The client
A publisher running several content properties across SE Asia. Their economics rest on a constant drip of long-form articles and practical guides for travelers, expats, and people researching moving to the region. The content team was six people, all capable, all quietly burning out.
The problem
The business needed roughly forty pieces published across its properties each month, and the team could comfortably do about twenty. Everything else slipped, got rushed, or never happened. The bottleneck was not any one step. It was every step: finding the right angle for a topic, doing the research, drafting, fact-checking, adding references, tuning for search, formatting for the CMS, and running editor review.
Each of those steps took a writer or an editor meaningful time, and each had to happen for every piece. The editor-in-chief described the problem in a way that stuck: “None of it is hard. It is just that there is always more of it than people.”
The team had tried writing with the consumer AI tools already on the market. The output was serviceable for a first pass but needed so much editing that the time saved was swallowed by the review. It also kept producing the same kind of bland, inoffensive prose that their readers would not come back for.
The approach
The pipeline broke the job into the steps the team was already doing and put an AI layer on the right ones. A new piece started as a brief submitted through a simple internal form. The pipeline then ran the brief through a research step (combining an LLM with a search API for sources), produced a structured outline, drafted the piece in the house voice using examples from their own archive, ran a fact-check pass against the cited sources, and applied their SEO checklist.
The editor saw the output inside a purpose-built dashboard, where every fact was linked to its source, every stylistic call the AI had made was flagged for confirmation, and the editor could accept, reject, or rewrite at the paragraph level. The editor remained the gatekeeper. Publishing always required a human pressing the button.
What was automated was the mechanical work. What was preserved was every part of the process that required editorial judgment.
The result
Production time per published piece dropped by around seventy percent. The team went from twenty pieces a month at capacity to forty-five pieces a month comfortably, with editorial quality holding steady in the internal review scores. Two of the six team members shifted into roles that use their judgment rather than their typing. Those roles turned out to be commissioning, editorial strategy, and partnerships. The other four team members work shorter days.
The editor-in-chief now spends most of her week on strategic editing and voice training of the drafting layer, rather than on production. She has called it, more than once, the first year in a decade that feels sustainable.
What this is not
It is not AI replacing writers. The team is still the same size, doing better work, on a sane schedule. It is also not a magic content machine. Every piece still goes through a human editor, and that is the step that matters most. What the build did was remove the mechanical bottlenecks that were quietly eating the team alive.
"For the first time in a decade, my Friday afternoon does not feel like a panic. The team is publishing more, the editorial quality is holding, and two of my writers got their lives back."