Cases

The data usually disagrees with the first story.

We publish how each case was built, what the evidence showed, and where we changed our own mind. Client names stay with the client. Every figure was measured against a documented baseline, or is labeled as an estimate.

Case / 01 Pharmaceutical group

DiagnosticRGM Supply & Planning

Data everywhere, decisions nowhere.

R$22–42Mgross profit a year at national scale, 3–6% of adjusted EBITDA

R$4.4–8.55Mone-time cash from stock parked at the distributor

97%of revenue covered by the rebuilt sell-in to sell-out mapping

23xhow much we cut our own stock-out estimate when the data said so

The question: Prices went up and nobody could say what happened next. Sell-in and sell-out had no shared key, so the two halves of the business could not be read together.
What we did: Rebuilt the link between internal product codes and retail barcodes, then measured stock in the channel, field execution against territory potential, and the official forecast against a naive one.
What the data showed: Much of what looked like weak demand was stock parked at the distributor, 79% of it in SKUs the channel was draining. Rep performance turned out to be 62% territory and 38% execution. The official forecast erred 60.3% (WMAPE) where a naive average erred 31.7%.

Case / 02 Fuel retail network

PilotOrchestration Measurement

Moving the premium fuel mix, one attendant at a time.

1station in the first wave

11–14attendants, each with their own daily message

30days, with treatment and control fixed up front

Pendinglift, reported only after the readback

The question: Can a daily message to each station attendant, shaped by how that attendant sells, move the mix toward premium fuel?
What we did: Signal decides who receives which message and when. A separate execution layer delivers it. Attendants who receive messages are compared with a control group.
Where it stands: The hypothesis, the metric and the control group were fixed before the first message. The lift is reported only after the readback.

Case / 03 Premium sportswear retailer

Pilot deliveredRGM Measurement

Media decided on 10% of the sales.

232%of sales credited to a single channel by the old attribution model

90%of sales happen in stores, while media was decided on the 10% online

60%of the media budget sat below an ROI of 1

+15%digital sales from reallocating the same budget, model estimate

The question: Which media spend actually drives sales, before deciding what to take in-house from the agency?
What we did: Replaced return on ad spend with incremental return on investment, over 797k rows of orders, stores and media by campaign and state, in one week.
What the data showed: Branded search fell from first to seventh place once incrementality was counted. The +15% is the model’s estimate for moving the same budget, and has not been read back against a control yet.

Case / 04 Multinational beverage group

Proof of conceptP&L Twin

Value written into contracts nobody enforces.

10contracts in the proof of concept

34.9%of the sample addressable

4xthe 9% WorldCC benchmark for contract value leakage

172minors billed as adults

The question: How much value sits in supply contract clauses that are never checked against what is actually billed?
What we did: Read the contracts in PDF, extracted every clause with an economic effect, and checked each one against what was billed and what was exercised.
What the data showed: One contract had expired four months earlier and was still running. Scopes overlapped, profit-share clauses had never been exercised, and minors were billed at adult rates.

Before Pragma

Results our team delivered in prior roles.

Programs with a number

01
DanoneR$72M
02
PepsiCoR$35–85M
03
PhilipsLogistics cost down 15%
04
Axia Agro600%+ return, R$70M+

Also delivered for

NidecKraft HeinzMondelezAmbevJ&JKellogg’sBRFNetflixDHL

Delivered by members of our team before Pragma existed, in prior roles. These companies are not Pragma clients, and the work was not done in Signal.

Your case
next

Every case on this page started the same way: one data extract, and 14 days later a Value Tree with an owner and a date on each number.

Vinícius GaleraCo-founder

Start a Value Tree

hello@pragma.do