TL;DR
The editorial-priority system I built for an ad-funded media site kept working after I pointed it at a lead-driven training business. The gaps it exposed forced it to adapt, and the maintenance load turned it into a product.
The situation as found
Melty is an ad-funded media site whose economics depend on daily audience, at tens of millions of visitors a month. CréActifs is a small training business that needs leads, at around 150k a month. The same editorial-priority system had to serve both.
- Scale: tens of millions against around 150k.
- Objective: ads and audience against leads.
- Counterpart: an engineer-CEO against a young marketing lead.
A system proves itself reusable only when pointed somewhere it was not designed for, the promise behind systems that scale. The tool ranked; a human took the final call.
The diagnosis
The objection was that the tool belonged to a big media site, but the obstacle was mechanical. The key performance indicators (KPIs) sat on incomparable scales: thousands, units, percentages, so nothing could be aggregated or compared honestly. At the smaller shop, per-article conversion was too sparse to decide on, forcing aggregation by type and topic. A working tool still demanded manual upkeep. The same logic could be re-aimed once the aggregation and the objective were re-pointed.
What was decided, and why
The logarithmic-scaling pivot at Melty came first. My early attempts forced the magnitudes onto one scale with multipliers weighted by experience, and they never worked. A conference remark by an R&D director at a search-tooling company gave the unlock: logarithmic values become comparable. I tested it, and the KPIs came together into one composite score. The rejected option was the multiplier route.
The same logic was then re-aimed onto conversion at CréActifs. The client’s question was not which update wins the most clicks; it was what moves leads. I balanced a search-and-audience score against a conversion score, content aggregated by type and topic. The rejected option was treating each business as a bespoke build, which defeats the reuse.
Who I built it with, and what that changed
At Melty I paired with the engineer-CEO on an asynchronous day-and-night cadence: he built his own API pulls so the dev teams were not mobilized, and his emulation pushed the scoring further.
At CréActifs the counterpart was a young marketing lead who had recruited me, knew her product and objectives, and validated milestones on the spot. She was at the end of a cycle and needed an autonomy and training instrument more than a scorer.
Where the tool’s intelligence lived moved with the counterpart: the algorithm and live pipeline at Melty, the shared Sheet and the client’s production planning at CréActifs. I built the scoring logic myself.
What was actually built or changed
- At Melty: an aggregate scoring algorithm over search and strategic-audience KPIs, fed by static year-and-three-month extracts plus dynamic 30- and 7-day data in a Google Sheet, with the CEO’s live API pulls layered in.
- At CréActifs: the same logic re-pointed onto a two-sided search-plus-conversion composite, in a shared Google Sheet the client’s production planning used, static and manually refreshed at first.
Enough detail to recognize real, transferable work: the deep anatomy of the score is a separate subject.
What broke, and what I got wrong
- The multiplier route failed for a long time before the log pivot.
- The Melty Sheet grew to dozens and dozens of columns, so I built views to hide the noise.
- At CréActifs, per-article conversion was too sparse to decide on, a limit I found by using it, which forced the aggregation.
- The manual load (updating data, fixing, debugging, adding features for new clients) stopped scaling.
The tool held at a handful of websites and no further. At three, then five, then 10 sites, keeping each one up by hand outran what I could give it.
The outcome, with its status
The system was re-aimed and it held: the same editorial-priority logic, pointed at a market of unlike shape, kept working. The maintenance pressure pushed it into Edito Pilot, a back-office-only build with a freelance developer while the front end stayed a Sheet. My rule: automate when doing it manually costs more than the investment.
The scale contrast is a stated order of magnitude, not a measured count; the 3-to-10-site figures are class-level.
The human kept the shot: the tool was a decision helper, and the engineer-CEO or the marketing lead took the final call. That boundary keeps it honest, and it is the same builder behind the who I am page.
If this is the shape of your problem, let’s talk.