AI adoption and ROI: what can really be measured, and how.
Companies know which AI tools they have bought. Almost never do they know how far the tools have entered the work, what they changed and whether the result is the expected one. Delphis gives a method to answer, without promising more than the data allow.
A demo on your use cases, with our team and no commitment.
The gap between licenses and results
A license bought is not adoption, and adoption is not a result. Between the two steps, three pieces of information get lost that usually no one collects.
How much it is really used
Vendor console data covers company accounts only. Use on personal accounts stays invisible and makes any estimate based on licenses alone unreliable.
What changed in the processes
Without a measurement before AI is introduced, every comparison afterwards is an impression. Baselines and shared criteria are missing.
What changed for people
AI changes the work, and how people perceive it conditions adoption. It is rarely read in a structured way.
Three levels of measurement, kept apart
Adoption, impact on processes and impact on people need different methods. Mixing them produces numbers that look precise and are not.
Demo data, not real 1. Real adoption
It integrates the usage data read from vendor APIs and traces it back to users, roles and departments. Targeted questionnaires bring out shadow AI, that is use on personal accounts. The result is a map of who uses what, for which use cases and where AI has not yet arrived.
- Usage by department, role and use case
- Estimate of use outside company licenses
- Different tools for the same use case
Demo data, not real 2. Impact on processes
It is measured well on bounded processes. A baseline is set before AI is introduced, the process is observed afterwards and expected and realized results are compared, use case by use case.
- Baseline before, comparison after
- Expected against realized
- Register of use cases and processes involved
Demo data, not real 3. Impact on people
Surveys and sentiment indicators read how the work changes and how ready people feel. Data is always aggregated by role or department, never assessments of the individual employee.
- Aggregate sentiment by role or department
- Readiness and AI skills
- No individual assessment
An honest method, including about its limits
- Impact on broad, loosely defined processes cannot be measured precisely. We say so and treat it as an estimate, not as data.
- Data about people stays aggregated by role or department. It is a product choice, consistent with the Italian Workers' Statute, GDPR and relations with trade union representatives.
- Which usage data is available depends on what each vendor exposes in its enterprise APIs. In a demo we assess it case by case.
What each role gets
- CEO
- A concise reading of how much AI is producing against the investment.
- Chief AI Officer
- Use case portfolio, adoption status and priorities.
- CIO
- Where adoption is low relative to spend, and which licenses to review.
- HR
- Training, AI literacy and impact on people.
Request a demo
A demo on your use cases, with our team and no commitment.
Frequently asked questions
Can you calculate a single ROI for all of a company's AI?
Not credibly. Impact is measured on bounded processes and read together with adoption and spend. A single number across the whole organization hides more than it explains.
How is use on personal accounts detected?
With questionnaires that bring out shadow AI, alongside vendor usage data. It is an estimate, not an exact count.
Is data about people used to assess individuals?
No. It is always presented in aggregate form by role or department.