How to Accurately Know Which Topics Bring Profit to Tobacco Content Accounts

After 11 PM on September 15, 2024, I sat in my rented studio in Binjiang, Hangzhou, spreading out the export tables from Douyin, Video Account, private domain, and store backend. The total profit for that natural month was about **42,000 RMB**, which seemed okay. But after filling in the "topic tags," I discovered: **science content contributed about 61% of views but only about 18% of profit**; conversely, the "Real Records of Days 7–30 of Quitting Smoking" column accounted for only about 14% of total site views but nearly **37%** of profit.

If you are also creating tobacco-related content — more precisely, content about quitting smoking, harm reduction, and compliant alternatives — it is easy to be fooled by "impressive view counts." Profit attribution is not a button in financial software; it is a system: **tag content → trace every revenue source to a specific topic → use models to allocate assisted conversions → sort by gross profit rather than GMV**.

Below, I break down the methods based on the pitfalls I have encountered.

1. Why "Total Account Profit" Almost Can't Guide Topic Selection

Many people only look at three things at the end of the month: follower growth, total views, and total store/livestream revenue. For tobacco-related content accounts, these three things can jointly mislead you.

First, the conversion chain is very long.

A user might first scroll past a science post about "smoking causes periodontal disease" and bookmark it; three days later find "oral changes timeline after quitting smoking" through search; a week later ask in a livestream "how to use nasal-type/NRT without relapse." The money is transacted in livestreams or private domain, but if you only record "livestream conversion," you will mistakenly think science content is worthless.

Second, unit price and gross profit margin vary hugely.

For the same transaction, knowledge payment mini-courses, physical merchandise, compliant health-related products, and consulting services can have gross margins differing by more than double. Looking only at GMV will make you treat "high volume, low margin" topics as cash cows.

Third, platform incentives and real profit are two different accounts.

In Q3 2024, I had two weeks of good traffic incentives. One post comparing "e-cigarettes vs. traditional cigarettes on oral health" broke 800,000 views, and the platform incentive totaled about a thousand RMB; but the comment section was full of arguments, conversion was poor, and DM consultations took up customer service time. **Subsidies from views do not equal the commercial value of a topic.**

So the attribution question needs to be replaced with a harsher one:

For every yuan that came in this month, after splitting it by "final conversion + intermediate education + initial seeding," which topic tags does each portion hang on?

2. My Working Definition of "Profit Attribution"

In the content account context, I define profit attribution as:

Within a clear statistical period (recommended dual-track: natural week + natural month), allocate **realized gross profit** (revenue − returns − platform deductions − advertising fees − directly attributable content costs) to **content topics/columns**, and separately record the "unattributable" portion.

Note three boundaries:

1. **The attribution target is the topic, not a single video.** A single video fluctuates too much; trends only stabilize after 8–15 posts on a topic (column).

2. **Profit takes priority over engagement.** Likes and saves are process metrics, at most auxiliary weights; they cannot replace gross profit.

3. **Compliance is a hard prerequisite.** Do not target minors, do not promote prohibited content, do not lead to illegal transactions. No matter how good your attribution is, if the account is restricted or penalized, the model goes to zero.

3. Before Starting: Break Columns Into "Topic Accounts" That Can Be Booked

For tobacco/cessation accounts, the columns I have actually used and that can differentiate profit are roughly these:

|------|-----------|----------------------|-------------------------|

When I restructured tags in August 2024, I mandated: **each piece of content must have 1 primary tag + at most 1 secondary tag before publishing**. Primary tags go into the main profit table; secondary tags are only counted for "assist frequency." Before that, without tagging, attribution was fortune-telling by memory at month-end.

The tool stack I used is very basic:

**Key action: bind "money" and "topic" together at the moment of transaction**, not guess at month-end.

4. Data Collection: The Minimum System That Can Run on a Single Spreadsheet

Starting July 2024, I forced my team (actually just me + one part-time editor) to spend 15 minutes daily filling in a "content daily report," with as few fields as possible:

1. Publish date, platform, content ID

2. Primary topic code (H01–P07)

3. Production cost (outsourced copywriting/material amortization, or estimated hours × hourly rate if none)

4. Ad spend (feed/promotion, 0 if none)

5. 24h/7d views, private domain entries (by QR code)

6. Direct transaction order ID list (with topic code in notes)

The order side has a separate "transaction detail":

I unified the gross profit formula as:

`Topic gross profit = attributed net revenue for that topic − topic ad spend − directly attributable production cost for that topic`

Common costs (rent, basic editing monthly fee) are not hard-allocated first; they are distributed once at month-end by topic net revenue share. Early allocation of common costs distorts small-sample topics.

5. Four Attribution Models: How to Choose for Tobacco Content

The classic marketing models — first touch, last click, linear, and time decay — need to be adapted into "operable versions" for content accounts.

5.1 Last Click / Last Content (Easiest to Implement)

**Rule:** The topic of the user's last effective interaction before a transaction takes 100% of the profit.

**How I define "effective interaction":** Tapping the shopping cart, entering the homepage and converting within 24h, or explicitly saying in private domain "I saw your post xxx."

**Pros:** Quick reconciliation, suitable for weekly review.

**Cons:** Severely undervalues H01 harm education.

In August 2024, when I only used the last-click model, H01's profit share was compressed to about 6%, and I almost cut the science column by half. It later proved to be a false alarm.

5.2 First Touch (Seeding Model)

**Rule:** The topic from which a user entered the private domain or first left their information takes the transaction profit.

Suitable for measuring "who is supplying you with blood." I use topic-specific QR codes: when a user scans the Q02 code to enter, transactions within the next 30 days are all credited as Q02 first-touch contribution (can be run in parallel with last-click as two separate tables — don't mix them into one muddled account).

5.3 Linear Allocation (Assist-Friendly)

**Rule:** Topics that appeared in the user journey split the profit equally.

Example: Path H01 → R05 → L06 with a 300 RMB gross profit, each gets 100.

High operational cost; requires at least being able to trace "same-user multi-touch." I can only do this in private domain: enterprise WeChat customer timeline +customer service manually tagging path labels. It's very hard to fully trace in public-domain pure short video.

5.4 Time Decay (My Current Main Model)

**Rule:** The closer to the transaction, the higher the touchpoint weight. In private domain, I use simple weights:

This is not academically optimal, but **for four consecutive weeks in October 2024**, the "topics to increase investment" ranked by time decay correlated positively with actual profit increment after investment, much more stable than pure last-click.

Personal view:

For public-domain account weekly decisions, "last-click as main table + first-touch as blood-supply table" is sufficient; once private-domain monthly transactions exceed 40% of total profit, switch to time decay. Don't jump straight into multi-touchpoint algorithms — you'll burn out on data entry.

6. Profit Attribution by Column: The Role Division I Actually Observed

The numbers below come from my **September 2024** monthly review of my main account (cessation/oral health focus), calculated with time decay + private domain path adjustment as **gross profit share**, not view share. Platforms are primarily Douyin + Video Account; private domain uses enterprise WeChat. Figures are rounded.

6.1 Harm Education (H01): View King, Profit Assassin (If Not Closed Properly)

Issues encountered during operation:

In early September, I ran a continuous series on "smoking and oral problems," with single posts reaching 800,000+ views, but the comments section was full of fear-based interactions. Many leads came in, but with mixed intent — a high proportion asking "do I have cancer" — and conversions were slow. Customer service spent a lot of time explaining, **labor costs were underestimated**.

My fix:

At the end of each science post, attach a fixed "next step": not a hard ad, but directing to Q02 timeline or R05 myth debunking. In attribution, H01 is mostly recorded as "first-touch blood supply"; last-click profits are still low, but I separately track **H01 → private domain 7-day retention**. Science content with poor retention gets downgraded regardless of view count.

6.2 Cessation Methods & Timeline (Q02): Slow Cold Start, High Profit Density

The "Days 3–7 Body Changes After Quitting" post published on September 8 had only 70% of the site average views, but the Q02 QR code brought in 96 leads that week, with ~**9,200** gross profit from conversions within 14 days.

**The reason is practical:** These users are already taking action; they don't need more scare tactics — they need pacing and methods.

6.3 Product Comparison/Review (C03): Tight Compliance Line, High Profit Volatility

Pitfall: In August, one comparison piece was too absolute in its wording, triggering review and traffic restriction risks, costing two weeks of adjustments.

**Attribution lesson:** C03 must separately record "compliance incident costs" (suspension of updates, scrapped edits). After I started recording scrapped costs to C03 for that month, the real ROI was immediately exposed — some seemingly profitable reviews were actually draining account security.

6.4 Emotional Stories (S04): Add-Friend Champion, Weak Direct Sales

I changed S04's success metrics to **effective add-friend cost** and **30-day private domain payment rate**, and stopped forcing story videos to include shopping links. Forcing links ruins the atmosphere, drops completion rates and trust, and is not worth it.

6.5 Q&A & Myth Debunking (R05): Search Long-Tail Profit Patch

Individual posts have ordinary views, but search-driven users have specific intent.

In September, R05 gross profit was ~9% with low production cost (many from high-frequency comment questions), **per-unit-hour profit** often ranking first.

Recommendation: Schedule 2 fixed posts per week, specifically answering pinned comment questions.

6.6 Livestream/Video (L06): Profit Booking Window, Not Profit Source

Livestreams are easily recorded as "all profit belongs to the livestream."

I analyzed a 2.5-hour session on September 12: on-site gross profit was ~11,000 RMB, but tracing lead sources revealed **~64% of converting users had viewed Q02 or R05 in the previous 14 days**.

If you attribute 100% to L06, you will livestream obsessively while starving the blood-supply columns.

**Approach:** Reallocate 50%–70% of livestream gross profit back to the content-side lead source topics (I use 60% back to content, 40% retained for livestream operations).

6.7 Private Domain Retention (P07): The "Last Mile" of Profit, Also Where to Test Models

Check-in group weekly reports, repurchase reminders, cessation calendars — this type of content has almost no public-domain views, but **September repurchase-related gross profit of ~12%** was mainly here.

Without P07, all the education from earlier topics leaks away.

7. My "90-Minute Weekly Review" Process

Every Monday morning (I fix Monday 10:00–11:30):

0–20 min: Data Cleaning

Export last week's orders, refunds, and ad spend; cross-check orders with missing topic codes. If the missing rate exceeds 15%, stop analyzing and fix the tracking first (notes/QR codes).

20–45 min: Generate Three Tables

1. Last-click topic gross profit ranking

2. First-touch blood supply ranking

3. Time decay comprehensive ranking

Only use the comprehensive ranking for investment decisions, but if it severely diverges from last-click, check if the livestream reallocation ratio is set incorrectly.

45–70 min: Column Diagnosis (Four Questions)

70–90 min: Assign Actions

Only decide three things: this week's primary topic, the topic to reduce frequency, and one mandatory closure mechanism (e.g., H01 must guide to Q02).

More than three actions means no action.

8. Common Mistakes (I've Made All of Them)

Mistake 1: Using view count as profit weight.

Result: the whole team makes scary content, the account becomes a "tobacco horror channel," and the private domain is a mess.

Mistake 2: Only attributing to platform, not to topic.

Conclusions like "Douyin makes money, Video Account doesn't" are too coarse. Q02 might be very profitable on Video Account while H01 might be a loss; lumping them together misjudges the platform.

Mistake 3: Messing up the statistical window.

Content explodes in 3 days, but conversion takes 14 days. I unified: **exposure window = 7 days, profit attribution window = 14–30 days**. Too short a window means science content never gets credit.

Mistake 4: Counting ad spend against content profit but not counting organic traffic from content.

Paid topics get double-penalized, organic topics get mythologized. Spend must be attributed to topics, and organic volume must also be attributed to topics.

Mistake 5: Topic tagging by feel.

Sloppily adding "oral + cessation + product" tags makes aggregation impossible at month-end. There can only be one primary tag.

9. Copy-Paste Table Headers (Markdown Format)

Content Ledger

|------|----------|------------|--------------|----------------|----------|-------------|-------|-------|

Transaction Ledger

|-----------------|-------------------|--------|----------|------------|-------------|--------------|----------------------------|

Column Monthly Report

|-------|-------|-------------|------------------------|-------------------|-----------|---------------------|----------|

The decision column only allows: Increase Investment / Maintain / Reduce Frequency / Pause and Restructure.

10. My Clear Judgments (No Double-Sided Flattery)

1. **Profit in tobacco content typically doesn't come from "the scariest post," but from "the pacing and methods that people who already want to change need."** Q02-type topics have been consistently more valuable on my books.

2. **Science content must exist, but should be assessed by blood-supply metrics, not cut by direct ROI.** Without H01, Q02's cold start would be more expensive.

3. **Livestreams are the cash register, not the water source.** The water source is in timelines, myth debunking, and private domain check-ins.

4. **The first goal of an attribution system is "to give you the confidence to cut videos," not to build a perfect model.** If it can consistently point out 1–2 columns that should be reduced in frequency, it's worth the time spent on data entry.

5. **Compliance costs must go into topic accounts.** Otherwise, C03 will always look attractive — until the account has problems.

11. The Minimum Experiment You Can Do This Week

If you have no system in place, don't jump to complex models. Use 14 days for this experiment:

1. Add primary topic tags (H01–P07 above) to your last 30 pieces of content.

2. Switch private domain add-friend to topic-specific QR codes, at least dividing H01 / Q02 / Other.

3. Force a topic code note on every transaction (lock thecustomer service script).

4. On day 15, generate a gross profit ranking with the last-click model and a blood supply ranking with the first-touch model.

5. Compare: View Top 3 topics vs. Gross Profit Top 3 topics — are they misaligned?

When I first ran this minimum experiment in July 2024, the misalignment was very obvious: View Top 1 was scare-based harm content, Gross Profit Top 1 was weeks 2–4 of cessation records. Since then, at every topic selection meeting, I first ask:

"Is this piece going for views, or for profit? If it's going for views, which profit topic does the closure connect to?"

If you can answer that clearly every time, your tobacco content account has only just begun to "know where the money comes from."

Accounts that don't know where the money comes from are just swapping health anxiety for traffic — lively in the short term, but the books tend to look ugly.