Someone sends you a portfolio. It has screenshots of workflow canvases, a list of tools and the word automation several times. Here is how to read past it.
Question 1: did it run in production?
Ask how long it ran, how many records went through it, and who used it. The answers separate a demo from a deployment immediately.
Things that ran for months have stories: the week the API changed, the edge case that only appeared at volume, the migration that had to happen without downtime. Things that ran for an afternoon have screenshots.
Good answer: "Eighty three days, about twenty nine thousand records, used by reps across twenty eight locations. Two outages, both upstream, and here is what we changed after the second."
Pause on: a confident description with no numbers, no duration and no incident.
Question 2: who operates it now?
This is the question that predicts what you are buying. If the honest answer is nobody, or only them, you are looking at a dependency rather than a system.
Good answer: "The operations team runs it. There is a runbook. I get called about once a quarter."
Pause on: "It still works, I check it sometimes." That usually means it decayed and nobody noticed.
Question 3: what business result can you trace?
Not executions. Not messages sent. A result, with a path back to it.
The strongest version is someone walking you backwards: money received, to the deal, to the handoff, to the conversation, to the list it came from. People who have built attribution talk in stamps and fields. People who have not talk in dashboards.
Good answer: includes a number, the baseline before it, and how the connection was established.
Pause on: volume as the headline. Twenty thousand calls is not a result, and anyone who has carried a number knows it.
Question 4: what did you decide not to build?
The most useful question in the whole conversation, because restraint is the expensive skill and it cannot be memorised.
Good answers are specific: this stayed manual because the process was still in debate. That needed a human approval because it was irreversible. This data was not clean enough to automate against, so we fixed the data first. The guidance from the people building these systems is the same: start with the simplest thing that works and add autonomy only where it earns its cost.
Pause on: "I would automate anything." It sounds like ambition and it reads, in twelve months, as a cleanup project.
Two smaller tells worth noticing
How they talk about failure. Ask for an automation of theirs that failed silently. Everyone has one. What matters is how they found out, because that answer tells you whether they build alerting or hope.
Whether they ask you questions. A strong candidate will want to know about your data before your goals. The ones who present a solution before seeing a record are selling.
What a portfolio should contain
If you are advising someone on what to show you, it is short: the problem in the company's language, what they did, what it produced, and who runs it now. Four paragraphs beat forty screenshots.
Profiles built this way
MitHub publishes open profiles where the work is written out in exactly that shape, with the technologies attached to each deployment rather than listed as a wall of logos. Reading them and contacting anyone is free.