Forward-deployed AI · Paris

The reasoning layer your ERP never had.

Fieldreason builds AI systems for multi-site operators: plants, camps, platforms, vessels, depots, clinics. We keep every site stocked, compliant and audit-ready, and we ship it live inside your environment in weeks.

Site 114 · resupply runlive
event · 06:12Supplier out of stock on reference 40-118. 12 units short for next month's configuration.
1
ResolveEquivalent found: 40-118B, same specification, on the approved list. Alternative 41-002 held in reserve.
2
VerifyDestination rules checked: item permitted, licence on file, no controlled-item constraint.
3
ReplenishSix-month window applied against stock and expiry: 9 units needed, not 12.
4
OverseeSubstitution routed to a human for sign-off. One decision, with the reasoning attached.
Proposed order ready · audit trail attachedhuman on exceptions
Who we work with

Dozens or hundreds of sites, far from head office.

The ERP holds the record. The people hold the reasoning: what each site should have, what expires first, what may legally cross which border, which alternative is acceptable when the exact item is gone. That knowledge sits in heads, spreadsheets and email. We put it in software.

industryPlants
energyCamps
offshorePlatforms
maritimeVessels
logisticsDepots
servicesClinics
What the ERP can't do

Stocked. Compliant. Audit-ready.

Three promises, one system. Rules where rules work, models where judgment is needed, and a human on every exception.

01 · resolve + replenish

Stocked

Match the exact item, then the equivalent, then the acceptable alternative. Compute what each site needs from what it holds and what expires. Propose the order. A person only touches the exceptions.

02 · verify + route

Compliant

Check every line against the rulebook: specifications, licences, certifications, controlled items. Route it legally across borders, with the paperwork the auditor will ask for already attached.

03 · oversee

Audit-ready

Every decision explained, every exception logged, every override attributed to a person. Controls and traceability are built in on day one, not added after the first audit finding.

How we work

Map the jobs. Build in six weeks. Run it.

We do not sell workshops or prototypes. We put two or three jobs live inside your environment, then stay to run them and add the next ones.

Map the jobs

We map the actual work, not the org chart: the jobs to be done, the rules, the documents, the systems, and where it fails today. Ranked by return, readiness and effort. One to two weeks.

Foundational Build

Two or three jobs live, integrated with your systems of record, controls included. Not a pilot. Four to six weeks from kick-off to production.

Run and extend

We operate what we built and bring the next jobs onto the same foundation. Every build makes the next one cheaper. You subscribe to a system that runs, not to a licence you have to make work.

you own

Your data, your integrations

Everything specific to your operation is yours: the data, the rules we encode for you, the integrations into your systems. It lives in your environment.

we own

The reusable modules

The resolution, verification, resupply, routing and oversight components we bring in. They get better with every build, and every client benefits.

you subscribe to

A system that runs

We operate it and keep it right as your suppliers, rules and sites change. If you ever leave, you keep your data and integrations. We take our modules.

The method in detail

Where it runs today

Our first client operates 260 remote sites, each holding 500 to 650 references, supplied through thousands of orders a year. Half the sites send no product codes. The reasoning was in people's heads. We are putting it in software.

260remote sites in the first deployment
500–650references held per site
6 weeksfrom kick-off to a live build
1human decision per exception, not per line
Principles

How we think about the work.

  • No workshops, no prototypes.We build live systems that run in your environment. A pilot is a way to postpone a decision.
  • Map the actual work.Not the org chart, not the system diagram. The jobs, the rules, the failure points.
  • Rules where rules work.Deterministic code for what is deterministic. Models only where judgment over lots of information is needed.
  • Humans on exceptions.The system proposes and explains. A person decides the hard cases, and only those.
  • Controls in from day one.Audit trail, attribution and traceability are part of the build, not a compliance retrofit.
  • No change programme required.We plug into the systems you have. If a vendor needs your operating model to change first, it is the wrong vendor.