Digital transformation for SMEs: a route you can actually walk

Digital transformation for SMEs has a bad reputation, and it has partly earned it. The label has been used to sell €12,000 websites, ERPs nobody ever finishes rolling out, and grants spent on licences that expire before anyone opens them. This guide goes the other way: the real order the work has to follow, what can be done with the team you already have, and the point at which — only then — it makes sense to bring artificial intelligence into the picture.

We wrote it for a company of 10 to 150 people with a business that works: it bills, it has loyal customers, and its people know their jobs. It does not need reinventing. It needs to stop losing hours on tasks a machine does better, and start deciding with data instead of gut feel. That is the whole promise, and it is plenty.

What is digital transformation in an SME (and what it isn't)?

Digital transformation in an SME is the process of turning a manual, scattered operation into one supported by systems that talk to each other. It is not buying software. Software is the consequence; the work is something else.

In practice it comes down to four concrete changes:

• Data stops living in one person's head and starts living in a system others can reach.
• Repetitive tasks stop being done by hand and start running on their own, with a person reviewing exceptions.
• Decisions stop being made on hunches and start being made against a figure everyone accepts as correct.
• The customer stops depending on who picks up and starts getting the same answer through any channel.

And what it is not: it is not having profiles on five social networks, it is not swapping the ERP every three years, and it is not a project with an end date. It is also not digitising chaos. If a process is badly defined, automating it only makes the errors arrive faster and in greater volume.

Digitising a broken process does not fix it: it industrialises it.

There is a simple test for whether your company has genuinely started. Ask two people from different departments how many orders were fulfilled last month. If they take more than two minutes, or give different figures, the work is still ahead of you, no matter how many tools you are paying for.

Why do so many digitisation projects fail?

Almost always for the same five reasons, and none of them is technical:

• They start with the tool. Someone sees a demo, falls in love, then goes looking for somewhere to fit it. The correct order is the reverse: measured problem, defined process, chosen tool.
• Nobody has the project in their job description. If everyone owns it, nobody owns it. It needs a named person, allocated time, and authority to decide.
• It is rolled out without migrating the real work. The new system runs alongside the old spreadsheet for months "just in case", and the old spreadsheet always wins.
• Nothing is measured beforehand. If you don't know how many hours the process costs today, you can't prove it improved — and the project dies in the first budget cut.
• Everything happens at once. Three open fronts in a 40-person company means three half-finished fronts.

On top of that sits an expectations problem: transformation is sold as a leap and experienced as a run of uncomfortable months in which people do the usual work plus the new work. Whoever fails to prepare the team for that discomfort loses the project in week six. We break the pattern down in why AI pilots fail, where it repeats almost identically.

What are the four phases of a realistic route?

This is the sequence we follow, and the reason each phase sits where it does. None can be skipped: each one produces the material the next one needs.

| Phase | What happens | Sign it's done | Typical timeline | |-------|--------------|----------------|------------------| | 1. Foundation | Invoicing, CRM, storage and access tidied up; one source per data point | Nobody hunts for a file in their inbox | 1-3 months | | 2. Processes | The 5 most time-consuming processes are mapped and 2 automated | The process runs itself and someone reviews exceptions | 2-4 months | | 3. Data | Systems feed a single place; a short dashboard is built | Management looks at the same figure every Monday | 2-4 months | | 4. AI | AI is applied where there's volume, clean data and a repetitive decision | One specific task drops from hours to minutes | 1-3 months per case |

Phase 1 — The foundation. Boring and non-negotiable. Electronic invoicing working, a CRM holding the real customers (not one spreadsheet per salesperson), documents stored somewhere with permissions, and backups someone has actually tried restoring. A typical example: a distributor with three salespeople, each keeping their contacts on their own phone. Before discussing anything else, those contacts go into a CRM with shared fields. It costs two tedious weeks and saves two years of unusable data.

Phase 2 — The processes. This is where the first visible return shows up. List processes by hours consumed per month and attack the top two. In a services firm that is usually customer onboarding and quote generation; in a manufacturer, delivery-note intake and work-order tracking. We develop the selection criteria in process automation: high volume, stable rules, costly errors.

Phase 3 — The data. Once processes run through systems, data exists in a usable format. That is the moment to bring it together and build a dashboard of eight to twelve metrics, not forty. If this phase feels distant, start with what business intelligence is and with defining the source of truth for each figure.

Phase 4 — AI. It only makes sense with the previous three reasonably done. With ordered processes and queryable data, AI stops being an experiment and starts solving specific things: classifying inbound email, extracting data from PDF invoices, drafting reply templates, flagging which customer is about to stop buying. In AI for SMEs we detail what can be attempted with no technical team.

What does it cost and how is it funded?

There is no single figure, but there are honest ranges for a 20-to-100-person company, counting external services and first-year licences:

• Phase 1 (foundation): €3,000 to €15,000. Most of it is time spent tidying up, not software.
• Phase 2 (processes): €5,000 to €25,000 for the first two processes, depending on integrations.
• Phase 3 (data): €6,000 to €30,000, heavily driven by how many systems have to be connected.
• Phase 4 (AI): from €4,000 for a tightly scoped case; more if it needs deep integration or a compliance review.

Two warnings about the money. First: recurring cost matters more than implementation cost. A €8,000 solution with €600/month in licences costs €15,200 over two years. Always ask for the 24-month total. Second: public digitisation grants change their conditions with every call, so verify deadlines and requirements at the official source before counting on them — and never design the project to fit the grant. A project shaped around a subsidy is usually a project you didn't need.

What does the team have to provide (and what can't be delegated)?

The technical part can be contracted out. The human part cannot. Three things have to come from inside:

• An owner with real time. Not a sponsor who shows up at the kick-off: someone spending four to six hours a week for the duration.
• The business decisions. What counts as an active customer, when an order is closed, which margin is used for comparisons. No supplier can decide this for you, and without it the dashboard means nothing.
• Adoption. Training, support and, above all, removing the old alternative. As long as the parallel spreadsheet survives, the new system is optional.

One detail that makes a difference: involve the people who do the work today, not only those who manage it. The person who raises invoices knows where the odd exceptions live — the ones no diagram captures. If they appear during design, the system handles them; if they appear during training, the system fails in month one.

How do you start next month without opening a huge project?

A 30-day plan any management team can run without new budget:

• Week 1: inventory. List the systems in use, who administers them, what they cost monthly and which data lives in each. Paid software nobody uses tends to surface.
• Week 2: hours. Ask each area which three tasks eat the most time and estimate hours per month. Precision isn't needed; order of magnitude is.
• Week 3: source of truth. Pick five management figures (sales, margin, orders, average lead time, overdue payments) and decide which system each comes from. One only.
• Week 4: one bet. Choose a single process from week 2 with stable rules and give it a date, an owner and a before/after metric.

That alone gives a good supplier what they need to put a serious proposal together, and gives you what you need to dismiss the ones who only want to sell licences. The criteria for that conversation are in how to choose an AI consultancy, and the groundwork on data in data strategy.

Frequently asked questions

How long does digital transformation take in an SME?

The four phases usually span 12 to 24 months, but the first visible result arrives in 8-12 weeks if you attack a single process. If a supplier promises full transformation in six weeks, they are selling a software installation, not a change in how the company operates.

Can we start with AI and leave the foundation for later?

You can try, and it usually gets expensive. AI needs accessible data and defined processes; without them the project becomes a pretty demo nobody can put into production. The exception is tightly scoped cases that don't depend on internal data, such as drafting replies or transcribing meetings.

Do I need to hire someone in IT?

In a company under 50 people, usually not at the start. What you do need is an internal owner with allocated time and a supplier who takes responsibility for maintenance. Past a certain volume of in-house systems, an internal profile stops being a luxury.

What about customer data and GDPR?

Every new system that processes personal data enters your record of processing activities and needs a data processing agreement with the supplier. It is not paperwork you resolve at the end: it shapes which tools you can use and where data may be hosted, so raise it during selection.

If you are at the point of deciding where to start and would rather not do it blind, the audit is exactly that: we review your systems, your processes and your data, and hand back a map with the order of the bets and what each one costs. If you already know your direction and want to test the plan against someone who has done this before, let's talk for half an hour and we'll tell you what we'd do and what we'd skip.