the quarter by pasting together three Excel files, they are not.
- Decisions are checked afterwards. If you change a price list or a process, you watch what happens to the numbers in the following weeks and adjust.
People's judgement stays at the centre. The data tells you that a long-standing customer has halved their orders, but working out whether it is down to price, service or a new purchasing manager is a job for the sales rep. With your data in order you notice it after two months, instead of when the annual accounts are closed.
The first obstacle: data scattered across too many systems
Almost every SME has more data than it uses. The trouble is that it sits in different places: the ERP for orders and invoices, the CRM for deals, the online shop for web sales, an Excel sheet for agents' commissions, another for consignment stock. Each system knows one part of the story.
As long as they stay separate, every report needs someone to export, paste and tidy up. The work is slow, errors are frequent and the report usually arrives too late to act on. We covered this in scattered data, blind decisions: the first step towards becoming a data-driven company is getting your systems to talk to each other through integrations and APIs, before you even think about charts and algorithms.
That is what happened with the management system built for Pulltex. Customer records, agent assignments and pro forma invoices lived in systems that did not communicate, and management had no clear, up-to-date view of the sales network. Today this information sits in a single platform, synced with HubSpot via API.
Business dashboard: the first tool to build
Once the data is connected, a business dashboard is the simplest way to use it. It is a screen that shows, updated automatically, the few indicators decision makers need. Five or six numbers you actually look at every Monday morning, not thirty charts.
For a distributor these might be revenue and margin by customer and by product family, late orders and stock that has not moved for more than 90 days. An installation contractor will look at open jobs, average response times, billable hours per technician and maintenance contracts coming up for renewal. A manufacturer will want machine status, parts per shift, stoppages with their cause and scrap. Anyone working with a network of agents needs open deals by agent, closing times and customers who have not ordered for three months.
In the case of Buzzi Connect, Buzzi SRL's high-frequency welding machines are connected to a single web interface that shows each machine's parameters, production statistics and alarms in real time. Production data is stored over time and can be exported to the customer's own systems (MES, ERP) to track performance.
For a dashboard to be used, every indicator has to correspond to a decision. Ask yourself what you would do differently if that number went up or down; if you don't know, remove it. Then assign each number to a person: if delivery time gets worse, it must be clear who deals with it. And make sure it updates by itself, because a dashboard that has to be filled in by hand gets abandoned within a few months.
A clear position: if you use a single ERP that already has decent reporting, learn to use that before buying a business intelligence tool. A dedicated tool makes sense when you need to combine data from several systems.
Business data analysis: from the past to forecasts
Business data analysis comes in three levels, and it pays to climb them in order.
Descriptive analytics: what happened
This is the dashboard level: monthly sales, margins, delays. It sounds trivial, yet many companies never get there with reliable numbers. It is also the level that pays back fastest.
Diagnostic analytics: why it happened
This is where you cross-reference data. Are sales falling in every region or only in Emilia-Romagna? Across all product families or only spare parts? With every agent or with just one? To answer, you need connected data, which is another reason why system integration comes first.
Predictive analytics: what will happen
This is the level of machine learning, meaning algorithms that learn from historical data to estimate what will happen. Some realistic cases for an SME:
- A food company with fresh products uses previous years' sales, seasonality and the promotions calendar to estimate how much to produce each week and cut waste.
- A distributor spots customers whose ordering pace is slowing compared with their usual rhythm and alerts the area sales rep before the customer moves to a competitor.
- A machine shop collects temperatures, vibrations and alarms from its machines and recognises the anomalies that usually come before a breakdown, so it schedules the repair instead of suffering the stoppage.
- A building materials wholesaler that offers deferred payment uses customers' past payment punctuality and seasonality to decide credit limits and terms.
- A farm stay starts from past bookings, local events and the time of year to suggest the rate for each night.
- An order desk uses AI to read emails and documents and route them, generating clean data for later analysis along the way. The topic is covered in the guide to RPA robotic process automation.
None of these models works without a clean and long enough history. If your reliable sales data covers six months, a demand forecast will be little more than a guess: work on the descriptive level first.
Where Italian companies stand
Interest in data is growing fast. According to the Istat "Enterprises and ICT" 2025 survey from Italy's national statistics institute, over the last two years the share of Italian companies with at least 10 employees that carried out data analysis, with in-house or external staff, rose from 26.6% to 42.7%. In 2025 the use of at least one artificial intelligence technology among Italian SMEs doubled compared with the previous year, from 7.7% to 15.7%.
The same survey reports the obstacles cited by companies that considered AI without adopting it. Lack of skills comes first (58.6%); 45.2% point to the unavailability or poor quality of the data needed. In almost half the cases, then, the problem is the data.
How to become a data-driven company in 5 steps
1. Start from a decision you currently make badly
Don't ask yourself what you can do with AI. Ask which decision you currently make on gut feeling because information is missing: how much credit to give a new customer, which jobs to accept in peak season, how much material to order in September. Then set a measurable goal.
2. Map and connect the data
List where the data needed for that decision lives, what format it is in and who updates it. Then connect the sources so they feed themselves. It is the longest and least visible part of the work, and without it every later analysis starts from wrong numbers.
3. Build a first dashboard
A few indicators, updated automatically, shared with the people who have to act. Use it for a few weeks and watch which numbers get looked at and which get ignored.
4. Try a predictive analytics pilot
Pick a single case with a good balance between impact and complexity. Train the model on part of the historical data, measure how accurate it is on data it has never seen and compare it with the method you use today. If it doesn't improve decisions, stop or change approach.
5. Put the result where people work and keep it in check
The forecast has to reach the ERP, the CRM or a notification to the person in charge, not a file nobody opens. It then needs monitoring over time: the market changes, and a model trained on data from two years ago loses accuracy. When that happens, it has to be retrained on recent data.
The most common mistakes
The most frequent mistake is buying a business intelligence tool before having clean data: the charts look great and are wrong. The second is measuring everything, because nobody looks at a hundred indicators. The third is handing the project to IT, while the right questions come from sales, production and accounting. The last is mistaking a forecast for a certainty: it is an estimate, and it should be read together with the experience of people who know the customers and the market.
For examples of AI applications that suit small and medium-sized companies, see the article on artificial intelligence for SMEs. How Selli connects data, automations and predictive models to business processes is described on the page about AI automation.
Let's talk: which decision do you want to base on data
A company becomes data-driven by starting from a precise question, not from a piece of software. Tell us which decision you make on gut feeling today: together we will look at where the data you need lives, how to connect it and whether the reporting you already have is enough or a dedicated dashboard is needed. If there is development to do, we work in two-week sprints with a test environment where you check the numbers before they reach production, and the code remains your property. Book a call with Selli and let's start from your first indicator.