TL;DR
Melty, a media site whose economics rest on daily audience, lost a number of the decision-support tools it used to order what to publish in a deep migration. I chose to rebuild that capability from the data, designing an aggregate scoring algorithm over SEO and audience key performance indicators (KPIs) that ran in a tool Melty used for months. The tool held the audience at full potential, and it outlived the engagement it was built for: that is the promise behind systems that scale. It became my own platform, reused across 10+ client sites.
The situation as found
A media website lives on its daily audience, because that drives the ad revenue. Melty’s edge was deciding early and correctly where to spend editorial effort, across thousands of pages, using decision-support systems that ordered the publications day by day. After a deep migration that rewired the front end and back office, a number of those decision-support systems were lost and could not be migrated.
The CEO wanted a decision-support tool to pilot the update of roughly 6,000 editorialized tag pages, the dossiers that gather every news item on a single subject. Three constraints made it hard:
- There was no surviving decision machinery to bring back.
- The audience could not be allowed to dip while the work ran.
- A business model with no tolerance for a wrong priority order.
The diagnosis
The presenting symptom was that the editorial team simply decided what to update. The real cause ran deeper: the machinery that made that judgment consistent and fast was gone, and deciding by feel does not scale. Effort went to the wrong pages at the wrong time. This was a data-and-ordering problem, not an opinion problem.
What was decided, and why
I proposed to work from the data, announcing up front that it would be an R&D process built on SEO and audience KPIs. This ruled out the apparently safer path of reconstructing the lost systems from memory, or handing the ordering back to people, which the constraints made unaffordable: no surviving system, no staffing surge. The approach is the same data-first, R&D design I used for a fraud-detection algorithm.
Two choices made the data path cheap and fast. I designed the aggregate scoring algorithm over SEO and strategic-audience KPIs. And I worked with the CEO as a peer: an engineer by training, he chose to build the live data-retrieval through APIs himself rather than mobilize Melty’s technical teams. I worked alone, with no team of my own to hand the build to.
What was actually built or changed
- Data. A full statistical extract on dossier performance from five years of aggregated internal data, crossed with 12 months of analytics and SEO data pulled with Screaming Frog SEO and API connectors, including Ahrefs.
- Proof of concept. A matrix over the top 100 current dossiers, crossed with the top keywords of the moment and the top themes and verticals by audience and ad revenue.
- The algorithm. An aggregate scoring algorithm weighing several SEO KPIs together with strategic-audience KPIs. The scoring formula is out of scope here; its anatomy is a technical deep-dive’s subject.
- The decision matrix. Built iteratively through daily alignment meetings and presented to the editorial manager and section heads.
- The beta tool. Static data exports in a Google Sheet, plus live data refreshed through the CEO’s API calls.
What broke
The tool shipped as a beta on a deliberately provisional architecture, static exports plus live additions, not a hardened product. The first weeks of R&D ran quietly, under the radar, before coordination began, a real risk of building in the wrong direction. And the migration had wiped a number of the previous decision-support systems, so the exercise started from zero trust in anything prior.
The outcome
- Melty’s teams ran the beta for several months to pilot strategic-SEO maintenance of their dossiers, keeping the audience at full potential alongside the daily news flow. There is no number to quote; that is what the source states.
- The secured audience was part of what let the CEO, mandated by his investor to find a buyer, close the sale: Melty was sold to a major media group specialized in people-news and entertainment.
- Because the algorithm was my own, financed by a client’s R&D, it became my own editorial-prioritization platform, its scoring algorithms improved and made available to 10+ client sites. The one documented reuse turnaround is the CréActifs, which reuses this algorithm. The system outlived the engagement it was built for and kept operating after I moved on.
If this is the shape of your problem, a repeatable judgment call that today lives in people’s heads, let’s talk.