From pilot to production: how to measure the value of AI so that Finance accepts it
In Latin America, AI is no longer an experiment for technology companies. According to a BCG report dated November 2, 2026, 85% of the region’s tech companies already have AI agents in production. In addition, one third sell AI solutions to their customers. But only 22% financially validate the impact of AI (BCG, 2026).
If that happens in the most advanced companies, the challenge is greater for everyone else. We have already explained why to measure the return on generative AI (Wasivia, 2026). Here we focus on how to measure it so that the Finance team accepts it.
What the studies show
| Finding | Population | Source |
|---|---|---|
| 85% have AI agents in production; only 22% financially validate the impact of AI | Technology companies in Latin America | BCG, 2026 |
| Only 18% have formal AI governance frameworks | 461 organizations surveyed in 2025, 91% in Latin America | RSM, 2026 |
| 40% cite the difficulty of building business cases with measurable economic impact as the main barrier | 130 organizations in Spain and Ibero-America | NTT DATA and CIONET, 2026 |
| Only 25% of AI initiatives delivered the expected return, and 16% were scaled across the enterprise | 2,000 CEOs in 33 countries | IBM, 2025 |
Sources: BCG, 2026; RSM, 2026; NTT DATA, 2026, in Spanish; IBM, 2025.
The conclusion repeats itself: using AI is easy; proving its value is not. For BCG, the next leap will depend less on technology. It will depend more on CEO ownership, on transforming the operating model and on disciplined management of generative AI costs (BCG, 2026).
Why Finance does not accept the numbers
Teams often present results that Finance cannot validate:
- There is no baseline. If no one measured how long the process took before AI, any improvement is an opinion.
- Hours are counted, not money. “We saved 200 hours” does not say whether those hours went to productive work or whether costs went down.
- Everything is attributed to AI. A sales increase during peak season does not prove the assistant caused it.
- Costs are missing. The license is counted, but not usage, integration, team time or maintenance.
How to measure the value of AI in five steps
- Define the baseline with Finance. Before starting, agree on what will be measured, how and with which data: time per case, cost per transaction, errors, sales or collections. Having Finance sign off on the starting point avoids arguments later.
- Link each operational indicator to a financial one. Less time per inquiry matters if it translates into fewer overtime hours, more cases handled by the same team or better sales. Write down that chain for each use case.
- Isolate the effect of AI. Compare against a group that works without AI over the same period, or measure before and after in comparable seasons. Without comparison, there is no attribution.
- Count all costs. Licenses, model usage, integration, team hours, training, supervision and maintenance. Net value is the benefit minus all of that.
- Review every quarter. Present results in the same format, adjust or stop what does not perform and scale what does. A use case without periodic review becomes a fixed expense.
What changes with AI agents
Agents act on their own, and that makes measurement harder. According to the NTT DATA and CIONET observatory, 80.8% of organizations in Spain and Ibero-America lack operational mechanisms to control the autonomy of their agents (NTT DATA, 2026, in Spanish). Without a log of what each agent does, there is no way to know how much value it generates either.
That is why it pays to design a log of every action from the start, with its cost and its result. It serves control, auditing and value measurement at the same time.
What it means for companies in Ecuador
The BCG figures refer to technology companies, the most advanced in the region. Elsewhere, measurement is usually weaker. But the logic is the same for an Ecuadorian company of any size: a pilot that is not measured by Finance’s criteria will hardly get the budget to grow.
The good news is that measuring well does not require large systems. It is enough to agree on the indicators at the start, keep the before-and-after data and review it with discipline.
What to do now
- Choose an AI use case that already works and ask whether anyone can prove how much it is worth.
- Bring Operations and Finance together to agree on the baseline and the indicators.
- List all costs, including those that do not appear on the vendor’s invoice.
- Define how you will isolate the effect of AI: a control group or a comparison between periods.
- Schedule a quarterly review with an explicit decision: scale, adjust or stop.
At Wasivia, we design dashboards that link the operational indicators of AI to financial results through our data analytics and business intelligence service. We also help you prioritize and measure your use cases through our AI consulting and strategy.
References
- BCG — Turning AI Ambition into Reality (Nov 2, 2026)
- RSM — AI readiness in Latin America survey report 2026 (Sep 3, 2026)
- NTT DATA and CIONET — Agentic AI Observatory 2026 (Jun 16, 2026), in Spanish
- IBM — IBM Study: CEOs Double Down on AI While Navigating Enterprise Hurdles (May 6, 2025)
- Wasivia — Generative AI ROI: from enthusiastic pilot to measurable value
