Anti-patterns
反模式（禁止行為）
1
NOT: Pull topic ideas from generic "construction trends" headlines without checking whether they map to architectural hardware buyers — INSTEAD: Mine topics from vertical-specific signals like door hardware code changes, access control integration questions, spec-stage pain points by audience segment
2
NOT: Let Topic Miner optimize for high-volume keywords that attract broad consumer or residential traffic — INSTEAD: Prioritize low-volume, high-intent B2B topics tied to commercial openings, specification workflows, and project-stage relevance
3
NOT: Write every email as a product blast centered on features and catalog language — INSTEAD: Frame emails around buyer jobs-to-be-done such as solving code compliance risk, reducing install errors, or shortening spec review
4
NOT: Use the same email copy for architects, distributors, installers, and end users — INSTEAD: Tailor messaging by role — architects get spec and code guidance, distributors get demand and margin signals, installers get application guidance
5
NOT: Write subject lines using vague hype like "innovative solutions for modern spaces" — INSTEAD: Use concrete B2B value signals in subject lines tied to code update relevance, application fit, or retrofit guidance for the recipient's segment
6
NOT: Segment only by firmographic fields like company size or industry code — INSTEAD: Combine firmographics with behavioral and lifecycle signals such as product page depth, CSI section interest, quote requests, and inactivity windows
7
NOT: Collapse all product interest into one "hardware" bucket — INSTEAD: Track and segment behavior by solution family (closers, exit devices, electrified hardware, access control) so nurture paths match actual interest
8
NOT: Treat all opens and clicks as equal engagement — INSTEAD: Score behavior by intent, distinguishing weak signals (single open) from stronger signals (repeated visits to spec-sheet pages, downloads, distributor page clicks)
9
NOT: Trigger follow-ups immediately after any click without checking context — INSTEAD: Use behavior tracking rules that account for recency, frequency, asset type, and page sequence
10
NOT: Ignore the difference between research behavior and purchase readiness — INSTEAD: Separate early-stage education journeys from late-stage intent journeys using content type, revisit cadence, and conversion depth
11
NOT: Declare a topic or segment "winning" from one campaign send — INSTEAD: Compare performance across cohorts, send times, language versions, and segments, then validate with repeated tests
12
NOT: Translate English emails word-for-word into Chinese and assume the same CTA and tone will work — INSTEAD: Localize bilingual emails with architecture/hardware terminology, region-specific compliance references, and CTAs matching how Chinese-reading B2B buyers evaluate products
13
NOT: Mix Chinese and English inconsistently inside the same segment without knowing language preference — INSTEAD: Segment by explicit language preference, engagement language, and geography so bilingual delivery is intentional
14
NOT: Route every task to the same model regardless of complexity or risk — INSTEAD: Use lighter/faster models for draft ideation and summaries, route high-stakes work (bilingual copy, compliance claims, performance analysis) to stronger models with review steps
15
NOT: Measure email success only by open rate — INSTEAD: Measure by downstream B2B outcomes: qualified clicks, gated asset completions, quote/demo requests, unsubscribe rate by segment, and influenced pipeline