TL;DR
I automated the mechanical assembly of a recurring newsletter first, and later the work behind deciding what content deserved the production capacity. Both times the trigger was the same: automate when the manual work costs more than the tool.
The situation as found: two costs
In 2005, at Net-iris, the bi-monthly Veille RH newsletter cost an editor-in-chief five half-days a fortnight: manual selection among around a hundred publications, text adaptation, a desktop layout pass into a PDF, proofreading, distribution. It was too much preparation to be profitable.
In 2021, at CréActifs, years of content had accumulated untouched, the marketing team was two people, and the client felt at the end of a cycle; the merged site’s search traffic was preserved but flat.
- 2005: five half-days a fortnight, unprofitable.
- 2021: two people unable to decide.
Both were low-leverage manual work that grows with content volume and is paid for silently, under the promise behind systems that scale.
The diagnosis
In both cases the fix was a threshold, not a headcount question. The hand doing the task was the bottleneck.
The newsletter’s real cost was the recurring assembly-and-integration step, not the editorial selection: the editor still chose the content, but every issue forced the same mechanical pass. At CréActifs the real cost was the inability to decide what to create versus what to update, and in what order, across a corpus too large to hold in one’s head.
What was decided, and why
In 2005, I decided to build an in-house WYSIWYG back office that gathered each period’s content by theme for fast selection, to add an embedded text editor, and to replace the layout-and-PDF step with an automatic full-HTML newsletter generator. Distribution was automated too.
In 2021, I decided to crawl the whole corpus, fold in audience and conversion data, score it with a house algorithm, and hand the client a shared matrix plus a create-and-update rule sized to its real production capacity.
Both times, the rejected option was the same: keep paying the recurring labor. Automation becomes obvious when the manual cost exceeds the investment, even when the budget is not there yet.
What was built: the assembly (2005)
The mechanics were plain. A WYSIWYG backlog of themed content let the editor assemble an issue from a list instead of hunting across sources. An embedded editor kept text adjustments in the same place. An automatic full-HTML generator removed the desktop integration step entirely, and distribution ran without a hand on it.
The bi-monthly went from five half-days to half a day, and proofreading got simpler. I built it alone.
What was built: the decision layer (2021)
At CréActifs the build went one layer deeper. A Screaming Frog crawl with iterative calibration produced the corpus data, tied to page performance through the analytics and conversion accesses. I scored the whole inventory and returned it as a readable matrix in a shared Google Sheet the client later pushed to become a dynamic dashboard.
The scoring reused and adapted a house algorithm I had built earlier for a media client, the editorial-priority algorithm; its mechanism is not this piece’s subject. Reuse across clients later became Edito Pilot, a back-office-only platform that automated generating and updating such documents, the front end staying a Sheet.
What broke, and what I got wrong
The friction is real. At CréActifs the document started static and largely manual, and the client was right to demand it become dynamic. Per-article conversion data were too sparse to decide on, so I aggregated content by type and topic. The marketing team was thin and the client was overwhelmed, which made the tool an autonomy and training instrument, not just a scorer.
The honest boundary: the tool digested and ranked the data, and a human took the shot. Production capacity, the tension on objectives, the easy shot against the long one, those are human knowledge the tool cannot carry. For 2005 the record holds no breakdown beat, and I will not invent one.
The outcome, with its status
In 2005 the bi-monthly ran on minimal human resources for proven profitability, at half a day. In 2021, traffic on the merged site resumed its growth as historical pages were updated and new ones published, and the client internalized that updating mattered as much as creating. The create-and-update rule stayed qualitative: X created per month for Y updated, on a 12-month plan scaled to the client’s real production capacity.
The recurring low-leverage labor kept getting automated, the people doing it kept getting freed, and reuse pressure turned a scatter of sheets into a product. That habit, not either single build, is what the who I am page is about.
If this is the shape of your problem, let’s talk.