Skip to content
JMA AI, artificial intelligence division of JMA Integra
ESEN

Company

Who answers for this, and with what proof

The group behind the system, the public evidence about the landscape, and where your operation fits.

Who we are

The artificial intelligence division of the JMA Integra group

JMA Integra brings together the group’s lines of business: private security, canine units, cybersecurity, logistics and project management. JMA AI grew out of the cybersecurity practice, not the other way around: first we learned to protect a company’s information, and then to put it in order and automate it. It’s a difference of origin, and it shows in how the work is built.

The starting point

Three public figures, with their sources in plain sight

None of these figures are ours: they come from public agencies and peer-reviewed research, with source and year. They describe the landscape; they don’t measure what we do.

In the same survey, 28.2% track between three and five indicators and 35.2% reported tracking none at all.

A company that doesn’t record has no history to give any model. Producing that record is the work that comes before any foresight.

Self-reported in a survey, sample of 22,188 companies. Most recent published edition of the exercise.

17.4%

0100%

of senior management’s time, at mid-sized Mexican companies

World Bank · Enterprise Surveys, Mexico country profile2023 survey covering 1,322 companies

Time that management reports spending on regulatory requirements: taxes, customs, labor, licensing and registration.

That’s why we start by taking manual work away from people who have no margin left, instead of handing them one more dashboard to attend to.

Time as declared by management in interviews, not measured with a stopwatch. General regulatory burden, not one specific filing.

17% to 33%

0100%

of answers, in AI tools working on an organization’s own documents

Magesh et al. · Journal of Empirical Legal Studies, vol. 22Published in 2025, data collected in 2024

Three commercial tools working on the organization’s own documents returned 17% to 33% of answers that were incorrect or cited sources that did not support the claim.

Connecting AI to your documents reduces error; it does not eliminate it. That’s why every answer arrives with its origin and the decision stays with a person.

202 queries designed to be demanding, graded by hand. The authors caution that this is not a population rate. US legal domain, in English.

In a separate evaluation, only 51.5% of sentences were fully supported by their own citations. A visible citation is not the same as a verified one.

Liu, Zhang and Liang · Findings of the ACL: EMNLP 2023

Every claim on this site states what date it is from and who stands behind it. If a source changes edition or corrects its figure, the claim comes down until it has been checked again.

Where it fits

Chosen by process, not by industry

The system is designed around the vocabulary, the steps and the screens of your operation: it fits wherever there is a recurring process, not wherever an industry says so.

  • 01

    Regulatory obligations and filings

    Permits, licenses and expiry dates, with monitoring of what changes in the regulation that applies to you.

  • 02

    Case files and document control

    Integration, versions, owners and control of what is missing from each file.

  • 03

    Operational follow-up

    Indicators, deviations and real progress, with a permanent view for management.

  • 04

    Administration and support areas

    Data entry, reconciliation, inboxes and recurring reports that consume hours of qualified work.

The same order serves a ten-person firm and a thousand-person maquiladora. And we work where your operation is: in your language, on your schedule and, when the project calls for it, on site.

Start with the process that costs you most

You don’t need a diagnosis of the whole company: one specific process is enough.