Personalisation at scale without sounding like AI: 8 patterns with real examples

Personalise one thing that is true, specific and tied to why you are writing: their role, their employer's products, their customers' problem, their own words. Skip the fake familiarity of a name and a compliment. Real messages from our case studies show eight patterns that read as human at scale, and each rests on a fact the agent actually had.

Everyone can tell when a message was mass produced. A first name merged into a template is not personalisation, and neither is "I loved your recent post!" with nothing behind it. What works is one detail that only fits this person and shows you looked. Below are eight patterns. For each, a generic version written for this post, and a real message from one of our published case studies (people's names are changed, company names are in brackets as in the studies).

1. The role specific question

Generic: "Hi Matt, I help engineering teams improve efficiency. Do you have 15 minutes?"

Real (Lamothermic, US foundry): "Matt, as Director of Engineering at [company], what's the biggest challenge you're currently facing with the high-performance metal components for your drilling equipment?"

It names the role and the product line. The question is a diagnostic one an engineer can answer in a line, and it has no pitch and no link. The engineer replied, and the agent then narrowed with an either/or question.

2. The employer's products

Generic: "I see you work at a great company in manufacturing."

Real (Lamothermic): "Dan, given [company]'s focus on mission-critical components for aerospace and defense, I'm curious. Do you ever find it a challenge to source complex investment castings that come fully machined and finished from a single, reliable partner?"

What makes it work is that "given X" is true and checkable, and the question connects it to a real problem the sender can solve.

3. The customer's problem

Generic: "We help sales teams close more deals."

Real (Lamothermic): "Hi Cole, I saw your work at [company] with industrial automation solutions. Are your manufacturing clients running into any challenges with component reliability or sourcing for critical parts right now?"

This one is for a person who sells to the people you want to reach. The agent asks about his customers' pain, not his own. Reply, then a narrowing question one minute later.

4. The peer frame

Generic: "I'd love to add you to my professional network."

Real (wealth advisor referral network, connection note): "Hi Dana, I saw your profile as a Registered Migration Agent in Australia. I'm also in the immigration advisory space and would be interested in connecting."

A true statement of common ground, no pitch. In that study the second message turned the acceptance into a question about the person's clients.

5. The headline question

Generic: "How do you approach business growth?"

Real (same study): "Hey Rob. With your focus on [his niche], how do you approach identifying unique international real estate opportunities for your HNW clients?"

The question is built from the person's own headline. It can only be asked of him.

6. The observation

Generic: "Great profile!"

Real (Sellinger, our own outbound): "Niko, I was looking at your work in AI and NoCode for business automation. Seems like your focus on scaling outreach and cost savings is making a tangible difference for your clients."

A statement, no question, true to the profile. It is the style that out-replied questions by 10 to 15 points in our data. See observation openers beat questions.

7. Their words back

Generic: "Thanks for your reply! That's really interesting."

Real (Sellinger, a later message): "That's a really comprehensive system you've built, especially the self-improving prompts idea. [...] That's exactly what Sellinger does, but as a dedicated solution. Might be worth a quick 15-min look?"

After a person has written, the best personalisation is what they said. Repeating two or three specifics shows the message was read, and it is the cheapest form because the data is in the thread. See what a good AI reply looks like.

8. The easy out

Generic: "Just checking in!"

Real (Sellinger): "Following up on this. Is it fair to assume outbound sales isn't a focus for [his products] right now?"

It names his products and gives him a way to say no. The product leader replied two days later and booked after a further nudge. A follow up personalised to the situation gets answers that a bump does not.

Which pattern fits which person?

Pick the pattern from the data you hold, not from taste. A rough map, built from the eight above:

The table also shows the limit. If the row says "nothing else", the answer is pattern 6 or a shorter message, never an invented detail.

A made up example: thin data against rich data

The two leads below are invented for this post. They show how the same agent should behave when the data differs.

Lead A (made up): a head of operations at a regional food distributor. The lead record has a role, a company name and what the company sells. Pattern 2 fits: "Given [company]'s focus on chilled distribution, do you ever find it hard to keep delivery windows reliable when order volumes swing?" Every clause is checkable.

Lead B (made up): a consultant whose profile holds a name, a city and a job title with no description. There is nothing true to build a pattern on. The right first message is short and plain: "Hi, I work with consultants on outbound. Is finding new clients something you are focused on right now?" That reads as honest because it makes no claim it cannot back.

The mistake to avoid is writing Lead A's message for Lead B and filling the gap with guesses.

What should you not personalise?

• Flattery. "Impressive background!" costs nothing and signals nothing.
• Invented details. If the agent does not have the fact, it must not write one. A wrong personal detail ends the conversation.
• Trivia. The school they attended in 2009 adds creepiness, not relevance.
• Everything at once. One detail. Not five.
• The trigger you do not mention. Only 5 of about 30,000 of our own openers mentioned the post that put the person on the list. If you have the signal, use it.

How does it work at scale?

Personalisation at scale depends on data, not on typing speed. For each person the agent needs a role, a company, what the company sells, a recent post or event, and later the conversation itself. The agent researches each prospect, writes the first message and every follow up, and uses only facts present in the lead data. Where the data is thin, the answer is a shorter message, not a made up detail.

How do you check personalisation after it ships?

Do not judge a campaign by reading three good messages. Read a sample and sort it.

1. Pull twenty first messages from one campaign, taken at random.
2. For each, find the one detail that ties the message to the person. Mark it true and checkable, true but generic, or missing.
3. Count the missing ones. Each is a data problem or a rule problem, not a wording problem.
4. Compare reply rates by opener style once you have enough volume. On our own account the style comparison was the clearest signal: observation openers out-replied question openers in every month compared.
5. Change one rule, then sample again.

Personalisation is a loop between the data the agent holds and the rules it follows. Fix the data first, then the rule.

A quick test

Read a message. Remove the first name. Could it have been sent to anyone in the list? If yes, rewrite it. Then read it as the recipient. Is there one sentence that shows you know something about them? If not, the message is not personalised, whatever the merge fields say.