Influencer performance scattered across 5 platforms, condensed into one advertiser report | VROF

Background

The request came in with something already built once. The previous vendor had delivered a Windows-only desktop program, and what the client sent over, organized, was not a review but a defect list. Only one of the five platforms was marked “no issues”.

What we saw in the field

Reading the defect list as per-platform bugs gives more than ten things to fix. But laid out side by side, the blocked points were almost all the same place.

One was “post access blocked by Cloudflare verification”
One had the post body, comments, and thumbnails not captured at all
One had comment authors marked @user instead of their actual names

The previous program was built in Python and launched a separate browser, then accessed the platforms without logging in. When the platforms judged that browser not to be a person, whatever the code after that did, it received empty values. It was not a bug — it was a problem of the access approach.

The problem we defined

Instead of fixing the collector, we decided to move where it runs. The client was already logged in to each platform with their own account. If it was a Chrome extension running inside that browser, there was nothing new to authenticate — it reads the screen the person is already looking at.

Since it was a decision to drop the Python desktop app and build from scratch, we explained first why we were not patching onto it before starting.

Collection screen. On the left is the list of supported platforms and each platform’s login status, and in the middle is an input field for registering a collection target address, with a window open for checking platform login status. The window says “If you are logged in to the browser, it is recognized automatically.” The two example address lines written in the input field are blurred out.

The extension does not create new authentication. It reads the logged-in state the person in charge has already set up, and only marks the platforms whose login has expired, guiding them to log in directly.

What we built

Collector for 5 platforms — 3 used domestically and 2 in Chinese-speaking regions. Each had a different authentication method and data shape, so each was built separately. Instead of the screen structure, it references the internal data the page already holds. One of them had no official channel and its screen layout changed frequently. It finds values even when the layout changes — because it reads the values that screen already holds, not its appearance
Database inside the browser — the collected material stays in the person in charge’s browser. There is no server, so advertiser data never leaves
Multiple posts by the same author — the previous program let later-collected data overwrite earlier data. Changed to key on each post’s unique identifier, so it accumulates post by post
Report for advertiser submission — per-platform aggregates, top-performing content, and trends over time in one sheet. The client’s first item was “it’s meant for advertiser reporting but feels very shabby”
Excel export — exports to Excel files instead of the CSV that kept breaking
Collection time recorded alongside — metrics of the same post are kept per point in time. Values that cannot be known from a single check, like the reaction speed right after posting, can be reviewed later

Result

The one platform that worked in the defect list became five. We left a verification record checking comment collection on all five platforms item by item (2026-02-08).

Other work done the same way is collected under web data collection.

Other work on the same topic is collected under data analysis.