Skip to content
Lavínia Alencar Belém, Pará, Brazil · working remotely Message me

The report is correct.
The number isn’t.

Your system doesn’t lie. It answers the question someone programmed into it, which is almost never yours. The gap shows up when someone sits down and reads the data line by line.

I want an X-Ray See the work

Three years in data inside iFood and Nuvemshop, after two in tech as an IT technician.

Coffee shop one month
Card machine fee
2.08%2.68%
Average ticket per visit
R$ 39.2053.40

Neither one was bad faith. The billing rule had never been audited, and the count added up transactions instead of visits.

Three stages, and you can stop at any of them

The first is short on purpose. It tells you whether something is hiding in your data before you sign anything bigger.

01 · short project

Data X-Ray

I read your data from start to finish and tell you what it really says: real fee rate, real average ticket, the hours that carry the register, the customers who carry the revenue. You get a dashboard, a report and a conversation explaining what to do with it.

from $750 Start here
02 · fixed project

Custom Automation

I build the machine that does the repetitive work in someone’s place. Sometimes it’s a service that takes orders and answers on its own. Sometimes it’s a routine that runs overnight and leaves everything ready before you arrive.

scope per phase
03 · monthly

Data Partnership

Short, recurring cycles, looking at what changed and what it means for the next decision. It’s for those who already have the machine running and want someone reading alongside them, not for those who want yet another dashboard.

from $250/month

Seven phases, always in the same order

In any stage and for any client. They exist so you can hold me to account midway, and so someone else can pick the work up if I’m not around.

1

Context

What the data source covers, and above all what it doesn’t cover. Starting here avoids the pretty conclusion built on top of a hole.

2

Audit

This is where the number changes

Business rule validated line by line, never assumed. This is the phase where the 2.08% fee turned into 2.68%.

3

Modeling

Loose transactions become the concept you live by at the counter: visit, customer, shift, month. One single metric layer, so everyone adds up the same way.

4

Delivery

Dashboard and report reading from the same source, recalculating nothing on the side. If two places show the same number, it’s because it is the same number.

5

Translation

This is where it becomes a decision

The number becomes action, separating what the data supports from what it doesn’t support yet. Saying "I can’t claim that yet" is part of the delivery.

6

Automation design

The right architecture for the size of your business, not the prettiest one to show off. The size of the solution follows the size of the problem, never the other way around.

7

Continuity

Everything documented, so the work can be picked up by anyone. Including you, if one day you want to bring it in-house.

Three projects, and what changed in the decision

+36%on the real average ticket, counting visits instead of transactions
47.9%of revenue concentrated in three hours of the day
40%of revenue coming from 13% of customers
Coffee shop · Data X-Ray

The fee wasn’t what it seemed,
and neither was the ticket

Starting point

One month of sales, all through the card machine, and no reading of their own data. The owner decided staff schedules and supply purchases from a memory of how busy it felt.

What I did

I audited the billing rule line by line, redid the count by grouping transactions from the same visit, built a dashboard and a report, and designed the architecture that would run it on its own every day.

What changed

He stopped scheduling staff by gut feeling and started scheduling around the three-hour block that pays the month. And he now knows what every real that comes in costs.

Absolute values rescaled at the client’s request. Proportions, percentages and conclusions are the project’s real ones.

6restaurants on a single channel, each with its own menu
3failure cases solved on paper, before any code existed
56hof scope quoted phase by phase, no guesswork numbers
Restaurant gallery · Automation design

Six restaurants serving customers
from a single WhatsApp number

Starting point

Orders jotted down by hand, customers waiting without knowing whether theirs was seen, and each restaurant with a different way of serving.

What I did

I designed the journey for both sides, customer and restaurant, step by step. I chose each component with a written justification, and handled the annoying cases from the design stage: abandoned cart, cancellation midway, and two orders being written in the same second.

The decision that matters

Use the official WhatsApp API, even though it costs one more registration and a few days of approval. The unofficial alternative works instantly and carries a real risk of the number being banned without notice, which would halt the whole gallery.

Project designed and quoted, not built. The numbers above are about scope and decisions, not results.

7.6xthe value of an ordinary purchase, in each fraud
1h13between the fraudster’s transactions, against 4h24 for the legitimate customer
50%of frauds concentrated overnight
0.5%of transactions, which makes fraud vanish in any average
Credit card issuer · Analysis and modeling

The fraud’s modus operandi,
turned into three executable rules

Starting point

Two years of transactions, fraud in 0.5% of them. An event rare enough to vanish in any average, and costly enough to hurt.

What I did

Dimensional modeling, a full pipeline for ingestion, cleaning and transformation in PySpark on Databricks, and descriptive and diagnostic analysis on top of the modeled layer.

What changed

Each factor became a written alert rule, ready for production, not a vague recommendation. A value above the limit goes to review, a short interval adds points to the score, and overnight goes to review.

California
22%
Missouri
14%
Nebraska
12%
Oregon
11%
Washington
7%

Four states concentrate almost 60% of the cases. Concentration is what makes an alert rule viable: you can tighten where it hurts without bothering people who have nothing to do with it.

Does your data hide a number too?

I want an X-Ray

What I build with

I pick the tool your team can keep running on its own, not the one that looks good in a portfolio.

Engineering phases 3 and 6

Taking data from where it’s born to where it’s read, every day, without anyone pushing it by hand.

SQLPythonPySparkSparkAirflowDatabricksAWSETL and ELTAPI ingestion

Modeling phase 3

Turning loose transactions into the concept the business owner lives by, with a single definition per metric.

dimensional modelingfact and dimensiongranularitysingle metric layerdata dictionaryquality monitoring

Automation phase 6

Connecting systems that don’t talk to each other, and making processes run without people in the middle. It applies to data that needs to move, to service that needs to answer, and to routines that need to happen on schedule.

n8nAirflowApps Scriptsystem integrationAPIswebhooks and scheduled triggersself-hosted server

Reading and dashboards phase 4

Where the number shows up for whoever decides. Including in the spreadsheet, which is where most people actually work.

Looker StudioTableauGoogle SheetsGoogle SlidesGoogle DocsExcelrecurring reportsresults presentations

Organization and management phases 1 and 7

The part nobody sells and everybody needs: making clear who owns which number, where it lives and how it is recalculated.

metric definitionprocess mappingredesign before automatingprocess documentationreview routinesource and access governancedecision versioninghandover to the internal team

Is one of these your question?

  • How much do I really pay in fees on each sale?
  • Do my customers come back, or is it always new people?
  • What time of day pays my month?
  • Is it worth opening earlier, or would I be paying staff for nothing?

If one of these is yours, message me. We’ll look at it together, and I’m the one who answers. I work in writing, by WhatsApp or e-mail.

Mobile · WhatsApp+55 91 98546-7984
E-mailcontato@lavinialencar.com.br
LinkedInin/lavinialencar