Mehdi Bakkali
Founder — Flowmetrik
Eight years at the intersection of AI, data and large accounts. Designs the data foundation, the agents, and their path to production.
BNP Real Estate · Société Générale · KPMG
FlowEnergy — building data & AI
Energy savings across a portfolio, proven site by site.
We bring your building data — assets, meters, invoices, weather — together in a single foundation. We put AI agents to work on it. And we train your teams to use it, so they no longer need us.
AI is only worth the data you feed it: that is why the data foundation comes first. FlowEnergy is the Flowmetrik unit dedicated to buildings — layered on the tools you already run, deployed in your environment, hosted in France.
The problem
The data already exists. It is scattered across six tools that do not talk to each other.
The utility holds the invoices, the facility manager holds the BMS, the technical lead holds a spreadsheet, the engineering firm delivered a PDF report. Everyone is right in their own corner, and nobody can answer the only question that matters: did the retrofit deliver what it was supposed to deliver?
On the portfolio we track today, that answer used to live in a thirty-minute Excel macro — per site.
What we do
Four workstreams, and they hold in this order.
You cannot govern data you have never measured, and a platform without a trained team dies within six months. That is why the four travel together.
Measure01
Prove a saving instead of estimating it.
An energy saving cannot be read off an invoice: a mild winter manufactures it, a cold winter erases it. We compute a weather-adjusted baseline, compare it to actual consumption, and display model quality next to the result.
IPMVP protocolDegree daysR² and CV(RMSE)Savings in MWh and in euros
Comply02
Obligations, read out of the documents and then tracked.
Energy-efficiency mandates, environmental certifications, contractual commitments: the deadlines sleep inside PDFs nobody re-reads. An AI layer extracts them, attaches each to the right site, and tracking becomes a campaign with reminders rather than a spreadsheet.
Document extractionDeadlines per siteCompliance campaignsAudit trail
Govern03
A foundation the rest of the organisation can query.
Assets, leases, meters, works, contractors: the same objects live in five systems under five identifiers. We map what exists, then set a target model and exposed views with proper access rights — not one more warehouse.
Source mappingTarget data modelExposed viewsAccess rights and traceability
Enable04
A platform nobody uses is worth nothing.
Upskilling is part of the delivery, not an option: tracks by level from executive to daily user, a library of validated skills, a usage charter, and a point of contact reachable every week.
ExecutivesManagersUsersChampions
The proof
A national portfolio, measured meter by meter.
The IPMVP protocol says how to prove an energy saving. Doing it across a whole portfolio is a data problem before it is a thermal one.
Public reference · French State property
AGILE — the French State real-estate management agency. A platform analysing energy savings across the State portfolio, in service.
Measuring the energy savings of a national portfolio, site by site.
The IPMVP calculation, until then a thirty-minute Excel macro, became an online laboratory: a degree-day-adjusted baseline, model quality shown next to the result, controlled exports. In parallel, an AI layer reads regulatory documents, extracts obligations and deadlines, and tracks risk-control campaigns — OCR and a sovereign model, data hosted in France.
Scale of the tracked portfolio: several tens of thousands of sites and meters, millions of invoice lines, around a hundred weather stations.
ARP AstranceSustainable real-estate consultingEnvironmental certification, from the document pack to the report.
A BREEAM In-Use tool that indexes the certification scheme and the project documents, and proposes for every criterion an answer sourced to the document that supports it. Alongside it, an AI-lead engagement: usage charter, credit management, reusable business skills, and team upskilling in waves.
Initial certificationRenewalPortfolio assessment
Engagement in progress · around a hundred people supported · three workstreams in parallel
Real-estate asset managementEngagement in progress
The target data architecture, before buying one more tool.
Design of the real-estate data model and the governance that goes with it, then a map of AI use cases prioritised into an executable backlog — enough to decide what to build, in what order, and what not to build.
Target modelGovernancePrioritised backlog
A real client, unnamed — permission to cite is asked for beforehand, never after
The agents
Three tasks nobody should still be doing by hand.
Meter readings, work orders, engineering reports: this material is produced by people who have no time to re-enter it. An agent is only useful if it takes over a precise, timeable, repeating piece of work.
Here are three — and we say which one already runs.
01
Reading a planning document
Working out what the local planning code allows on a plot takes half a day of reading — and starts over with every asset.
The agent reads the regulation, extracts the rules that apply to the plot and writes the report, quoting the paragraph behind every point.
Live at a client
02
Watching a fleet of meters
A consumption drift is discovered quarterly, when the invoice lands — and by then it is too late to understand why.
The agent compares each meter to its own weather-adjusted baseline and raises only the deviation that clears the noise, with the site context attached.
Components already shipped elsewhere — not yet deployed as such
03
Preparing a committee memo
Preparing a committee means hand-collecting figures that already exist, across four different tools.
The agent gathers, formats and sources every figure back to its origin. The memo goes to a human for review — it never leaves on its own.
Components already shipped elsewhere — not yet deployed as such
What FlowEnergy is not
Saying it upfront saves everyone time.
A technical department that hears “AI and energy” thinks first about what it has already bought, and about what it is not allowed to let out. So here is the list, first — and it is written into the contract.
- We do not replace your BMS, your asset-management software or your engineering firm. We plug into them.
- We do not rent out a platform. The system is built for you, deployed in your environment, and the code stays yours at the end.
- We do not announce a saving we have not measured: a gain is proven against a weather-adjusted baseline, or it goes unsaid.
- No data leaves France unless you decide it does — sovereign models, separate environments, isolated secrets.
- No agent decides in your place. It reads, extracts, compares and alerts; the decision stays with you.
The support
Software ships in weeks. Adoption has to be built.
Flowmetrik is an AI consulting firm: the platform and the team upskilling belong to the same contract, because neither holds without the other.
AI and data audit
What exists, what is usable, what is blocking. A prioritised and costed map of use cases, and a backlog you can start on Monday.
Training by level
Executives, managers, daily users, champions: four distinct tracks, worked on your own files rather than on generic examples.
Agents in production
Usage charter, security audit, production release rules, a library of validated skills, and a point of contact present every week.
48 people trained in a single session
At ARP Astrance, a first session reached 48 people, followed by waves targeting the teams that produce the deliverables — around a hundred people supported in total. The AI lead has held a weekly clinic ever since.
How we work
A workshop, an assessment, then production.
Every stage ends with something you can look at. If the assessment says the data is not there, we say so — that outcome is planned for, and it is cheaper than finding out later.
2 hours
Frame
A workshop, your sources on the table, one business question to settle. You leave with a scope and a go / no-go, not with a specification document.
4 to 6 weeks
Prove
An assessment on a real scope: is the data there, is it reliable, does the calculation hold? This is where you find out what is genuinely missing.
Then continuously
Ship to production
Connectors, access rights, tests, logging, team training. In your environment, under your security rules — and with a handover planned.
Sovereign by default
Portfolio data says where the buildings are, what they consume and what is planned for them. It stays in France, in your environment, on open and documented components. See how.
Who delivers
A FlowEnergy engagement rests on two skill sets.
Data and modelling on one side, building physics on the other. A computed saving is only worth something if the two agree — otherwise it is a number, not a saving.
Alexandre Mertz
Energy efficiency engineer — consultant
Dynamic thermal simulation, audits and feasibility studies. Checks that the statistical model stays consistent with how the installation actually behaves.
ECOME Ingénierie · Politecnico di Milano
Contact
One portfolio, one question, and the data you already have.
Describe the portfolio and the question in three lines. We come back with what would need connecting, how long it takes, and whether it is worth it.
Mehdi Bakkali — Flowmetrik · mehdi@flowmetrik.com