How do you measure ROI if the manual process was never measured?

You estimate a baseline before automating, even a rough one, and measure the AI process rigorously. With no starting point, any return figure is an opinion.

Is AI’s cost just the models?

No. The real cost includes models, integration, human supervision and maintenance. Ignoring those line items makes a project look profitable when it isn’t.

← All capabilities

ROI & measurement

Phase: Improve

Prove with numbers that AI saves or earns more than it costs.

ROI measurement connects AI to the business: how much it costs to operate and how much it returns in time, cost or revenue. It’s the capability that decides what to scale, what to switch off and what budget to defend to leadership.

What it includes

Baseline before automating: cost and time of the manual process.
Total cost of operating the AI (models, integration, supervision, maintenance).
Return per process: time saved, cost avoided or revenue added.
Portfolio decisions: which agents to scale and which to switch off for underperforming.

When you need it

Leadership asks what the AI investment returns and there’s no clear answer.
You have several agents and don’t know which are worth it.
AI cost grows and isn’t tied to any measurable result.

How it’s measured

ROI per process (return / total cost to operate).
Cost per resolved case vs. cost of the manual process.
Payback: time to recover each agent’s investment.

Common mistakes

Counting gross savings and ignoring the cost of supervision and maintenance.
Having no baseline, so no improvement can be proven.
Keeping agents alive out of inertia even when they don’t perform.

FAQ

How do you measure ROI if the manual process was never measured?: You estimate a baseline before automating, even a rough one, and measure the AI process rigorously. With no starting point, any return figure is an opinion.
Is AI’s cost just the models?: No. The real cost includes models, integration, human supervision and maintenance. Ignoring those line items makes a project look profitable when it isn’t.

Related capabilities

Observability Workflow redesign Optimisation

Recent signals

Trends, practices, tools, cases and market data for this capability, with sources. Updated weekly.

Data

Gartner: el coste de inferencia por flujo agéntico crecerá más de cinco veces hasta 2028

Predicción de Gartner (17-ago-2026), sin línea base pública. Gartner describe una «paradoja de la inferencia»: el coste unitario baja, pero el gasto total sube porque los flujos multipaso consumen muchos más tokens. Citar como previsión, no como medición.

Gartner · 2026-08-17 · E2

Los marcos claros de medición de ROI son el principal acelerador de adopción (30 %)

En la encuesta de Ortec a 400 directivos norteamericanos de logística, el 30 % señala los marcos claros de medición de ROI como principal acelerador, por delante de los casos de pares (25 %) y la integración con los sistemas de planificación (24 %).

DC Velocity (encuesta Ortec) · 2026-01-15 · E2

El 44% escala IA en toda la empresa, pero el impacto en EBIT sigue clavado en el 37%

La proporción de organizaciones que escalan IA a nivel de empresa sube del 38% al 44%, y las que la usan en tres o más funciones del 51% al 56%. Aun así, el porcentaje que atribuye algún impacto en EBIT a la IA se mantiene en el 37%, prácticamente igual que en 2025. A nivel individual el efecto sí se nota: el 80% dice que la IA ha mejorado su productividad y el 50% que le ayuda a tomar mejores decisiones. Además, el 20% declara que los costes operativos de IA —tokens incluidos— están limitando su uso. La brecha entre productividad individual e impacto financiero es el argumento de rediseño.

McKinsey — The state of AI in 2026: On the road to ROI · 2026-08-25 · E2

Missing any of these?

Start with an assessment: which capabilities you have, which you’re missing, and where it pays off most to start.

Talk to us →