Most SMEs don't have an AI problem — they have a prioritization problem

Talk to enough small and mid-sized businesses about AI and a pattern shows up quickly: the tools aren't the bottleneck. Off-the-shelf AI features are now bundled into email clients, accounting software, CRMs, and customer support platforms — access is no longer the barrier it was a few years ago.

The actual bottleneck is deciding where to apply automation first, in a business that doesn't have a dedicated data science team or an unlimited software budget. Get that wrong, and you end up with an expensive chatbot nobody uses, or a "digital transformation" project that stalls after the kickoff meeting.

This article lays out a three-stage framework — crawl, walk, run — built around one rule: never automate a process you don't already understand well enough to explain in plain language.

Stage 1 — Crawl: narrow, boring, high-frequency tasks

The best first automation projects are the ones nobody would put in a pitch deck. Look for tasks that are:

  • Repetitive — done the same way, dozens or hundreds of times a month.
  • Rule-based — a human could write down the steps as a checklist.
  • Low-risk if imperfect — an occasional miss doesn't damage a client relationship or create a compliance problem.

Typical crawl-stage candidates for an SME:

  • Sorting and tagging incoming support emails or contact-form submissions by topic.
  • Extracting data from incoming supplier invoices or receipts into a spreadsheet or accounting system (OCR + structured extraction).
  • Auto-generating first-draft responses to common customer questions, reviewed by a human before sending.
  • Reconciling two systems that should agree but currently require someone to manually cross-check (e.g., orders in an e-commerce platform vs. entries in accounting software).

The point of this stage isn't to save enormous amounts of time yet — it's to build internal confidence and internal competence: someone on your team learns how to configure, monitor, and correct an automated process before you trust it with anything higher-stakes.

Stage 2 — Walk: connecting systems, not just tasks

Once a business has two or three crawl-stage automations running reliably, the next opportunity is usually integration between systems that currently don't talk to each other — a CRM and an invoicing tool, an e-commerce platform and inventory management, a support inbox and a ticketing system.

This is where automation starts compounding instead of just saving isolated pockets of time, because the value comes from eliminating the manual re-entry and reconciliation between tools, not just speeding up work inside one tool.

Warning signs you're not ready for this stage yet:

  • Nobody on the team can currently describe, start to finish, how data moves between your core systems today.
  • Your crawl-stage automations still require frequent manual correction.
  • You're considering this because a vendor pitched it, not because you identified the specific reconciliation pain yourself.

Stage 3 — Run: decision support, not just task execution

The most advanced (and most misunderstood) stage is using AI for decision support — demand forecasting, pricing suggestions, churn-risk flagging, prioritizing which leads a sales team should call first. This is where generic AI hype usually starts the conversation, and where most SMEs should actually finish it, not start it.

Decision-support AI is only as good as the operational data feeding it. If your business hasn't been through the crawl and walk stages — meaning your data is scattered, inconsistent, or trapped in disconnected systems — a "run" stage AI project will produce confident-sounding recommendations built on unreliable inputs. That's a worse outcome than doing nothing, because it looks like progress while actively degrading decision quality.

The governance question nobody skips in Belgium/EU

Because Robust Code operates from Belgium, this needs saying plainly: any automation touching customer or employee personal data needs to be evaluated against GDPR obligations from the start, not retrofitted afterward. Practically, that means before automating a process:

  • Know what personal data the automation touches, and why it's necessary.
  • Confirm whether the AI tool or platform processes that data outside the EU, and under what safeguards.
  • Keep a human in the loop for decisions with meaningful impact on individuals (e.g., automatically rejecting a job application or a credit request is a very different risk category than auto-tagging a support email).

This isn't box-ticking — a genuinely useful automation project should be able to explain its own data handling in one paragraph. If it can't, that's a signal to slow down.

A simple prioritization exercise

Before adopting any tool, list your candidate processes and score each on two axes:

  1. Frequency — how often does this happen per month?
  2. Time cost per instance — how many minutes/hours does a human currently spend on it?

Multiply the two. The highest-scoring items are your crawl-stage candidates — not the process that sounds most impressive to automate, but the one quietly consuming the most hours every month. This is deliberately unglamorous. It's also usually right.

If you want to put real numbers behind this instead of a gut estimate, our Automation ROI Calculator turns hours-saved-per-week and hourly cost into a payback estimate in about thirty seconds.

What to actually avoid

  • Buying a platform before identifying the process. The tool should follow the prioritization, not the other way around.
  • Skipping straight to "run" stage projects because they sound more impressive in a board meeting than "we automated invoice data entry."
  • Automating a process you don't understand. If nobody on the team can currently explain the process end-to-end without checking, automating it will encode the confusion, not remove it.
  • Treating AI adoption as a one-time project. The businesses that get real, compounding value treat it as an ongoing operating habit — a new candidate process every quarter — not a single initiative that gets a press release and then stalls.

Where Robust Code fits

We help SMEs run exactly the prioritization exercise above — honestly, including telling a client that a proposed AI project isn't worth doing yet — before recommending any specific tooling or building anything custom. If you want a second opinion on where automation would actually move the needle in your business, get in touch.