Data, the new capital of SMEs: collect and make reliable

Every day, your company produces hundreds of data points: orders, invoices, deliveries, payments, inventory movements, exchanges with customers. This raw material is worth its weight in gold — provided it is reliable and put to use. Yet in most SMEs it lies dormant, scattered across Excel files, isolated software and paper notebooks, to the point where important decisions are still taken blind. Here is how to turn the situation around.

Data, an asset that goes unvalued

People readily talk about human capital, financial capital or plant and machinery. But there is a less visible and equally decisive asset: data. Knowing precisely who your best customers are, which products are the most profitable, what your real payment terms are or the exact turnover rate of your inventory means holding an advantage that competitors who steer “by feel” do not have.

The particularity of this asset is that it depreciates fast. Data that is wrong, out of date or impossible to find at the right moment is worth nothing — worse, it misleads. Where a machine loses a few percent of its value each year, a customer database riddled with duplicates and obsolete information can become unusable within a few months. Making the most of your data is therefore not a one-off project: it is a continuous discipline.

That discipline is organised around a three-stage value chain, which must be tackled in order, because each step conditions the next:

1. Collect clean data, only once, as close as possible to the event that produces it.
2. Make it reliable through single master data, automatic checks and clearly assigned responsibilities.
3. Use the whole to decide — indicators, dashboards, analyses.

This article is devoted to the first two stages, the ones almost every SME skips because they are thankless and invisible. The third — choosing your indicators, building a dashboard that is actually useful, drawing decisions from it — assumes the upstream work is done: it is covered in a separate article, dedicated to Business Intelligence and steering with dashboards. We therefore stop precisely at the moment data becomes usable.

Step 1 — Collect: single data entry at source

Collecting reliable data means recording it once only, at the moment and place where the event occurs, then letting it circulate without ever re-entering it. The sales rep who confirms a sale, the storekeeper who receives goods, the driver who collects payment in the field: each enters the information they are responsible for, and that information then serves everyone.

The great enemy of collection is double data entry. When the same order is copied into an inventory file, then into invoicing software, then into accounting, you multiply not only wasted time but above all the opportunities for divergence. A missing decimal point, an amount rounded differently, a reference mis-copied, and the versions no longer match.

A few principles make collection sound from the outset:

• A single entry point per type of information: the customer is created in one place only, never duplicated across five files.
• Data entry at the right moment: capture the data when the event happens, not the next day from memory — this is the whole point of field teams equipped with synchronised mobile tools.
• Structured fields rather than free text: a drop-down list of categories is a thousand times better than a “comment” box where everyone writes as they please.
• Data entered by the person who knows it: whoever is responsible for the operation, not a department transcribing blind.

Step 2 — Make it reliable: single master data and checks

Even when properly collected, data degrades if the organisation imposes no safeguards. Making data reliable means guaranteeing that at any given moment there is one single correct version of each piece of information, shared by every department.

Single master data

The heart of the matter is master data: the master list of customers, items and suppliers. As long as each department maintains its own list, duplicates proliferate. “Société Ben Salah”, “Ben Salah SARL” and “ETS BEN SALAH” designate the same customer, but count as three in the statistics — and their real revenue remains invisible. Single master data, shared and maintained, removes that fog at the root.

Automatic checks

A good system prevents the error instead of correcting it after the fact. Detecting a duplicate tax identification number when a customer is created, refusing a negative inventory quantity, flagging an aberrant amount, requiring the mandatory fields of an invoice: these preventive checks are worth far more than hours of after-the-fact clean-up.

Traceable history

Reliable data is also data whose origin is known: who created this order, who modified it, and when. This traceability is not merely a matter of internal control — it is the condition for trusting the figures and settling disputes without endless discussion.

What “reliable” means, precisely

The word is too vague to be actionable. Management data is in fact judged on five distinct qualities, and a file is frequently flawless on three of them and disastrous on the other two:

• Accuracy: the value matches reality. A purchase price entered excluding tax in a field meant to include tax is perfectly structured, perfectly up to date… and perfectly wrong.
• Completeness: the fields that matter are filled in. A customer record with no tax identification number or geographical area rules out any analysis by territory and will block compliant invoicing.
• Freshness: the information reflects the current state. Inventory that was correct on the 1^st of the month helps nobody on the 20th.
• Uniqueness: one reality, one record. This is the quality that duplicates destroy silently.
• Consistency: the same information says the same thing everywhere. If total sales in the sales module do not match those in accounting, one of the two is lying — and as long as you do not know which, both are unusable.

The value of this grid is practical: it turns a vague feeling (“our data is not great”) into a precise diagnosis. Take your three essential tables — customers, items, sales — and score them honestly on these five points. The result alone shows where to start, and avoids launching a major clean-up where the problem was not.

Who is responsible for what: governance in practice

Here is the point that determines whether the whole effort lasts. A database cleaned once degrades within a few months if nobody is accountable for it. Data governance — an intimidating phrase for a simple idea — consists in answering three questions, and an SME of ten people can do so on a single page.

Who is allowed to create?

Disorder almost always begins with uncontrolled creation. If five people can create a customer or an item, five naming conventions will coexist, and duplicates will come back whatever you do. Restricting creation to a limited number of people — or framing it with an approval step — costs a few minutes of waiting and saves months of clean-up.

Who arbitrates in case of disagreement?

Two departments show two different figures for the same month: who decides? Appointing an owner for each master data set — the sales manager for customers, the purchasing manager for items and suppliers — avoids endless debate. This owner is not an IT specialist: it is the person who knows the business and stands behind the definition adopted.

How do we name things?

A written coding convention, however rudimentary, is worth more than the best of intentions. How is an item coded, in what order, with which abbreviations? Do we write the full company name or the trade name? What do we do with accents and capitals? Half a page of rules, available to those who do the data entry, prevents divergence from creeping back the very next month.

The migration project: cleaning up existing data without blocking everything

There remains the question everyone puts off: what to do with years of already accumulated data? The temptation runs two ways, and both extremes come at a cost. Starting again from scratch loses the very history that gives data its value. Migrating nothing condemns you to dragging your duplicates along indefinitely.

The workable path comes down to a few principles, applicable without stopping the business:

• Turn off the tap before mopping up. Put the checks at creation in place first. Cleaning a database that keeps producing duplicates is a Sisyphean task.
• Deal with what is active first. A customer with no order for four years does not deserve the same energy as one who buys every week. Concentrate the clean-up on the scope that genuinely moves.
• Merge rather than delete. When two records designate the same customer, merging preserves the history of both; deletion would amputate real revenue.
• Date the migration. Note from what point data is considered reliable. Any analysis going back before that date must be read with caution — knowing this is better than discovering it in a meeting.

This project is never finished in the strict sense, and that is normal. The goal is not perfection: it is to reach a level where the figures stop being argued over in every meeting, then to maintain that level through the checks and governance put in place upstream.

The target outcome, at a glance

Collection at source, single master data, checks, governance and migration of existing data all converge on the same end state. The table below summarises what distinguishes an SME that endures its data from an SME that masters it:

Criterion	Scattered data	Reliable data
Information entry	Repeated across several tools	Single, at source
Customer / item master data	Multiple diverging lists	One single shared list
Duplicates	Frequent, undetected	Blocked at creation
Consistency across departments	Contradictory figures	A single version of the truth
Traceability	Unknown or partial	Complete history
Time needed to obtain a figure	Hours of consolidation	Real time

How to know that data has become usable

This is where the upstream work stops — and it helps to recognise the moment you have got there. Three signals never lie, and none of them is measured in a piece of software.

The first: nobody disputes the figures in meetings any more. The discussion is about what they mean and what to do, no longer about whether they are correct. The second: obtaining a figure no longer requires preparation; there is no more “I will get that ready for tomorrow” and no more manual consolidation between two files. The third, and the most revealing: an unusual question finds its answer. As long as you can only answer questions anticipated in advance, data is a frozen report; when you can freely cross-reference customer, item, period and channel, it has become an asset.

Once these three conditions are met, the subject changes nature entirely. It is no longer about data entry quality but about choices: which indicators to track, how to build a dashboard people actually look at, how often to read it, and how to move from observation to action. That is a discipline in its own right, and it is the subject of our dedicated guide to Business Intelligence and steering with dashboards.

A final word on what comes next, because it reinforces everything above: a clean database is also the only credible starting point for artificial intelligence in business. Fed on duplicates and contradictory figures, AI merely amplifies the noise. The SME that disciplines its collection today builds an information asset that no tool, however advanced, will be able to reconstruct after the fact.

The role of a unified management system

This entire chain — collecting and making data reliable — rests on one prerequisite: that the data lives in the same place. That is precisely what an integrated management software suite (ERP) provides. By bringing sales, purchasing, inventory, finance, accounting and HR together in a single database, it guarantees single data entry, enforces shared master data and applies checks at creation. Data stops being a scattered by-product and becomes a structured asset.

The contribution is not only about centralisation: it lies in the fact that quality becomes a side effect of everyday work, rather than an extra chore. The sales rep recording a sale is not “doing data quality”, they are doing their job — but that action feeds single master data, triggers the checks and leaves a time-stamped trace. It is the only mechanism that lasts, because it does not rely on each person's voluntary discipline.

This is the philosophy behind Swifto: a cloud platform where every operation entered — at the office as well as in the field — feeds a single database, with non-duplicated customer records, checks at creation and a complete history of changes. Designed for Tunisian SMEs and supported by a local team, it makes data reliability a consequence of everyday use rather than a separate project — the bedrock on which dashboards and indicators can then be built.

Frequently asked questions