AI Agent Marketing & Sales Case Hub: Verified-Source Records on Lead Gen, Ad Operations, Support Automation, and Full-Funnel Sales

Data basis and how to read this page. This domain holds 269 published cases — counted this round from the live published data (filtered by the "营销与销售" category on 2026-11-09 14:15; matches the category count in the site's llms.txt, updated 2026-11-09). This page draws only on 10 of them that we read line by line and re-checked for source accessibility (sources 10/10 reachable, all HTTP 200, verified 2026-11-09 14:20–14:45). Every figure here is source-reported; this site has independently re-tested none of them. Re-check the original source before citing any number.

Who this page is for, and what it promises. For founders, growth leads, and sales managers evaluating whether to hand marketing and sales work to AI agents. After this page you can state: which marketing/sales task types have publicly traceable records, where humans stand in each (the five-bucket model: setup / preparation / oversight / review / exception handling), who vouches for each disclosed number, and under what conditions these attempts fail.

Task map

Seven task types with verified examples (each with its boundary):

1. Lead research & SDR outreach automation — define the ICP → scrape and research prospects → draft personalized first emails → follow up → log to CRM. Boundary: "does ~80% of the work" is a ceiling claim — complex buying signals, live objections, and closing stay human; research hallucinations must be spot-checked on the first batch of emails.
   • Examples: OpenClaw as a junior SDR (under $20/month vs $6,000+/month human cost, author-reported); n8n LinkedIn-to-cold-email pipeline (20–30 qualified calls booked per month, promotional-blog figure).
2. Cold-start audience growth — zero ad budget, growing a list via outreach and community engagement. Boundary: every outgoing message is human-approved before posting; the source stresses running the process manually hundreds of times before automating, or the output becomes uncontrolled.
   • Example: Hermes-driven cold start (1,000 subscribers in 80 days with zero ad spend; $20/month Codex subscription as the backend).
3. Ad operations & ad-data auditing — agents managing budgets or auditing ad data. Boundary: measurable metrics ≠ business value — the autonomous ad experiment broke lead quality; ROAS and similar key figures get misread by agents and must be re-checked by a human, one metric at a time.
   • Examples: Claude Code running Meta ads solo (CPL $6.14 vs a $2.50 target); Manus auditing Meta ads (reported ROAS 14.25x, actual 1.76x).
4. Social content monitoring & topic scouting — scrape posts, rank by save rate, morning Telegram digest. Boundary: platform data gaps force proxies (Instagram has no public save count — use comments/likes); wrong field mappings corrupt the ranking; records of this type carry no conversion metrics.
   • Example: Grok Bot + Apify viral-post tracker (150 posts scraped, 128 ranked in one run; top save rate 2.7).
5. SEO execution — bulk Meta-title rewrites, internal links, content refreshes. Boundary: the source offers only task-level time comparisons, zero quantified results; every batch is human-approved before going live — automation without guardrails is just faster mistakes.
   • Example: StudioHawk's Claude Code SEO workflow (claims 6–8-hour tasks compressed to minutes; unverified).
6. Support triage & escalation — from single-session support investigations to 22 million enterprise calls a year. Boundary: "5–10x throughput" and "87% faster" are vendor/platform figures; complex cases still go to humans, with AI coaching in real time.
   • Examples: Cursor's own support team (75%+ of support interactions run through Cursor); TELUS resolving calls proactively with Gemini (30% of calls automated, 87% faster resolution).
7. Full-funnel foreign-trade sales — background checks, lead scoring, quotes, 24/7 multilingual replies in one chain. Boundary: the 3%→8–20% inquiry conversion is supplier-reported with no independent audit; deep CRM and knowledge-base integration is a prerequisite.
   • Example: Foton trade agents (inquiry handling from 2–3 hours to minutes).

For sales visit notes — the workflow this domain is most often searched for — the deep dive on this site is the guide How to choose AI sales visit notes tooling? Run one pilot before you buy. Its anchor case (Huide Tech) is not in the table above: the source link returned HTTP 403 in this round of checks, so its figures could not be verified to this page's standard — see "Next steps."

For the remaining 259 records in this domain, browse the full catalog: it is the paginated list of all 2,001 cases and does not support stable category filtering (the ?category= parameter does not work server-side). This page vouches only for the 10 rows above.

Comparison table

Cross-case conclusions

1. Every quantified result is source-reported; we have re-run none of them. 10/10; four of ten come straight from vendors or platforms (Apify blog, Cursor's official blog, a Google Cloud customer story, 36kr supplier coverage) — treat as promotional.
2. Human involvement concentrates in setup and review; zero fully hands-off success stories in this sample. The closest-to-autonomous ad experiment needed forced human intervention on day 16; TELUS's complex cases still route to humans with AI coaching.
3. The same specific failure patterns recur: research hallucinations (OpenClaw case), key-metric misreading (Manus ROAS), quality-vs-cost optimization failure (Meta ads experiment), field-mapping corruption (Apify case), automation without guardrails as faster mistakes (StudioHawk's own words).
4. "From hours to minutes" time comparisons have no control data. StudioHawk and Foton both use them without verifiable quantified results — treat as direction, not conclusion.
5. The big support-automation numbers have zero independent verification. 5–10x throughput, 30%/87%/$53.9M annual savings — all from platform solutions or vendor blogs.

Verify it yourself

1. Open any case in the table and click its original source: confirm it is reachable, note who the author is (individual experiment, agency, vendor, or platform solution), and note the measurement window (e., "last 30 days").
2. Read the evidence label on the case page: treat "vendor self-report" as promotional; treat promotional-blog numbers as anecdotes.
3. Compare each case-page figure against the original source, watching the units — a "save rate of 2.7" is saves÷likes; "30% of calls automated" is not "30% resolved with no human involved."
4. Run the case page's "try it" prompt against your own business for two weeks and log your own reply rates, CPL, and human hours instead of quoting someone else's savings.
5. Revisit this table as a reminder: this site has re-tested no numbers; replace every figure with your own pilot data before any buying decision.

Next steps

• Start small: pick SDR outreach or ad-data auditing (clear boundaries, cheap failure cost), run a two-week pilot with the case page's "try it," and log your own numbers.
• Sales teams handling visit notes: read the guide How to choose AI sales visit notes tooling? Run one pilot before you buy — a single-workflow selection deep dive built on the Huide Tech case. That case is not in the table above: its source is a supplier case-study post on Zhihu that returned HTTP 403 in this round, so the figures quoted on the case page (visit notes 10–15 min → ~3 min, completeness +60%, 60%+ autonomous adoption after two weeks, retention +10%, 14 managers issuing 375 ShareAgent queries in two weeks) could not be verified to this page's standard; the case page itself flags the 403 and the lack of independent verification. The guide's pilot-first method stands unaffected.
• Keep exploring: browse the full catalog for the remaining 259 "Marketing & sales" records (a paginated list of all 2,001 cases with no stable category filter; most not line-checked by us).
• After you try: share your practice and numbers via the submission form on the homepage, noting the source and how humans stayed involved.