Back to the cases

Hauling new trucks across Mexico and the United States

Their own development team, able to build it without me

One of the companies that moves the most new trucks in North America owns its own development company. It had plenty to build with; what it lacked was deciding what would actually serve the operation. We started with a workshop for its programmers, built together, and today its team carries on alone.

The goal

That its technology area deliver artificial intelligence solutions faster and with an impact on the group’s profitability, instead of buying whatever the vendor of the month advertises. And, in concrete terms, that the operation could ask questions about its own data without waiting for someone in IT to build a report.

How it worked before

• The operation ran on a robust in-house system, and every question outside the expected ones depended on IT building a report by hand.
• During discovery, their own people named the weak spots: processes known to only one person, errors when entering routes and travel expenses, and the difficulty of knowing what had been fully invoiced.

What we did

• November 2024: with leadership and the heads of development, projects and infrastructure, we went area by area (quality, shop, planning, logistics and finance) and sorted the ideas in a value versus effort matrix.
• December 2024 to January 2025: the workshop with their programmers. Artificial intelligence models running on their own machines, containers, agents that coordinate with each other, databases and reading scanned documents. At the close, each programmer presented their own exercise, and one of them already connected an assistant to their database.
• 2025: that exercise grew into the assistant that answers questions about the operation today. The split went like this: I designed the architecture, their team built the services that feed the data from their system, and my team built the agents that understand the question, answer it and review it.

The piece that makes it trustworthy is an evaluator that grades every answer before it goes out, and every answer shows the query and the data it came from, so anyone can verify it. The database is read-only.

The impact, measured with CREO

I measure the impact of every project with four questions, which I call CREO: did the company become more Competitive? Did it improve its Return, meaning its profitability? Is it more Efficient? Does it meet its Obligations better, the tax, legal and contractual ones? Not every project hits all four, and here are the ones that apply.

Competitiveness

Their technology area went from watching artificial intelligence from the outside to having its own, built by them and running on their own server.

Efficiency

The operation asks in Spanish about trips, units and travel expenses, from the chat, from Telegram or with a batch query form, without asking IT for a report.

Obligations

Every answer leaves a trail of the query that produced it, and the assistant has read-only access to the company’s information.

Return

The models run on their own account, at cents per query, and with no per-user licenses.

The result

• Their technical staff validated it first, and then their operations staff tested it with real questions taken from their daily work. The first version got four out of ten right. Every error they reported was corrected against the figure they expected, until the answer matched.
• The assistant went into production with four ways to query it, and the project was delivered and paid in full.
• Since mid-2026 their team operates and expands the automation platform on its own, and builds new workflows without me.

More with less

What stayed in the company is the capability, beyond the system itself: a team of its own that understands how a trustworthy assistant is built and already builds without depending on a consultant. For a group that already pays for its development area, that multiplies what the area delivers with the same people.