Santiago Serrano
← All projects
/ ProjectIn development

Júpiter Studio

A data structure that serves different brands

17
tables in the data structure
9
analysis rules
28
days recalculated every day

/What it is

A platform for several brands that brings together advertising results, social conversation and strategy, and each week produces a few conclusions ready to act on. Part of a twelve-month programme for a corporate client.

/The real problem

The client can't verify its vendors' monthly reports nor measure, with a single figure, how many people it reaches and how often. Each brand measures similar things under different names, and ad platforms correct their numbers days later.

/Decisions I made

  1. D-01

    The system doesn't know the brands

    Each brand is described in a configuration file and the system has no special logic for any of them: adding a new brand means adding a file. No limit or threshold is written inside the code.

  2. D-02

    Reach is never added up

    Platforms don't let you add reach without counting the same person twice, so it is stored separately for each period. And a campaign name is never used as an identifier, because it changes.

  3. D-03

    Every day, 28 days are downloaded again

    Meta corrects its figures backwards, so every day the last 28 days are downloaded again instead of trusting what is already stored.

  4. D-04

    AI writes, it doesn't calculate

    Conclusions come from nine fixed rules. The AI only writes them in plain language from numbers that have already been checked, and what can't be verified isn't published.

/How it works inside

  1. 01 Sources

    • Meta
    • TikTok
    • Agency file
  2. 02 Download

    • Daily download of the last 28 days
  3. 03 Structure

    • Common data structure (17 tables)
    • Configuration per brand
  4. 04 Results

    • Analysis rules
    • Panel per brand
Target architecture, simplified.

The internal screens are confidential: a diagram of how it works is shown.

/Result

A 17-table data structure where the brand is part of all the data, nine analysis rules, four human review points and a navigable prototype with six screens.

/ Status

In development. The specification and the screen prototype are done; data download and the internal side are still being built. No brand or agency names, for confidentiality.

Team project; I am responsible for the technical architecture and the data structure.

Got a similar problem?

The first conversation is free. Tell me how your business runs and we'll see what can be built.

/ Next project

MyTicket MX

The theatre ticketing company that left Shopify to sell on its own system