Limbral · Digitalization and AI adoption for firms and municipalities

Your sector's digitalization frees up scale.
The first firm through takes it.

Every sector that digitalizes frees up a surplus of scale: margin, customers and speed that couldn't be served before. That surplus does not get shared. It goes to the first firm that arrives with its systems ready, and from there it buys market share at the price it sets itself.

Limbral takes established firms and municipalities from talking about AI to running systems in production: a diagnosis in two weeks, the first system live in sixty days, and an in-house team that no longer needs us the following year.

Request the two-week diagnosis See the work
First throughThe rest of the sectorWindow open
The thesis

Three things that already happened in every sector and every administration that digitalized before yours.

Digitalization frees up capacity — it doesn't share it

When a process goes from manual to automatic, the operating ceiling moves up. A company serves more customers and more products with the same people; a municipality resolves more procedures, and resolves them without the resident having to come in and ask what happened to theirs. That new ceiling exists for everyone at once, but only whoever has the system running actually occupies it.

The first mover sets the bar for everyone else

The firm that gets there first doesn't just win volume: it wins the cost benchmark, the expected response time, and the data it learns from. In local government the same thing happens in a different currency: whoever digitalizes first sets what residents, the press and the province will expect of the municipality next door, and arrives at the next budget discussion with its own numbers instead of estimates. Whoever arrives later isn't competing against the system — they're competing against its learning curve.

The window closes on adoption, not on technology

The technology becomes available to everyone on the same day. What separates them is how many months an organization — a company or an administration — takes to change the way it decides. That is where the window is lost, and that is where we work.

We don't sell tools or licenses. We work on the distance between what your organization could already be doing and what it actually does at ten o'clock on a Tuesday morning.

What we do

A short path, with dates, that ends in production.

Four moves. Each one has a date and a deliverable you can use the day it lands. If a move doesn't produce a result, it stops there: we don't drag projects along.

  1. Week 0–2

    Threshold diagnosis

    We map where your sector sits on the curve, who has already moved, and which decisions inside your firm are still made by hand. You walk away with three opportunities ranked by return and by adoption friction — and with the one we ruled out, and why.

  2. Day 15–60

    First system in production

    We pick the opportunity with the best ratio of impact to resistance and build the whole thing: data, model, interface and the process around it. Not a pilot, not a demo. Something a team uses to do its work and won't want to go back from.

  3. Month 2–6

    Real adoption

    A system nobody uses costs more than never having built it. We redesign the process, train the people who will use it every day, and measure usage rather than licenses: how many decisions went through the system this week, and how many hours came back to the team.

  4. Month 6–12

    Your own engine

    We leave the capability installed: the people, the data architecture and the judgement to pick the next project without us. A successful Limbral contract is one where the next contract is smaller.

If the data isn't inside, we go out and get it

A good part of what a firm decides does not live in its own systems: what the competition is doing, how prices move across the sector, what came out in the official gazette, what a public registry holds, what is happening around each branch. We do not work only with what you already have loaded. When it is needed, we build the ingestion of those external sources, normalize them and cross them with your internal data — every figure traced to its source, its URL and its capture date, so it can be audited.

The last case below is exactly that: the firm did not have its competitors' prices and there was nowhere to buy them. We built the ingestion of geolocated prices — which product, which branch, which day — and made it available to query in real time: the pricing models consume it, and so does anyone on the commercial team who wants to see how a particular market looks this morning.

The cases below started there: with a piece of data the client did not yet have.

What we built

Three times the window was open and someone went through it.

The third one we built on our own initiative, from public sources. All three are systems actually running in production.

Mass-market retailArgentina · nationwidePricing · Forecasting · Demand planning

Shelf prices, recalculated every day

A retail chain was competing against rivals who changed prices by hand, once a week, looking at three or four stores. Nobody in the sector knew what was happening that same day to the price of a product across the rest of the country. Neither did they.

What we built

First, what nobody had: the price of every product, every day, in every supermarket in the country, store by store. On that base, demand forecasting and dynamic pricing — models that learn how people react to each price change and anticipate how much will sell next week. That forecast is what then drives demand planning: how much to order, when, and which store to send it to. And on top, something for whoever does the shopping: a bot you send your list to, and it tells you which store it is best to buy it at, what to replace whatever is missing with, and without sending you further than you are willing to go.

Why it was a threshold

By the time the rest of the sector started watching prices in real time, the client already had two years of its own history: it knew how people had responded to every change, while the others were only starting to find out. The advantage was never the tool that collects the prices — anyone can buy that. It was the archive, and the archive isn't for sale.

Geospatial dataData infrastructureBillions of records

Billions of points that no query could reach

The organization had the data and could not use it: the geospatial volume grew faster than the capacity to query it, and any question that combined territory with time ended in an hours-long process someone ran by hand and nobody reproduced the same way twice. Decisions depending on that information were made from samples, or not made at all.

What we built

The infrastructure that was missing underneath: ingestion and partitioning of the data at a scale of billions of records, spatial and temporal indexing so that a query over an area and a period resolves in seconds rather than hours, and an access layer the business areas can use without asking the data team for anything. On top of that, the processes previously run by hand were versioned and scheduled, so the same calculation gives the same result two months later.

Why it was a threshold

The change was not one of speed, it was one of which questions were worth asking. When a query takes hours, you only ask what you already know to ask; when it takes seconds, you explore. That jump is what turns a data archive into a decision-making capability, and it is what later sustains anything you want to build with AI on top: without data queryable at that scale, no model is any use.

Public archivesMunicipalities · JudiciaryDocument digitalization

Public archives you could only read one document at a time

All the information was published and none of it was usable: spending decrees and rulings in scanned PDF, daily bulletins, systems that did not talk across areas or jurisdictions. Any aggregate question — how long a procedure really takes, where the backlog piles up, where the money is going — could only be answered with anecdotes.

What we built

The full circuit, over two different archives. On municipal spending: reading the official bulletins of Argentine municipalities, province by province, until what was bought, from which vendor, for how much and who signed it all became searchable. On a judicial archive: daily automated ingestion of bulletins and rulings, OCR of the scans, normalization of the entities appearing in them — courts, case titles, parties, subject areas — and full-text and semantic search across the whole body, with dashboards of timing, volume and composition. Every figure traced to its body, URL and capture date, and verified against the original document before publication.

Why it was a threshold

An institution that structures its archive stops arguing from perception and starts arguing from evidence: it allocates resources where the backlog actually is, and answers with the figure rather than the version. We did this from the outside, fighting somebody else's scans and whatever the institution does not publish. Inside there is none of that to fight — the originals are there, the systems are there, and so are the people who know how each field gets filled in — which is why the work from the inside is easier for us, not harder.

Firms and municipalities

Digitalizing a company and digitalizing a municipality are the same job.

The vocabulary changes — case file instead of purchase order, ordinance instead of internal policy — but the problem is identical: an archive nobody can search, systems that share no common key, and decisions taken without knowing how long anything takes. We work on both, and in a municipality there is also somebody watching from outside.

The case file is still a folder

A procedure that passes through four departments still moves by hand, and nobody can say where it is without going to look for it. The resident who asks how much longer does not get a fact: they get an estimate.

The ordinances are published and cannot be searched

Decades of ordinances, decrees and bulletins in scanned PDF. They are public and illegible to any search. Knowing which rule applies today on a given subject depends on somebody remembering.

Every area has its system and none of them talk to each other

Land registry, revenue, public works and licensing hold data on the same resident and the same plot without sharing a single key. Cross-referencing them is manual work redone every time somebody asks.

There is no way to know whether anything improved

Without measured times there is no management: no one knows which procedure runs late, where the backlog piles up, or whether the last decision helped. The argument runs on perception, and the loudest voice wins.

What you are hiring is that third case done from the inside, on your own archive and systems: ingestion and OCR of whatever is on paper or in PDF today, full-text and semantic search across the whole document base, a common key linking the registries or master records of the different areas, and dashboards of processing times by procedure or process with every figure traced to its source. And the in-house team involved from day one, because neither a company nor a municipality can depend on us coming back.

Transparency is not a by-product of this: it is the shape it takes. A searchable archive, measured times and figures traced back to their source are, at once, better management inside and better accountability outside. If you run a company or work in a municipality that has not taken this step yet, the two-week diagnosis ends with a map of what is worth digitalizing first and how much each item changes.

Let's talk about your organization

The partners

There are two of us. We both work on your project.

Limbral has no pyramid. There's no partner who sells and a junior who delivers: whoever presents the diagnosis is the one who writes the code and sits down with your plant team.

Denise Sandel, MBA

Partner · Adoption

She works on the other side of the system: the organization that has to change in order to use it. The problem is never the tool — it is the process, the incentives, and who loses control when a decision gets automated. She designs the redesign: which role changes, which meeting disappears, which metric the board starts watching.

LinkedIn profile
What she brings
  • Decision mapping: who decides today, on what information, and how fast
  • Designing the role transition for the people who will use the system
  • Data and AI governance: what is allowed, what is not, who answers for it
  • Week-by-week adoption measurement, and the call to stop when something does not take

Micaela M. Kulesz, PhD

Partner · Implementation

She builds the system the organization will use. A PhD in experimental economics and more than ten years putting pricing, forecasting and geospatial data products into production. She combines revenue-management modeling with rigorous experimental measurement: if a change can't be measured against a counterfactual, she doesn't consider it finished.

LinkedIn profile
Teaching

Senior lecturer in Industrial Organization (UBA), Quantitative Finance (SLU, Sweden) and experimental economics at AIMS Senegal and CeMarin Colombia.

Formats

Three ways to start. None of them lasts longer than it has to.

Two weeks

Threshold diagnosis

  • Where your sector sits on the curve and who has already moved
  • Three opportunities ranked by return and by friction
  • What we ruled out, with the reason in writing
  • Fixed price, no commitment to continue
Sixty days

Crossing

  • A whole system in production, not a pilot
  • Data, model, interface and the process around it
  • The team that will use it, trained
  • Usage measured from week one
Twelve months

Scale

  • Fractional data and AI leadership
  • One new project per quarter, with a quarterly stop-or-go
  • Training the in-house team through to autonomy
  • It ends when we are no longer needed
FAQ

What people ask us before the first call.

What exactly does Limbral do?

Limbral digitalizes established firms and municipalities, and leaves decision systems running in production. The concrete work is making the archive searchable, connecting systems that today share no common key, and measuring times nobody measures today. We do not sell tools or licenses.

How long until the first system is running?

Sixty days. The diagnosis takes the first two weeks, and the first system goes into production between day 15 and day 60. It is not a pilot and not a demo: it is something a team uses to do its work every day.

Do you work with municipalities as well as companies?

Yes, and it is the same job. The vocabulary changes — case file instead of purchase order, ordinance instead of internal policy — but the problem is identical: an archive nobody can search, systems that share no common key, and decisions taken without knowing how long anything takes. We work in Argentina.

What does the two-week diagnosis include?

Three opportunities ranked by return and by adoption friction, a map of where your sector sits on the curve and who has already moved, and the list of what we ruled out with the reason in writing. Fixed price, with no commitment to continue.

Do our data need to be in order before we start?

No. A good part of what an organization decides does not live in its own systems, and when the data is not inside we go out and get it: official gazettes, public registries, competitors’ prices, whatever happens around each branch. Every figure is traced to its source, its URL and its capture date.

What if we already have a data team in house?

We work with that team from day one, not alongside it. The stated goal of the contract is that they can choose and build the next project without us: a successful Limbral contract is one where the next contract is smaller.

How do you measure whether a system worked?

By usage, not by licenses: how many decisions went through the system this week, and how many hours came back to the team. If a move does not produce a result, it stops there and the project is not dragged along.

Who are the partners?

Denise Sandel (MBA), who works on adoption, and Micaela M. Kulesz (PhD in experimental economics), who works on implementation. There are two of us and we both work on every project: whoever presents the diagnosis is the one who writes the code and sits down with the plant team.

Contact

Your sector's window is either closing already or hasn't opened yet. The two answers change what's worth doing this quarter.

Write to us and in a thirty-minute call we'll tell you which of the two we think it is, based on what we see in your sector. If we think it isn't your moment yet, we'll say so on the call and there will be no proposal.

[email protected] Request the thirty-minute call