Blog · AI & Outreach
December 3, 2025 · 5 min read
AI-powered personalisation at scale has become the dominant promise in outbound sales technology. Feed in a list of prospects, run it through a language model, and get back emails that feel individually crafted. The promise is real — but most teams implementing it make a predictable set of mistakes that turn a potentially powerful approach into expensive slop.
These are the five most common mistakes, based on reviewing the sequences of teams that tried AI personalisation and gave up — and the ones that tried it and actually saw it work.
The biggest mistake is handing the writing entirely to AI. "Generate a personalised cold email for this prospect" produces an email that sounds personalised but contains no specific insight that the AI could only have written for that person. The personalisation is structural — it uses the right name and company — but the content is generic.
AI works as a co-writer or editor, not as the author. A human decides what observation to make about the prospect and why it is relevant. AI turns that observation into a well-crafted first line. The research and judgment stays human; the copy execution is AI-assisted.
"I saw that [Company] recently expanded to a new market" is a personalised opening. If the rest of the email is a generic pitch about your product, the personalisation is a veneer — and buyers can feel the difference. The opening hook needs to connect to the problem you are addressing and the offer you are making. Personalisation that leads to a generic pitch is worse than no personalisation because it sets expectations the email then fails to meet.
Every personalised element in an email should connect to why that specific prospect, in their specific situation, should care about what you are offering. If the connection is not clear, the personalisation is decoration.
Teams often find one personalisation signal that works — recent job changes, LinkedIn activity, company news — and apply it universally. When 80% of your sequences start with a reference to the prospect's recent LinkedIn post, it stops feeling personal and starts feeling like a pattern. Prospects talk to each other, and they compare notes.
Build a library of personalisation signals and rotate them. Some prospects get a hook based on their company's recent hiring activity. Others get a hook based on a market trend specific to their industry. Others get a hook based on a specific result relevant to their segment. Variety in signal type is as important as variety in copy.
Mistake four is skipping human review of AI output before it goes to prospects. AI hallucinate facts. It invents details, misattributes quotes, and states things as confident facts that are wrong. An email that confidently states an incorrect fact about the prospect's company — a wrong revenue figure, a non-existent product launch — is worse than a generic email. Build a review step.
Mistake five is failing to measure whether AI personalisation actually improves results. Run a controlled test: the same list and message with AI-personalised first lines vs. without. Measure reply rates over four weeks. If personalisation is not lifting results, the approach needs to change. Many teams assume personalisation is working and never check.