Skip to content
Baguals

Perspectives · November 17, 2026

Why every AI transformation starts with one workflow

Big AI programs promise to change everything at once and often change very little. One workflow in production teaches a company more than a year of planning.

By Equipo Baguals

There is a familiar way to bring AI into a large company. A steering committee forms. Consultants map dozens of use cases. A roadmap appears with three horizons and a budget for each. Months go by, the slides get better, and the operation keeps running exactly as it did.

We work the other way around. We start with one workflow, get it into production in weeks and only then talk about the next one. Here is why.

A transformation is hard to measure

When the goal is to “transform the company with AI,” almost anything counts as progress. A pilot that never left the lab, a dashboard nobody opens, a chatbot that answers three questions. None of it is clearly a failure, so none of it is clearly a success either.

One workflow is different. You can name it: customers asking about the status of their shipment, invoices typed into the ERP, payment reminders sent by hand. You can measure it before and after: how long a response takes, how many cases get solved without a person, how many hours the team gets back. And you can agree on that metric with leadership before anyone writes a line of configuration.

When the metric is clear, the conversation changes. It stops being about whether AI is the future and starts being about whether this agent is doing this job well.

Production teaches what planning can’t

Every operation has rules nobody wrote down. The supplier who always sends invoices with the wrong tax ID. The customer who calls instead of writing. The exception that applies only at the end of the month. None of these show up in a workshop. They show up the first week an agent handles real cases.

That is why getting to production fast matters more than getting the plan perfect. A workflow running on real data, with real customers, surfaces the exceptions, the missing rules and the places where a person really does need to decide. Those lessons are what make the second workflow faster and better than the first.

A long planning phase delays that learning. Worse, it often designs around an operation that only exists on paper.

Small scope, real control

Starting with one workflow also makes control simpler. It is easier to decide which actions need human approval when you’re looking at one process instead of twenty. It is easier to read the log of every action, review the conversations and see where the agent hesitated. It is easier to set a spending cap and know what it covers.

For a leadership team that has never had an AI agent working inside the company, that matters. Trust is built one visible result at a time, not with a promise about the whole organization.

What one workflow looks like in practice

Imagine a freight company where customers write on WhatsApp to ask where their container is. Today, someone on the team reads each message, looks up the booking in a spreadsheet exported from the operating system and writes back. On a busy day, answers take hours.

A first workflow could be exactly that question and nothing else. The agent reads the message, looks up the booking in the same data the team uses, answers with the status and hands the conversation to a person when something doesn’t add up. Before it goes live, the company and the team building the agent agree on what success looks like, for example the share of questions answered without a person and the time to first response.

That is a small scope. It is also a real one. If it works, the company has a measured result, an agent its team knows how to supervise and a clear idea of what comes next: maybe freight quotes, maybe arrival notices.

Expansion earns its way in

Once the first workflow is measured, expanding becomes a business decision instead of an act of faith. What the agent learned, the rules, the exceptions, the documents, carries over to the next process. The team already knows how to supervise. Leadership already knows what a result looks like.

This is how we think about scale: not as a big launch, but as one solved problem after another, each one making the next one easier.

The question to ask

If you’re weighing how to bring AI into your operation, try replacing “what’s our AI strategy?” with a smaller question: “which workflow eats the most time, and how would we know if an agent did it well?”

The answer to that question is a better first step than any roadmap.

Let’s talk about your operation.

Get in touch