How to Adjust When Content Team Labor Costs Are Too High Without Affecting Tobacco Column Output

In mid-March 2024, I spread the February personnel costs of the content team across the "Tobacco Column" in a Feishu spreadsheet, and the per-article comprehensive cost had already reached **equ 380-420 yuan** (including social insurance allocation, topic meeting hours, and second-review rework). The column's commitment at the time was: one in-depth/popular science article on weekdays, one short commentary or Q&A on weekends. Readers don't care whether your labor costs are high or low; they only care whether that article on "nicotine metabolism / withdrawal timeline / oral risks" is there on Monday morning.

My judgment was firm: **Cost reduction can cut headcount, but not the "fact bottom line" and "pace bottom line."** Once tobacco-related content gets dosage wrong, regulation wording wrong, or turns individual cases into general rules, it damages the account more than a week of missed updates. Below is a round of adjustments I actually made -- splitting positions for part-timers, locking frameworks with templates, bringing AI into the pipeline -- and which steps nearly derailed.

I. Where the Money Is Really Going: Not "Expensive People," But "Human Efficiency Fooled by the Vertical Threshold"

Many people cut full-timers first. I first analyzed 128 column articles from November 2023 to February 2024, allocating work hours across four types of actions:

What really consumed the budget was the third item. At the time I had 1 full-time lead writer (pretax 14k), 1 full-time editor (pretax 11k), 1 intern (4.5k), plus occasional freelance contributors. The lead writer could produce 1 "seemingly complete" article per day, but the editor's second review averaged **0.7 rejections/article**, with the worst areas being: writing "possibly related" as "inevitably causes," directly applying foreign research to domestic scenarios, and blurring the harmful substance comparison between e-cigarettes and traditional cigarettes.

**Personal view:** The "high" labor cost of a tobacco column is often not that people write slowly, but that you are using full-time salaries to buy large quantities of standardizable intermediate processes. What full-timers should buy is judgment -- topic selection, risk expression, brand tone -- not building outlines from scratch every day.

II. Three Red Lines: Write Them Into the Schedule Before Adjusting

At an internal meeting on the afternoon of March 18, 2024, I set only three rules, written on the first page of the Feishu "Column SLA" document:

All subsequent part-timer contracts, template fields, and AI prompts were hung under these three rules. Cost reduction without red lines inevitably becomes "cheap but untrustworthy."

III. How to Split Part-Timers: Only Outsource "Verifiable Segments," Not the "Final Review Brain"

1. Cut Positions into Four Segments Instead of Finding One "Versatile Writer"

From late March to April, I split the production chain into:

In April 2024, I simultaneously tried 6 part-timers in remote collaboration (some in Shanghai, Chengdu, Guangzhou, all via Feishu + Tencent Docs), and eliminated 3 within two weeks. The reason for elimination was almost identical: **Fast delivery, but the fact card was empty, or the first paragraph of a Baidu Baike entry was rewritten and passed off as "research conclusions."**

2. Lock Trial Rules in Writing to Avoid "Favor Hiring"

3. The Pitfalls I Encountered

In the second week of April, a part-timer with excellent writing skills turned "a certain cessation pathway case" into universal advice. The editor did not discover it until 11 PM. We removed the pre-published piece overnight and swapped in a stock short commentary. **The pace was saved, but the reputation was nearly damaged.** Afterward, I wrote the word pairs "case/population" and "correlation/causation" into the review checklist as mandatory items. Part-timers can save money, but **the final review position cannot be downgraded to part-time.**

Using part-timers + piece-rate to reduce content labor is common in the industry; my experience is that piece-rate prices can be lower than the hourly equivalent of full-timers, but you must turn acceptance standards into a table, otherwise the cheap price buys rework.

IV. Templating: Turn the Tobacco Column Into a "Fillable Framework" Instead of "Inventing a New Style Every Day"

Templates do not make articles generic and empty -- they prevent 70% of structural decisions from consuming full-time brainpower.

1. The 4 Frameworks I Actually Rolled Out (Launched April 2024)

Phenomenon -> layer one mechanism (cellular/circulatory, avoid scare tactics) -> common misconceptions -> actionable daily steps -> when to seek medical/professional help

0-72 hours / 1-4 weeks / 1-3 months / 6 months+, each period only writes "observable changes + individual variation disclaimer"

Comparison dimension table (nicotine delivery, combustion products, usage scenarios) -> evidence level explanation -> do not make "absolutely safe" conclusions for readers

Scenario -> action -> frequency -> note boundaries

Each framework corresponds to a Feishu document master, with fixed fields: **target readers, prohibited wording, required source types, image suggestions, internal link column articles**. Writers open the master and start working, instead of staring at a blank page for 40 minutes.

2. "Fact Card" Is More Important Than the Outline

Before writing each article, the researcher or writer must fill out a fact card (I required it to be completable within 15 minutes):

May 2024 statistics: after implementing the fact card, second-review "factual" rejections dropped from about **0.7 rejections/article to 0.25 rejections/article**. Templates do not save literary flair; they save rework.

3. Review Checklist (Editor Goes Through in 8 Minutes)

**Personal view:** Many teams only make templates with "beginning -- middle -- ending." The focus of tobacco column templates should be on **risk wording and sources**, otherwise templates will only mass-produce acceptable empty words.

V. AI Assistance: Put in the Right Process, It Is Leverage; Put in the Wrong Process, It Is a Disaster

Around 2024, using AI for outlines, first drafts, and multilingual rewriting had already become routine for content teams; some teams calculated that streamlining the tool stack could save hours of repetitive work each week. What I care about is not "whether to use AI," but **whether it touches the final review responsibility**.

1. What I Allow AI to Do (Written into Policy Starting April 2024)

Per-article lead writer time from "blank page to ready for second review" dropped from an average of **3.5-4 hours to 1.8-2.2 hours** (regular popular science articles). Difficult articles still took close to 4 hours; I did not forcibly compress them.

2. What I Strictly Forbid AI to Do

3. Crash Record (Worth Writing Into Training)

In early May, an intern used AI to expand on the "tar and periodontal" section. The model fabricated a sample size and conclusion from a "2022 journal cohort study." The editor discovered the source did not exist through cross-referencing. Our handling was simple: **The entire section was voided, the intern's AI first-draft access was revoked for the week, only allowed to do outlines.**

Also, AI tends to write in a tone that is "encyclopedia-correct but lacking humanity." Many readers of the tobacco column are people who are quitting or have repeatedly tried to quit; they need a concrete sense of time ("what to do when your mouth is dry on day 3 after quitting"), not "you should maintain good living habits." So we stipulated: **AI drafts must go through an additional round of "personal scenario sentences" -- at least 3 instances landing on executable time/actions.**

Some in the industry also calculated that pure tool chains can compress per-article comprehensive costs to a fraction of traditional commissioning; I agree that "tools can reduce costs," but only if your manual review is focused on facts and tone. **What should be saved is typing time, not verification time.**

VI. Combined Approach: 4-8 Week Transition, Column Never Stops

I did not lay off people, add AI, and switch templates all on the same day -- that is a three-piece disaster set. The actual pace was as follows.

Weeks 1-2 (2024.3.18-3.31): Only Build Standards, Do Not Cut People

Weeks 3-4 (April): Part-Timer Trials + AI Enters Outline Level

Weeks 5-6 (Late April to Mid-May): AI Enters First-Draft Polishing, Lead Writer Unburdened

Weeks 7-8 (Late May): Accounting and Headcount Finalization

The steady state we achieved roughly:

**Personal view:** The sign of successful cost reduction is not "minimum headcount," but "per-article cost reduced and red-line incidents at zero." Reducing costs while allowing the accident rate to rise is shifting risk liabilities into the future.

VII. Three Things Particularly Important for the Tobacco Column

5 Things You Can Do Right Now

When labor costs are too high, I first ask: **Is full-timer time spent on judgments that only full-timers can make?** If the answer is no, then split part-timers, lock templates, and let AI enter the front-end processes. A tobacco column can be thinly staffed, but cannot be thin on facts. Pace is supported by stock and templates; trust depends on keeping final review in your own hands -- this is the approach I still follow after that round of adjustments in 2024.