- Is it worth continuing to invest in AI without knowing what pays off?
- What the data says: how widespread is the invisible ROI problem?
- Where does AI really work and where does it still not?
- How much is the AI market growing in Italy and worldwide?
- Three concrete scenarios: what happens without metrics in an Italian company?
- Who actually checks AI content before it reaches customers?
- How do you build an AI business case that survives the budget meeting?
- What are the most common mistakes made by those who approve AI budgets without metrics?
- How to figure out if your AI project is actually working?
- So, is it worth setting up AI metrics before spending?
- How we see it at SHM
- Frequently Asked Questions
- 40.2% of marketing leaders fail to prove the financial value of AI investments, according to the Enterprise Martech Outlook 2026.
- Those who measure ROI get growing budgets 56% of the time; those who don't stop at 37.5%.
- In Italy, only 8% of companies have defined complete metrics to evaluate the impact of digital, despite growing investments.
- The solution is not to spend less on AI, but to set up KPIs and a business case before kicking off any project.
Two-thirds of large companies already have a dedicated AI budget, but 40% of those managing these investments cannot prove a single euro of return. Are you about to approve the next AI expenditure without knowing how you will measure the result?
Is it worth continuing to invest in AI without knowing what pays off?
No. Data shows that those who don't measure ROI get less budget and risk blocking useful projects along with useless ones. Setting up the metrics before spending is the move that separates those who scale from those who stall.
Approving an AI budget without a tracking system is like opening a shop without a cash register: money goes out, but you have no clue how much comes in. The issue isn't AI itself, but how most companies adopt it: they spend first, then try to figure out if it was actually useful.
The Enterprise Martech Outlook 2026 analyzed by MarTech it states with precise numbers: 40.2% of martech leaders fail to demonstrate a clear and measurable financial contribution. Yet budgets keep growing anyway.
| Your situation | Should you measure right away? | why |
|---|---|---|
| You are evaluating a new AI project | Yes | Defining KPIs before launch is much easier than adding them later |
| You already have active AI projects without metrics | Yes | Without data you risk cutting what works and keeping what doesn't |
| Your team uses AI only for internal uses (texts, reports) | Yes | These are the easiest cases to measure: time saved, fewer errors |
| You are scaling AI agents toward customers | Yes | The risk of errors increases: metrics become a quality safeguard |
| You have AI budget but no governance manager | Yes | Without ownership, costs grow and results remain invisible |
What the data says: how widespread is the invisible ROI problem?
The problem is global and very widespread. Almost half of marketing teams cannot prove the value of AI, and in Italy the situation is even more critical.
95.3% of large organizations have AI agents (autonomous programs that perform tasks without human intervention) on their roadmap. But only 10.8% have brought them to full production. The rest are still in the planning, testing, or limited production phase, according to the Enterprise Martech Outlook 2026.
The gap between those who measure and those who don't has a direct impact on your wallet: teams that demonstrate value get budget increases in 56% of cases. Those that rely on vague promises stop at 37.5%. The difference is nearly 20 percentage points.
In Italy the picture is even more fragmented. According to Politecnico di Milano Observatories , only 8% of businesses have defined metrics to fully evaluate the impact of digital, despite investments being on the rise. Spending without measuring is not a strategic choice: it is a hard habit to break.
Where does AI really work and where does it still not?
AI performs best on internal and repetitive tasks. Where processes are already complex or data is messy, AI amplifies problems instead of solving them.
69.3% of respondents say AI has a clear or substantial impact on internal marketing operations. Only 59.9% say the same for the customer experience. Almost a 10-point difference that shows where AI is mature and where it is not.
Internal uses — writing, editing, data analysis, reporting, creative variations — are easier to test, control, and integrate into existing workflows. Agents that handle customer journeys, automated decisions, audience selection, and campaign creation are still lagging behind.
| Type of AI usage | Maturity status | Measurable ROI? |
|---|---|---|
| Assisted writing and editing | In advanced production | Yes |
| Data analysis and reporting | In advanced production | Yes |
| Data management and cleanup | In advanced production | Yes |
| Video generation | In advanced production | Yes |
| Customer journey optimization | Still in development | No, for now |
| Automated decisioning | Still in development | No, for now |
| Audience selection | Still in development | No, for now |
| Autonomous campaign creation | Still in development | No, for now |
There's a principle to keep in mind: if the process is already chaotic, the AI agent will just make it chaotic faster. Before automating, you need clean data and stable workflows. This applies whether you're an e-commerce store looking to automate post-purchase emails or a manufacturing company thinking about trusting AI with order management.
To dive deeper into how AI agents are spreading across sales and marketing teams, also read how some companies are unifying budgets and teams around AI agents .
How much is the AI market growing in Italy and worldwide?
The market is growing at a very fast pace, but the ability to measure results is not keeping up. Those who invest without structure risk funding experiments with no return.
In Italy, the AI market reached 1.8 billion euros in 2025, with 50% growth compared to the previous year, driven 46% by GenAI (generative artificial intelligence, the kind that produces texts, images, and code), according to the Artificial Intelligence Observatories of the Polytechnic University of Milan .
Globally, enterprise GenAI spending has tripled: from $11.5 billion in 2024 to $37 billion in 2025, according to data from Menlo Ventures on about 500 US enterprise decision-makers . Numbers that make the question even more urgent: what is this money for?
In Italy, 71% of large companies started at least one AI project in 2025, but only 1 in 5 uses it pervasively across multiple business functions. Among SMEs, the share drops to 8%, according to the Politecnico's own Observatories. The gap between those who have started a project and those who have truly integrated it is the clearest sign of the measurement problem.
Three concrete scenarios: what happens without metrics in an Italian company?
Without defined KPIs, even a working AI project risks being cut at the next budget cycle. These examples show the real cost of lacking measurement.
Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.
| Company type | AI project started | Without metrics | With metrics defined beforehand |
|---|---|---|---|
| E-commerce (fashion) | AI for automatic product descriptions | Unquantified time savings, no conversion data | Time saved per card: -8 min. Conversion rate: +4%. Budget confirmed. |
| Professional firm (accountants) | AI for document summarization and drafts | Sporadic use, no comparison with the previous method | Billable hours recovered: +2/week per team member. Positive ROI in 3 months. |
| Manufacturing company | AI agent for supplier request management | Untracked errors, hard to attribute savings to AI | Supplier response time: -30%. Order errors: -15%. Project scaled. |
Who actually checks AI content before it reaches customers?
Most companies keep a human in the loop for customer-facing AI content. But this oversight has a cost that must be included in the ROI calculation.
Only 1.6% of companies allow fully automated AI content towards customers. 43.3% allow it only after human review, editing, and verification. 24.4% limit generative AI to internal use only, according to the Enterprise Martech Outlook 2026.
This creates a paradox: AI should boost productivity, but if the time saved in creation is spent double-checking the output, the net gain is zero. Or worse, negative. The cost of human review needs to be factored into the business case right from the start, not discovered afterwards.
To understand how to manage AI agent governance without losing control, it's also helpful to read the real risks of AI agents publishing content without consent .
| Company policy on AI content | Share of companies | Impact on ROI |
|---|---|---|
| Fully automated AI content toward customers | 1,6% | Potentially high ROI, high quality risk |
| External AI only after human review | 43,3% | ROI reduced by review costs, controlled quality |
| AI for internal use only | 24,4% | ROI easier to measure, limited external impact |
| No defined policy | Rest of the sample | Unmeasurable ROI, governance risk |
How do you build an AI business case that survives the budget meeting?
A solid business case starts with KPIs, not tech. Figure out what you're tracking first, then pick the tool.
The problem isn't that AI doesn't work. It's that most projects start with the solution ("let's use AI for X") instead of the question ("what problem do we want to solve and how do we measure it?"). This backward approach is the main reason for invisible ROI.
To avoid the same mistake described in the article on token governance and ROI in corporate AI budgets , the business case structure must follow a precise path.
- 1 · Management / Sponsor Define the problem to be solved Write down in one sentence the process you want to improve and the expected result. Without this, no KPI makes sense.
- 2 · Operations team Choose 2-3 measurable metrics Time saved, errors reduced, conversions increased, costs lowered. Metrics that already exist in your systems, not new ones to invent.
- 3 · Operational team + IT Record the baseline Before turning on AI, measure your starting point. Without a baseline, you can't prove improvement.
- 4 · AI vendor / Internal team Start a limited pilot Test on a subset of processes or clients. Gather data for 4-8 weeks before scaling.
- 5 · Management Present data at the budget meeting Compare baseline and pilot results. If the delta is positive and measurable, you have the business case. If not, you avoided waste.
What are the most common mistakes made by those who approve AI budgets without metrics?
The same mistakes keep happening, no matter the industry. Catching them early is the fastest way to dodge them.
The topic of AI results quality is closely tied to this: as the article on explains the AI “good enough” that costs a lot in marketing , lowering the quality threshold without measuring it leads to damage that is hard to quantify later.
| Error | Why it happens | How to avoid it |
|---|---|---|
| Starting from the technology, not the problem | Pressure to “do AI” before understanding why | Write down the problem in one sentence before evaluating any tool |
| Do not record the baseline | It is taken for granted that improvement will be seen | Measure the current process before turning on AI |
| Choosing overly vague KPIs | “Productivity” and “efficiency” are not metrics | Use numbers: hours, errors, costs, conversions |
| Do not include the cost of human review | Only time saved in production is counted | Add verification time to the net ROI calculation |
| Scaling before validating the pilot | Enthusiasm or vendor commercial pressure | Wait for pilot data before expanding the project |
| Do not assign a governance manager | It is assumed that "the system thinks for itself" | Assign an owner for each active AI agent |
How to figure out if your AI project is actually working?
There are simple metrics you can check every month without fancy tools. The trick is to set them up before you start.
| Indicator | How it is measured | Positive signal | Red flag |
|---|---|---|---|
| Time saved per task | Comparison of before/after hours on a sample | Steady reduction >15% | No difference or increase |
| output quality | Rate of necessary reviews/corrections | Revisions dropping over time | Revisions stable or increasing |
| Cost per unit produced | Total process cost / output volume | Cost dropping with stable volume | Cost stable or growing |
| Team satisfaction | Short monthly survey (3 questions) | Improving score | Resistance or abandonment of the tool |
| Impact on business KPIs | Conversions, errors, response times | Positive delta compared to baseline | No measurable delta after 8 weeks |
So, is it worth setting up AI metrics before spending?
It is worth it if you already have active AI projects or are about to approve a new one: defining KPIs and baselines upfront is the quickest way to protect your budget for the next cycle.
A middle ground you are still evaluating whether to adopt AI: start with a pilot on a measurable internal process (writing, reporting, data analysis) and build the business case from real data.
Not yet if your company doesn't have stable digital processes yet: AI will amplify the existing mess. Stabilize workflows first, then consider automation.
How we see it at SHM
For us, AI is a tool, not an end goal. And like any tool, its value is measured by what it produces, not by how much it costs or how much buzz it generates. We believe that integrating AI into a company without first defining metrics is the digital equivalent of buying machinery without knowing what it will produce.
We've worked with companies that had already started AI projects without a measurement framework. The starting point is always the same: figure out what concrete problem you want to solve, define how you'll measure the improvement, and only then pick the tool. Not the other way around.
If you are trying to figure out how to set up governance and track your AI investments, our service for strategic AI integration starts right here: business case, KPIs, and pilot before any scaling. For those who also want to train the internal team, the path of AI training for businesses helps build the skills needed to evaluate and manage projects independently. If instead you need to analyze already available data to build your business case, the service of Data processing and business intelligence can be the right starting point.
You can find more practical insights on AI adoption in business in the topic area AI and process automation .
Frequently Asked Questions
Start by recording the baseline today, before turning on the AI. Even a week of manual measurement of the current process is enough to have a point of comparison. Without a baseline, any improvement is a guess, not a fact.
The simplest ones are time saved per task (hours before vs. after), the number of revisions needed on the output, and the cost per unit produced. These are metrics that already exist in your systems and don't require new tools.
For internal uses (writing, analysis, reporting) the first signs show up in 4-8 weeks. For customer-facing AI agents, timelines get longer because more complex variables come into play. A 6-8 week pilot on a limited process is the fastest way to get real data.
Yes, always. If the team has to check every AI output before publishing or sending it, that time has a cost. An ROI that doesn't include human review is overestimated and hard to defend at the next budget meeting.
The most effective format is simple: problem to solve, current baseline, target KPI, project cost (including human review), expected result in 6 months. One page is enough. The important thing is that the numbers are real, not optimistic projections.
That's a useful outcome, not a failure. It means you avoided scaling an investment that wouldn't have paid off. Figure out why the pilot didn't work: unstable starting process, poor data quality, or the wrong tool. Then decide whether to fix it or stop.
The AI budget isn't the problem: the problem is getting it approved without knowing how to prove its value six months later.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
