Your AI project saves hours. Does it save cash?

A business buys AI software to cut the time spent preparing quotations, processing orders or checking documents. The pilot works. Tasks become faster. Management multiplies the hours saved by the hourly staff cost and presents an attractive return.

There is one missing step: the payroll may be unchanged.

The employees are still there, their salaries are still due and the extra capacity may not produce additional sales. A useful operational improvement can therefore have a weak cash return. For an owner deciding what to fund, or an investor evaluating a Portuguese company, that distinction changes the investment case.

Two ECB analyses published this month make the question timely. The 2 October post reports that 72% of euro area firms planning AI investment expect to use internal funds. This is a share of firms reporting a financing intention, not 72% of investment expenditure. Respondents could select several funding sources, and the survey does not establish realised returns. [1]

The 6 October analysis separates businesses adopting existing tools from those developing AI technology. Their investment needs and risks differ. Its sector comparisons are euro area evidence, not a financing offer for an individual Portuguese SME. [2]

For an established company using existing tools, the first financing question is whether the project produces cash before it consumes the cash needed elsewhere.

The European Commission's 2026 Portugal Country Report records AI adoption of 11.54% in 2025, against an EU comparison of 19.95%, rounded in the text to 11.5% and 20%. [3] Eurostat's enterprise survey covers businesses with at least ten employees or self-employed persons in specified sectors. It does not represent every Portuguese microbusiness. [4]

Those figures suggest room for wider adoption. They do not tell an investor which company can earn a return from it.

A distributor with repetitive order processing may have an identifiable process to improve. A consultancy may gain capacity but struggle to sell more work. A manufacturer may accelerate quotation preparation while production remains the real bottleneck. These are illustrative possibilities, not observed company cases.

The opportunity has to be tested at workflow level.

Start with technical time saved: how much faster does a defined task become?

Then measure usable capacity: how much of that saving survives review, exceptions, coordination and the way work is organised?

Finally, establish cash conversion: how much usable capacity reduces an actual cash expense against a credible baseline?

This last step might mean less paid overtime, lower contractor invoices or an approved hire no longer needed. If staff use the time to win and deliver additional business, model that separately through incremental contribution and collection timing. Extra revenue is not all profit, and profit is not automatically cash.

Counting the same hour as both a cost reduction and a growth opportunity overstates the return.

Consider a hypothetical Portuguese business with a twelve-person office team. Each person has 1,600 relevant working hours a year. A quarter of that time is spent on suitable tasks, and the proposed tools cut those tasks' time in half.

That releases 2,400 hours on paper:

12 x 1,600 x 25% x 50% = 2,400.

Assume 60% becomes usable capacity after review and operational constraints. The result is 1,440 hours. At an assumed loaded cost of EUR30 per hour, its annual capacity value is EUR43,200.

The project requires EUR36,000 of incremental implementation cash and EUR18,000 a year to operate. These are invented inputs for the model, not supplier quotes, Portuguese wage benchmarks or market prices.

Hold the technical performance constant. Change only the share of usable capacity that becomes a cash expense avoided.

The technology performs equally well in all three cases. The cash outcome differs because the company uses the released capacity differently.

At 25% conversion, the expense reductions do not cover running costs. The project could still improve service or resilience, but those benefits would need their own evidence. This model supplies no cash payback.

At 70%, it generates EUR12,240 annually once fully operating. That is useful, but far below the headline EUR43,200 capacity value.

The 100% case requires every usable hour's cost equivalent to become an expense avoided. Treat it as a demanding sensitivity, particularly when time savings are dispersed across twelve people.

Assume implementation cash is paid at the start, running costs apply throughout year one and that year's gross savings reach only half the steady annual level. Benefits then remain constant.

In the 70% case, year-one project cash is:

-EUR36,000 - EUR18,000 + EUR15,120 = -EUR38,880.

The company recovers EUR12,240 in each subsequent full year. Cumulative cash remains negative by EUR2,160 at the end of year four. With even benefits thereafter, payback occurs roughly 4.2 years after the project starts.

Dividing implementation cost by steady annual net benefit alone produces about 2.9 years. That calculation omits the first-year rollout cash deficit.

There are two useful thresholds. Cash conversion must exceed 41.7% simply to cover steady running costs. To recover implementation and all running costs by the end of year three, it must reach 83.3%, given the assumed half-benefit first year.

Neither threshold is a market benchmark. Both are consequences of the stated inputs. Higher operating costs, another implementation bill or slower adoption would raise the hurdle.

Ask the operating manager and finance team to agree a short evidence file.

First, establish the baseline: task volumes, time, error rates and the invoices or payroll items that could change. Separate existing salaries from incremental cash expenditure.

Second, test complete outputs. A faster first draft is insufficient if review and corrections absorb the saving.

Third, name the cash action. Specify the contractor work reduced, overtime avoided or planned recruitment deferred, with timing and an accountable owner. Savings against a planned hire are credible only if that hire was actually needed and budgeted.

Fourth, map monthly project cash alongside the company's other commitments. In the example, EUR36,000 of implementation spending understates the first-year funding requirement. The business must also absorb operating costs while savings build.

That file supports a more useful financing conversation. A small pilot may fit internal cash if it leaves adequate liquidity for the business. Borrowing requires testing repayments against the timing and downside of actual cash benefits. An equity discussion becomes more relevant when adoption sits inside a larger expansion or transformation with risks the existing balance sheet cannot comfortably carry.

Calling an expense "AI" does not establish a financing need, lender acceptance or investor interest.

For a buyer, establish whether the purchase forecast already includes these savings. If the seller's price assumes the benefit and the buyer's return model adds it again, the same improvement is being counted twice.

An adoption gap can identify somewhere to look. The investment decision needs a named workflow, a measurable improvement and a cash action that survives the cost of implementation.

This is a nominal incremental cash model before tax, financing, inflation, discounting and unmodelled transition costs. Excludes residual value, grants, revenue growth, working-capital changes and terminal benefits. Payback is not NPV or a return recommendation. VAT and tax timing should be added for a real company. Avoided hires require a documented credible baseline. Benefits from extra revenue require separate incremental margin, collection and cost modelling; do not double-count the hours.