For: Luxembourg SME owners deciding how much money, time, and credibility to risk on a first AI project
Maroun AlteklyFounder of MonyTek · Luxembourg SME consultingThe whole idea in one line
Size the loss, not the cost. When you set an AI project budget as a Luxembourg SME, the number that should govern the decision is not the vendor quote minus the grant. It is the one almost nobody calculates: if this returns nothing, am I still fine on Monday? Find that number first, before you choose a tool, a vendor, or a use case, and a first AI bet stops being a gamble you have to defend.
In short: to budget an AI project as a Luxembourg SME, decide what you can afford to lose before you decide what to build. Set a cash ceiling, a time ceiling, and a credibility ceiling, then choose the smallest project that fits under all three. This is the affordable-loss test, and it answers the real question behind “how much does AI cost”: what happens to you if the pilot returns nothing.
Here is the problem, and it is a real one. Search for an AI project budget and almost every result hands you the same thing: a setup fee, a monthly licence, a reminder that a grant can cut the bill by 70%. All of that answers what will it cost. None of it answers the thing that actually keeps an owner up after a good demo: what happens to me if this does not work? The quote sells you the upside in one clean number and leaves the downside completely unsized. So you end up committing to the wrong figure entirely.
“Don’t ask what the project costs. Ask what you can lose (in cash, in management time, and in credibility with your own team) and still be completely fine on Monday morning.”
That reframe is not a slogan. It is a named decision principle. Affordable loss comes from Saras Sarasvathy’s research on how expert entrepreneurs actually decide: you gate a commitment on what you can survive, not on a forecast you cannot really defend (Society for Effectual Action). It is the opposite of expected-return reasoning. For a first AI bet, where you genuinely cannot forecast the payback, it is the right discipline: decide the loss first, then go looking for the use case.
I will not invent a client story to dress this up, so let me give you the pattern instead, because I have watched it more than once. The damage from a bad first AI bet is rarely the money. An SME can usually survive losing a few thousand euros. What it survives less easily is the second cost: the owner stood up in front of the team, said “this is the future, we are doing AI,” the pilot stalled, and now every future change is met with folded arms. In a company of fifteen people, that costs you for years. So when I size a first bet, I am not only protecting the bank balance. I am protecting the owner’s ability to ask the team to believe them next time. The same gap between intent and follow-through is why AI interest stalls before execution in so many firms.
Short answer: no. SME Package — AI reimburses 70% of eligible costs, but only after the work is done and the bill is paid. So the grant does not lower what you risk up front. Your real exposure is the full pre-VAT outlay of a EUR 3,000 to 25,000 project, not the post-grant figure. Size your loss against what you pay first, not what you eventually keep.
That funding is real and generous, and it is the reason this bet is worth making at all: used well, you run a contained pilot and recover most of the cost later (Guichet.lu). The trap is that one word, after. Because the money comes back later, owners treat the post-grant number as the risk, size up, and quietly commit more than they can afford to lose.
EUR 3,000 - 25,000
Project value (excl. VAT) required to qualify for SME Package - AI (Guichet.lu).
70% reimbursed
Share of eligible costs returned by the Ministry of the Economy, after the work, as de minimis aid (Guichet.lu).
You front 100%
You pay the full bill first and recover 70% later, if it qualifies. That outlay is your real exposure.
Read the third card again, because that is where the napkin maths lies. The grant arrives after the work, so it does not reduce what you risk up front. For affordable-loss sizing, the honest number to put in your “can I lose this?” box is closer to the full pre-VAT outlay, not the comforting post-grant figure. The funding mechanics, eligibility, and how SME Package-AI compares with Fit 4 AI are covered in our Luxembourg AI funding guide for SMEs. This article assumes you have read that and focuses only on the sizing decision.
Two things are genuinely different here. First, the credibility ceiling is higher-stakes, because the market is small and people remember. The business community that decides things is a few thousand people, and they end up in the same rooms. An over-announced pilot that flops is not a private loss. It can become a story that arrives in the next room before you do. Second, the funding runs through a pre-analysis with the House of Entrepreneurship of the Chamber of Commerce, or the eHandwierk department of the Chamber of Skilled Trades and Crafts for craft businesses (Guichet.lu). That process is deliberate. The one tax this market really charges is patience, so your time ceiling has to account for the runway, not only the build.
Short answer: the affordable-loss test sizes a bet by what you can survive losing, not by what you hope to gain. You set three ceilings first (cash, time, and credibility), then pick the smallest AI project that fits under all three. If nothing fits, you shrink the scope or you wait. That is the whole discipline.
I will be straight with you: this part is not the easy bit. Most owners do it the other way round. They fall in love with a use case, then try to justify the loss. The test inverts that, and it is uncomfortable, because the bet is gated by what you can bear rather than by what you hope to gain.
How much could you write off entirely?
The euro figure you could lose without touching payroll, tax, or your operating buffer. Use the full up-front outlay, not the post-grant figure. The 70% comes back after the work, if it qualifies.
How many owner hours can it eat?
The hours per week the owner and the team can give before the project starts starving the business that pays for it. In an 8 to 25 person firm, the one person who understands everything is already busy. This ceiling is low.
How visible can a failure be?
How loudly you can announce this, and to whom, before a stalled pilot becomes a story people repeat. In a small market the social cost of an over-announced flop lasts longer than the cash cost.
Only after all three are written down do you go looking for the project. If the smallest sensible version of a use case still breaches one ceiling, the answer is not to stretch the ceiling. It is to shrink the bet, or to wait. The mistake I will never repeat in my own work is overriding my gut for an exciting opportunity. The room is loud, the upside looks enormous on paper, everyone is nodding, and something inside says no, and you talk yourself past it. The affordable-loss test is a formal way to let the gut win without calling it a feeling. If a project only makes sense when you assume the best case, that is the loud room. Choosing which workflow to test is its own discipline. Our guide to practical AI adoption for Luxembourg SMEs covers picking the one workflow worth sizing a bet around.
Short answer: small enough to lose without pain. Pick one repetitive workflow that hurts every week, then scope the smallest version that fits under your cash, time, and credibility ceilings. For most small firms that means a contained pilot in the lower half of the EUR 3,000 to 25,000 band, not a company-wide programme.
A composite, not one real client
Every euro figure in this article is meant as illustrative planning examples, not benchmark claims. The owner below, “Marc,” is a composite of owners I meet, not one real client: same firm shape, same pressure, no real person behind it. The only verified figures are the SME Package — AI funding rules sourced to Guichet.lu.
Picture Marc, who runs an owner-led services firm of about fifteen people. He is the hero of this story, and the bet is his to size. After a good demo, the easy move is to budget on the napkin: list price, minus 70%, looks like almost nothing. Marc does the harder thing and sets the ceilings first. Cash: an amount the firm could write off without touching payroll or the tax buffer. Time: a few hours of one manager’s week, and almost none of Marc’s own. Credibility: a quiet pilot known to two people, not a company-wide “transformation” announcement.
Only then does Marc choose the workflow: a repetitive, reviewable document-triage task that eats hours every week. The smallest version that fits under all three ceilings is scoped. Because it lands inside the EUR 3,000-25,000 band, Marc starts the SME Package-AI pre-analysis so 70% can be reimbursed later (Guichet.lu). I have left the figures deliberately vague, because the point is the order of operations, not a fake benchmark I would have to invent.
My job in Marc’s story is not to be the hero. It is to be the guide, and the gift I bring is this method. I have earned the right to hand it over. Years ago I helped introduce mobile banking to SMEs when it first launched, and the thing that decided whether it worked was never the technology. It was whether the owner had sized the change to something the business could absorb, and whether they were in the room while their people adjusted to it. A first AI bet is the same. The test does not calculate the return; sizing the loss and sizing the upside are two different jobs. For the payback side, use the four-criteria scorecard in our guide to automation ROI for Luxembourg SMEs. ROI maximises the expected return; affordable loss bounds the downside you must survive if that return never arrives. You want both.
Short answer: it is the right size when you could lose the full up-front amount and still be fine on Monday. Judge the options by survivability, not by sticker price. A contained bet you fully understand beats a big one you only took because the grant covered 70%. The three sizes below show the difference.
Same owner, same firm, same funding. Only one of the three passes the affordable-loss test. The euro figures below are an illustrative example, sized only against the EUR 3,000-25,000 SME Package — AI band (Guichet.lu), not a quoted benchmark.
Sized up because "the state pays 70% anyway."
Verdict: Wrong size. The up-front EUR 24,000 is what you actually risk, and the number was chosen by the subsidy, not by what you can bear. I would rather see an owner do a small bet they understand than a big one they only took because the state was paying most of it.
Sized to fit under all three ceilings, one clear workflow.
Verdict: Right size. You could lose all of it and still be fine on Monday, and you understand exactly what you bought. That is the only kind of first bet worth making.
No workflow hurts weekly. The only driver is fear of being left behind.
Verdict: No bet yet. Fear is not a use case. Find the painful, repetitive workflow first, then size a bet around it. Waiting one quarter to find the real problem beats spending now on a guess.
Seventy percent off is a strong pull. As an example, it makes a EUR 20,000 project feel like a EUR 6,000 project, so owners size up: bigger scope, more ambition, “while we are at it.” But the loss you must be able to bear is still the up-front outlay, because the 70% reimbursement comes back after, and only if it qualifies (Guichet.lu). The funding is a reason to execute the bet better, not a reason to make it bigger. Subsidised does not mean safe.
Short answer: agree the stop rule before you spend. Write down, out loud, the result that means scale and the result that means stop. Then run a simple 30-day plan that fixes the workflow, the baseline, and the smallest scope before a single cheque is signed, so a failure stays contained.
A bet you cannot stop is not a bet, it is a commitment, and a commitment you cannot walk away from is how the cash loss quietly turns into the credibility loss. Here is the sequence I would run in the first 30 days, before writing a single cheque.
Week 1
No tools yet. Write the three ceilings as real numbers. Find the one workflow that hurts every week.
Week 2
Define what a result looks like: the baseline today (hours, rework, delay) and the single number that proves it worked.
Week 3
Scope the smallest project that fits under all three ceilings. If it lands in the EUR 3,000-25,000 band, start the SME Package - AI pre-analysis.
Week 4
Agree the stop rule out loud: the result that means scale, the result that means stop. Then start.
By day 30 the goal is a bet you can survive losing, not a transformation programme you have to defend. If the only justification for the project is fear of being left behind, there is nothing to size yet, so find the painful workflow first. And when the bet is genuinely strategic enough to consider building rather than buying, our guide to AI build vs buy for Luxembourg SMEs applies the same survivability logic to that decision.
Short answer: you end up strong whether the pilot works or not. The cash, time, and credibility were all under your ceilings, the stop rule was set in advance, and the grant improved execution rather than ambition. Win or lose, you keep the standing to ask your team to back the next bet.
That is the payoff, and it is the reason the discipline is worth it. Here is where the owner who sizes first actually lands:
The cash, time, and credibility on the table are all under their ceilings. A complete failure is a survivable Monday, not a crisis.
You agreed in advance the result that scales and the result that stops, so the decision is already made before emotion arrives.
You used the 70% reimbursement to do a contained bet well, not to justify a bigger one you did not need.
Win or lose, the team watched you size a risk responsibly, which is what lets you ask them to believe the next, bigger bet.
That is not caution. That is how you earn the right to a second, bigger bet. The owner who can fail completely and still be fine on Monday is the owner the team will follow into the next decision.