MarketSplash (@marketsplashcom) header
MarketSplash (@marketsplashcom) profile picture

MarketSplash develops modern marketing playbooks for the agentic web era. Articles, workflows, and free skills for your AI tools. All in one place.

Joined December 2019
Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)Photo by MarketSplash (@marketsplashcom)
1/ AI writes a personalized sales email for almost nothing, so personalization stopped earning replies. Every email now names the recipient's title and company. Naming them earns nothing. What earns a reply is a reason this account should hear from you now. 🧵
1
8
5/ What fails is anything true of thousands of companies at once. A funding round with no stated use. An industry trend. A jump in how much the account is reading about your category. Those describe a market, not an account.
1
3
1/ Neither Meta nor Google Ads has a permission that pauses ads and nothing else. Each one gives you read-only access, or edit access that can also change budgets and bids. Nothing in between. So an AI agent you let pause a wasteful ad can also raise your spend. 🧵
1
11
6/ So the morning script adds up the per-ad rows and compares them to the account total for that day. More than 1% apart, and it writes nothing and prints the mismatch instead of a brief. A quiet morning and a broken feed look identical otherwise.
1
2
1/ An AI agent that sends your follow-up emails can't tell a failed send from a lost confirmation. Both look the same from its side: the request left, nothing came back. So it sends again, and the prospect gets two emails. Nothing errors. 🧵
1
3
6/ In the study, 432 retries reused the original reference. None produced a duplicate. Retries whose first attempt went out with no reference duplicated 68% of the time. When every write accepted a reference, the overall duplicate rate fell from 28% to 4%.
1