Can we measure it?
Hours, errors, delays, lost revenue. If nobody can say what the process costs today, nobody can prove it improved. The good news: in most companies the work is concrete and close at hand. You can watch it, time it and count the errors.
Is it worth it?
A task that eats a few hours a month doesn't need an agent. One that eats half a person's week or more usually does, and so does one whose mistakes and delays are costly even if it takes less time.
Does it repeat?
Agents earn their keep on volume: quotes, reports, document packs, inbox triage, onboarding. Work that comes back every day, in roughly the same shape, with the same small judgment calls.
Can it be built safely?
The data has to be reachable and mistakes have to be catchable. Where judgment matters, the agent hands the decision to a person. Some processes should stay human, and we'll say so.
Is someone in charge?
We need an owner, CEO or COO who wants the result and stays involved. "Let's explore AI" with no owner is a no, because nobody will be there to decide when the demo is ready.
Will the team work with us?
We sit with the people doing the job. If they can't spare a little time, the system won't fit how they actually work. It doesn't take much of it, but it has to be real.
Miss one and we pass
Not yet is an honest answer.
Miss one check and we pass, or we come back when it changes. It saves both sides months.
When all six line up, we measure the process with your team and build a working demo on it. You see it at work, with a proposal and a clear ROI, before you commit to anything.
Think you have a process that passes? Tell us about it. We'd rather hear the messy version.
Written by
Henrique Milli
Founder of HMD Digital. An engineer who ships production systems where failure is expensive, and builds and manages MidasCV, Ultravisible and Docable.
