Making AI Investments Measurable: Four Key Metrics for SMEs

Something shifted in the stock markets at the end of July. The big tech companies presented their quarterly figures, and investment sums for AI were higher than ever. Nevertheless, several stock prices fell. Apple reported a record quarter with a 16 percent increase in revenue, yet the stock opened 8.6 percent lower the following day.¹

The market is now asking a different question than twelve months ago: What is coming out of the investment? You should ask yourself this same question for your own company. This article shows four key metrics with which you can measure an AI investment before the next budget is approved.

What Quarterly Figures Reveal About AI Spending

Two figures from the last week of July 2026 tell the story.

The first is from Alphabet. The group reported a negative free cash flow of $5.85 billion for the quarter, the first since its IPO.³ Free cash flow is the money remaining after all investments. Negative means: the expansion of AI currently costs more than the ongoing business brings in. Meta shows the same effect in its profit line. Revenue rose by 28 percent, while profit fell by 14 percent.²

The second figure is from Amazon. The cloud business AWS grew by 37 percent, and the stock gained about 7 percent after hours.⁴ Here, the return on investments was visible, and the market rewarded it.

Both corporations spend similar amounts. What distinguished them was the proof that something tangible came out of it. This distinction will also arrive at your company as soon as the advisory board, shareholders, or bank inquire about the return on AI expenditures.

In SMEs, This Proof Is Often Missing

The Marketing Tech Monitor 2026 surveyed a total of 414 marketing and digital managers in the DACH region at the beginning of 2026.⁵ Three findings from it are important for the budget question.

The first concerns the success rate. 47 percent of respondents state that in their organization, 70 to under 80 percent of transformation and technology projects fail or miss their intended success.⁵ Almost half of the managers thus expect that seven to eight out of ten initiatives will not deliver. They name a lack of specialized know-how as the most common cause, at 56 percent, significantly ahead of too many parallel projects and unclear goals.⁵

The second concerns usage. 68 percent of companies operate a CRM system, but only 8 percent fully utilize its capabilities.⁵ The tool is in place, but the work with it is missing. You pay for this gap every month without seeing it on an invoice.

The third concerns the foundation. Only 6 percent of companies have high-quality data.⁵ The biggest challenge cited by respondents is the ability to even integrate data from separate systems. Without this foundation, even a good model only provides estimates.

Ralf Strauß, editor of the study, summarizes it as follows: "The mere existence of purchased AI software licenses is not yet an indicator of AI maturity."⁵

The survey dates from the beginning of the year. Little has changed in the situation since then, as shown by two newer data points. The KPMG Global AI Pulse for Q2 2026 measures that the proportion of companies using AI in daily operations almost doubled within three months, from 13 to 22 percent. Nevertheless, only a small group can demonstrate a reliable return on investment. According to the same survey, those who fully know the ongoing costs of their AI systems achieve an established ROI five times more frequently.⁶ And at the end of July, the trade press picked up on the Monitor's findings as a current diagnosis: AI projects fail due to organization and structure, rarely due to technology.⁷ Usage is thus growing faster than proof. The following key metrics close exactly this gap.

Four Key Metrics That Make Value Visible

If you want to prove the value of your AI investment, you need figures that you can collect yourself. The following four do not require an analytics project. They can be maintained in a table and updated quarterly.

1. Utilization rate per license. How many of the paid seats actually worked in the last 30 days? The benchmark is the completed task, not the login. The study figure serves as a comparative value: large marketing and sales organizations on average use only about one third of the capacities of their existing applications.⁵ If you are below that, the next license renewal is the wrong expenditure.

2. Time to regular operation. How many weeks pass between the start of a pilot and the point at which a named person responsible operates the system in daily business? Set a deadline, for example, twelve weeks. Pilots that miss this deadline and have no owner will be terminated or restarted. This limits the number of parallel projects, which appears in the study as the second most common reason for failed initiatives.⁵

3. Hours relieved per role per week. Measure based on a specific role, not the company. How long did the clerk need for a process before, and how long now? In our projects, two one-week samples are sufficient: one before implementation and one after eight weeks of operation. Estimated time savings do not count.

4. Data coverage per use case. For the use case, list which data fields the system requires. Count how many of them are complete, up-to-date, and available without manual export. Our rule of thumb from project work: Below 80 percent coverage, the result becomes unreliable, regardless of the model. This metric explains most of the disappointed expectations we encounter.

Four numbers, one sheet of paper, a quarterly rhythm. That's all it takes to start. We describe how such an entry is structured in the article on the AI implementation gap.

The Blind Spot: Management and Team Assess Themselves Differently

A finding from the same study deserves special attention because it explains why measurement often fails to overcome resistance in everyday operations.

53 percent of managers consider their teams to be rather poorly equipped to effectively use data and tools. Additionally, 45 percent assume a rather low acceptance for new technologies among their employees. Employees, however, assess this differently: 49 percent classify their own acceptance of change as high or very high.⁵

The justifications also diverge. Managers point to training that is too general and not very practical. Employees cite unclear instructions, overly complex tools, and a lack of contact persons.⁵

For you, this means: Before approving another training budget, gather perceptions from both sides. Five questions for management, the same five questions for users. Where the answers diverge widely, that's where the actual need for action lies. This work on roles and empowerment is the Rock Organization in our methodology. According to our project experience, it more often determines project success than the choice of the model. How we systematically assess maturity and gaps is shown in the article on AI sovereignty in SMEs.

What You Can Do

1. Step 1: For each paid AI tool, pull the usage report for the last 30 days. Note active users versus paid seats.
2. Step 2: List all ongoing AI initiatives. For each initiative, enter the start date, named person responsible, and status. Initiatives without a person responsible will either be assigned one or stopped.
3. Step 3: Choose a use case and measure the processing time for the affected role for one week. This is your baseline.
4. Step 4: Re-evaluate the four key metrics at a regular interval, for example quarterly, and decide on extension, expansion, or termination based on this.
5. Step 5: Before the next budget approval, clarify the data coverage for the planned use case before purchasing a license.

An addition with an eye on the calendar: Since August 2, 2026, the transparency obligations of the EU AI Act apply to AI-generated content. Anyone currently reviewing AI expenditures should include the labeling in the same process. Details on this can be found in our article on August 2nd.

From Measurement to Decision

Measurability is a basis for decision-making. If you know which four figures to look at, the discussion about the AI budget becomes significantly shorter.

Sources & References

1. 9to5Mac, „Apple stock opens down roughly 10% following mixed Q3 2026 results", 31.07.2026: 9to5mac.com
2. Meta Platforms, „Meta Reports Second Quarter 2026 Results", 29.07.2026: stocktitan.net
3. Search Engine Journal, „Google Q2 earnings show 5.85 billion negative free cash flow", 23.07.2026 (figure from Alphabet's investor documents): searchenginejournal.com
4. Yahoo Finance, „Amazon stock soars after earnings", 30.07.2026: finance.yahoo.com
5. Marketing Tech Lab GmbH, „Marketing Tech Monitor 2026", Survey January/February 2026, n = 414 complete responses (DACH) from 1,562 invited, additionally 2,822 employees surveyed; Editor Dr. Ralf Strauß. Study page: marketingtechlab.de · Results report via absatzwirtschaft, 27.05.2026: absatzwirtschaft.de
6. KPMG, „Global AI Pulse: Q2 2026" (Survey Q2 2026): kpmg.com
7. t3n, „AI projects in marketing: 85 percent fail – and it's not due to technology", 29.07.2026: t3n.de