Pick the one process that wastes the most staff time this month, fix that, and measure the hours you got back. That is digital transformation. Everything else — the roadmaps, the maturity models, the three-year vision deck — is scaffolding that small and mid-sized companies rarely need and almost never finish. The research backs the narrow approach: Boston Consulting Group’s September 2025 study found 60% of companies were getting hardly any material value from their AI investment, while only about 5% were generating value at scale.
What the word is hiding
“Digital transformation” survives because it is vague enough to sell. Underneath it sit four plain activities: moving work off paper and spreadsheets, connecting systems that currently need a human to copy between them, giving customers a way to serve themselves, and using data you already hold to make a decision earlier. That is the whole category.
Reframed that way, the work becomes tractable. You are not transforming an organisation. You are removing a specific piece of friction, confirming it stayed removed, and then doing the next one. The compound effect over two years looks like a transformation from the outside, which is fine, but nobody inside ever had to use the word.
- Which task do people complain about every week, and how many hours does it take?
- Where does someone copy data from one screen into another?
- What do customers phone to ask that they could look up themselves?
- Which decision do we make late because the number arrives late?
Where everyone actually is
It helps to know the baseline before you decide you are behind. Eurostat’s figures, extracted in January 2026, show that 49.3% of small EU enterprises bought cloud computing services in 2025, rising to 66.78% of medium-sized and 84.67% of large ones. AI is much earlier: Eurostat’s December 2025 extraction put AI use at 17% of small enterprises, 30.36% of medium and 55.03% of large, with the EU average at 19.95% — up 6.47 percentage points on 2024.
Two things fall out of those numbers. First, buying cloud is no longer a differentiator — half of small EU firms already do it, and among enterprises that buy cloud, 85.15% use it for email. Second, AI is genuinely early outside large companies, which means there is still an advantage available to a small firm that picks one use case and does it properly rather than broadly. We walk through that in Practical AI for Small Business: Where to Start.
Why most programmes do not pay
The failure pattern is consistent and it is not technical. BCG’s September 2025 report on the AI value gap found about 5% of companies generating value at scale and 60% reporting hardly any material value despite substantial investment. Gartner, reported by Computerworld in July 2025, predicted that more than 40% of agentic AI projects would be cancelled by the end of 2027 on escalating costs, unclear business value or inadequate risk controls — and noted that of thousands of vendors claiming agentic capability, only around 130 were judged genuine.
Read across those findings and the common cause is scope. Programmes that start with a platform and look for problems to apply it to run out of patience before they run out of budget. Programmes that start with a measured problem tend to finish, because there is a number that tells everyone when to stop.
What separates the 5% from the 60%A project with a number attached ends. A project with a vision attached gets renamed.
Start with the process that hurts most
Rank candidate projects on two axes: hours currently burned, and how contained the fix is. The sweet spot is a lot of wasted hours and a boundary you can draw a circle around. Avoid anything that requires three departments to agree on a new process in month one.
| Project | Adoption evidence | Typical effort | Main failure mode |
|---|---|---|---|
| Cloud email, files and identity | 85.15% of EU enterprises buying cloud use it for email (Eurostat, 2025) | Two to six weeks, mostly migration and user training | Migrating mailboxes without tightening identity, so you move the risk with the data |
| Document and data automation | Text mining is the most-used AI technology in EU enterprises at 11.75% (Eurostat, 2025) | Four to eight weeks for one document type | Automating a process nobody has standardised, so the exceptions outnumber the rules |
| Customer self-service | 26.08% of EU enterprises buy cloud as a platform for building and hosting applications (Eurostat, 2025) | Six to twelve weeks for a first useful slice | Shipping a portal that answers the questions you find interesting, not the ones people phone about |
| AI-assisted support replies | Only about 5% of companies generate AI value at scale (BCG, September 2025) | Three to six weeks with a human reviewing every reply at first | Letting it answer unsupervised before you have measured how often it is wrong |
Pick one. Give it an owner who is not also running the business, a deadline inside a quarter, and a number: hours saved per week, calls avoided per month, days off the invoicing cycle. If you cannot name the number in the first meeting, the project is not ready and no amount of software will make it ready.
Integration beats a pile of apps
The most expensive mistake in small-company technology is not buying the wrong tool. It is buying twelve right tools that do not talk to each other, then hiring a person to copy data between them. Each app looked cheap. The integration debt did not appear on any invoice.
- Check the integration before the features. Does it connect to your accounting system and your identity provider, natively, without a middleware subscription?
- Prefer one system doing two jobs adequately over two systems doing one job each brilliantly and never speaking.
- Insist on an export. If you cannot get your data out in a documented format, you have rented your own records.
- Count the logins. Every tool outside single sign-on is a password someone will reuse and an account nobody will remember to disable.
- Write down who owns each tool. Unowned subscriptions are how companies end up paying for software nobody has opened in a year.
Adoption is the whole game
You can buy the licence in an afternoon. Getting forty people to change how they do something takes months, and it is the part that gets cut when the budget tightens. The World Economic Forum’s Future of Jobs Report 2025 is pointed about this: employers surveyed named analytical thinking as the most sought-after skill, with resilience, flexibility and agility close behind, and the report’s headline recommendation is urgent upskilling rather than more tooling.
The practical version is unglamorous. Train in small groups on real work rather than demo data. Pick two or three people who will be the ones colleagues actually ask, and give them time for it. Keep the old process available for a fortnight so nobody is trapped, then switch it off on an announced date — if you never switch it off, you are running both forever.
Measuring it without a dashboard nobody opens
Measure three things and resist the urge to build a dashboard. Before you start, record the baseline: how long the task takes today, how often it goes wrong, and what it costs. After you ship, record the same three. A spreadsheet with six numbers in it has settled more arguments than any business-intelligence rollout.
- Time. Hours per week on the task, measured the same way before and after.
- Error rate. How often the output needs rework, as a share of the volume.
- Cost. Licences plus the implementation plus the training time, against the hours recovered.
Then be willing to call it. If the numbers did not move after a fair trial, stop and write down why. That note is worth more than the project was, because it stops you buying the same idea again in eighteen months with a different logo on it.
A twelve-week version you can actually run
Twelve weeks, one process, one owner. Weeks one and two: measure the baseline and write the current process down as it is actually performed, not as the handbook describes it. Weeks three and four: choose the tool, check the integrations, and agree the number that defines success. Weeks five to eight: build and pilot with a small group doing real work. Weeks nine and ten: train everyone and run both processes in parallel. Weeks eleven and twelve: switch the old one off, measure again, and write the one-page outcome.
That cadence is deliberately boring, and it is the reason it finishes. Resourcing it is usually the sticking point rather than the plan: Eurostat found 57.5% of EU enterprises that tried to recruit ICT specialists in 2023 had difficulty filling the vacancy, which is why so many of these projects get bought in. In-House IT vs an IT Partner: How to Decide sets out that decision with the costs attached, and Which Emerging Technologies Are Worth Your Investment? covers which newer technologies are worth a slot in the queue.
Eudora runs this work remotely for clients in several countries: we measure the baseline, build the fix, train your team over video and hand over the documentation. The full list of what we cover is on our services page. For on-site work in Sri Lanka, our sister business is at eudora.lk.
Frequently asked questions
How much should a small company budget for this?
Budget per project rather than per year, and expect implementation and training to cost more than the licences. A single contained project — one process, one tool, forty users — is usually a few weeks of specialist time plus the subscription. If a proposal cannot break the cost into those two halves, ask for one that does.
Do we need an AI strategy?
You need a use case and a way to check the output. BCG’s September 2025 research found around 60% of companies getting hardly any material value from AI, and strategy documents were not the missing ingredient. Pick one task, measure how often the output is wrong, and only widen the scope once that number is acceptable.
What if our team resists the change?
Usually they are resisting the extra work of running two processes at once, which is a fair objection. Fix it by setting a date to switch the old one off, training on real work rather than demo data, and making sure the people who answer colleagues’ questions have time allocated to do it.
Should we replace our old system or integrate with it?
Integrate first if the old system does its job and has a documented export or API. Replace when it blocks three or more projects, when it cannot be secured, or when it needs a person to copy data out of it every week. Replacement is the more expensive answer and sometimes it is still the cheaper one.
How do we avoid the project quietly dying?
Three things keep it alive: a named owner who is not also running the company, a number agreed before the work starts, and a deadline inside one quarter. Projects without all three tend to get renamed rather than finished.
Have one process in mind and no appetite for a transformation programme? Describe it to us and we will tell you what the fix would take. Get in touch with Eudora Technology to talk about your project.
.jpg){kind=link}
.jpg){kind=link}