Back to the cases

Freight transport · in progress

Process redesign and automation, pushing together

For a carrier on Mexico’s northern border, fuel is “the blood running through the veins” of the company, in the words of its chief financial officer. Even so, nobody could answer with data whether the liters paid for matched the kilometers driven. An outside firm redesigned its processes and, in parallel, we built the system that holds them up.

The goal

To know, every week instead of every month, how much fuel each unit consumed, whether that consumption is reasonable for the kilometers and hours it worked, and which job and which company in the group should be charged for it. Their chief financial officer described the existing process as “fragmented, disjointed and basically manual.”

How it worked before

• The fuel station and the units’ GPS already talked to each other, and accounting was left out of that conversation. Every fill-up was typed into the accounting system, record by record.
• Four pieces of data were outside every system: engine hours, the driver, the job and the signature on the ticket. Two people entered them by hand, from photos that arrived through WhatsApp groups.
• What the fuel truck dispensed at the job sites was kept in a paper log.
• Fuel efficiency was calculated by hand and with two different criteria, and the financial report arrived a month late.
• When one month of data was reviewed, seven out of ten fill-ups had the odometer at zero or one.

What we did

Before a line of code was written, every piece was sorted: what belongs to a person, what rule-based automation solves, what calls for machine learning and what calls for generative artificial intelligence. Anything that touches money goes into deterministic code, meaning that the same data always produces the same result.

• July 2026: five on-site sessions with their people to map the process as it really happens. Their team validated thirteen of thirteen points in the diagram, and that produced the scope: six modules.
• August 2026: a read-only connection to accounting, a portal of their own and, as its home page, a fuel efficiency traffic light for each unit with six rules anyone can explain.
• September 2026: field capture from the driver’s phone, which also works without signal. The driver takes a photo of the odometer, the artificial intelligence reads the number and the driver confirms it. We trained the drivers and the other offices. One stage was delivered four days ahead of the deadline.
• October 2026: field capture started for real; fuel station fill-ups come in on their own every night; and the system produces the file that accounting imports with its official template, with Mexico’s excise tax (IEPS) and VAT broken out. Each role has its manual inside the same portal.
• November 2026: formal handover, with its test checklist, and the move to a dedicated machine in their own data center.

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

Management went from waiting for the monthly report to getting a cutoff every week. In the words of their chief financial officer: “if I have a weekly cutoff, it lets me make decisions and correct deviations before we get to the financial report.”

Return

Every liter is tied to a unit, a job and the company in the group that should pay for it, and the system shows in pesos what is still pending review. An entire location, which was outside every report, came under control.

Efficiency

The sources come in on their own, with no typing. The driver captures from their own phone and the office stops transcribing WhatsApp photos.

Obligations

The system writes to accounting only through the official template and with its own series of record numbers, leaves electronic invoicing untouched, and breaks out the taxes, verified against the invoice. Every source file keeps its digital fingerprint, and cancelling a fill-up leaves a record of who, when and why.

The result

• In its first three weeks, field capture reached 87 out of every 100 fill-ups at the station. In the practice week it was 46.
• The first full week matched the fuel supplier’s invoice to the liter.
• At one of the locations, the station and accounting differ by one tenth of a percentage point.
• The system has already flagged purchases at other gas stations with no matching dispatch, entries that arrived hours late, and a gap between what was loaded into a fuel truck and what it delivered.
• Once the rules were calibrated with their people, false odometer alerts dropped to less than a third.
• Out of fourteen requests from their people in one week, eleven were resolved the same day.

The hard rule of the project: the system flags, and the final call is always a person’s. Their chief financial officer said it better: “very good alert data will be worth nothing to us if, consciously or unconsciously, we decide to do nothing with it.”

More with less

Six modules in operation in about eight weeks, with a team of two people. It made use of what they already had: the station, the GPS, their email accounts and the drivers’ phones. The way into accounting was proposed by their own outside accountant, and the project’s open questions went from twenty-eight to two.