The scalable personalization tactic that boosted email clicks by 117%

Today’s expert shares an experiment I would have wept tears of joy over back then. It’s the at-scale hyper-personalization tactic that boosted email click rate by 117%. Plus, an example of AI actually eliminating the miserable part.

And she’s not just sharing what to do, but the actual steps (including the prompts) she took to do it.

Getting Personal

Personalization is a force multiplier for any kind of email marketing, but to test this new tactic, Ranjani Nagarajan and her team chose a notoriously frustrating audience. Roughly 4,000 monthly free users who visit a pricing page, but never make the jump.

Ranjani is a senior marketer on HubSpot’s Global Self-Service Monetization team, so she’s an expert in helping folks help themselves. And knowing when they need a little nudge.

“Before, we would email hundreds or thousands of contacts on a monthly basis. They’re showing some signal. They’re going to the website, or the in-app portal, they’re looking at pricing. We used to send them the same email like, ‘Hey [first name], we saw that you viewed the pricing page. We think this could be a great fit for you.’”

You’ve probably been on both ends of a campaign like this. For a long time, it’s been about the best a marketer can do, but it wasn’t enough to move this stubborn segment because it wasn’t addressing their unique concerns.

We’ve long had the data to do true personalization, but never had the bandwidth. Ranjani and colleagues wondered if they could use AI to close that gap.

“Acknowledging an action and then adding how our product will benefit their business doesn’t really need AI. [But] without contextual analysis, it’s just template-level personalization. What drives lift is context,” she explains. “So if it’s a CEO reading it or a marketing manager reading it, the context changes because you need to approach them with different lenses. AI can do this for literally every single contact.”

“I wouldn’t say it’s perfect, but for the first time, we’re able to do true hyper-personalization at scale. And the results speak for themselves.”

In addition to improving email click rate, the experiment bumped the free-user-to-purchase conversion rate by 129% — adding several thousand dollars of monthly revenue from just one segment on just one page.

Building a Data Layer

There are two keys to this tactic, and it should surprise no one that the first is high-quality data. Ranjani says that to do this successfully, you’ll need to gather the following:

• The specific trigger action (and what it implies).

“Behavioral data is huge for personalization, because we’re not just sending them an email. We’re sending them an email that connects back to an action they took.”

In our case, it was which specific pricing page that a given visitor viewed. Your trigger might be adding an item to a cart, downloading a whitepaper, performing an in-app action, or a million other things. Choose something that’s a clear signal of buying intent. This is going to be really important in just a minute.

• A contact profile.

“This is basically gathering as much information as possible about this contact. We have a mechanism to do the research.”

I won’t belabor this one because the HubSpot Blog has a trillion articles on building prospect profiles. The important part is that you build a thorough profile and feed it into your CRM. (Can you guess which one we use?)

• A company profile.

To build real context, it’s not enough just to know who they are. You also need to understand their goals and pain points. That takes firmographic data.

“We’re talking about thousands of contacts, so basically every contact would [have their website crawled], then we would build a company profile for them. We would also look into the industry this contact belongs to, maybe pull some industry pain points.”

Now, what do we do with all this dang data?

Prompt Attention

The second key to this tactic is clear and concise prompting, which isn’t as straightforward as a lot of AI influencers try to make it sound. To put all this data to work, Ranjani and her team created two layers of prompts.

The System Prompt

“This is the standing instruction. It defines who the model is, what it’s doing, and what rules it always follows. It’s set once and runs in the background for every request,” Ranjani explains. “Think of it like the chef’s training. The techniques, standards, and rules that never change regardless of what’s ordered.”

The system prompt is part ground rules and part guardrails. It should include:

• Voice and tone instructions
• Email format
• Rules and limitations
• All do’s and dont’s
• Examples of good content

If you use an AI tool that is project based, you can set the system prompt as part of the project parameters.

The User Prompt

“This is the live input. It’s what gets sent at runtime. Think of it like tonight’s order at the restaurant. The kitchen uses its standing training to fulfill it, but differently each time and consistently well.”

While the system prompt addresses the overall campaign, the user prompt tells the AI how to use the data specific to this individual recipient.

This part of the prompt includes:

• Contact name
• Company name
• Industry
• Audience segment
• The trigger signal

A prompt in five acts.

All of this is great in theory, but what does it look like in real life?

Down below, I’ve included an actual, functional, (totally swipeable) prompt that Ranjani shared with me. But in her wild generosity, she’s also shared her team’s five-part formula so you can understand it deeply enough to adapt it to your business — or even make prompts of your own.

1. The Setup.

The AI writes to whoever you describe, so be specific.

2. The Blueprint.

This section is a detailed map of the email structure. It includes the number of paragraphs, the reason for each paragraph, rough word counts for each paragraph, and what each paragraph should accomplish.

• The opener. Tell the AI to mention the trigger action as soon as possible.

• The situation. Describe an industry challenge or pain point and what’s possible with your product or service.

• The close. Close with one reason it’s relevant for them, followed by one clear CTA.

3. The Rules.

“Left to its defaults, AI tends toward polished and generic. It ends up robotizing. I don’t know if that’s a word.” Ranjani laughs. “But these instructions keep the output grounded and human.”

These will change dramatically based on your industry, audience, and circumstances, but here are some examples:

• YES: Focus on the outcome for them, not the mechanics of your product.
• YES: One clear CTA. End there.
• NO: Benefits broad enough to apply to any company in any industry.
• NO: Multiple options or a softer alternative. (A single ask is always stronger.)

4. The Craft.

These are the best practices and writing principles that you might give to a human writer in a campaign brief.

Think of the AI as a junior writer that needs a little more hand-holding than a seasoned pro. You’ll want to include things like:

• Write to one person, not a segment.
• One idea per paragraph.
• Never use three words where one word will do.
• Mention their company or industry in context, not just in a greeting.
• Overall wordcount should be X.
• Include the trigger action in the subject line and first paragraph.

5. The Proof.

Lastly, show the AI examples of what good copy looks like and campaigns that have worked in the past.

If you really want to make it work, you can annotate your examples to highlight important structures and define why they’re effective.

Taste-making tips

If you’ve made it this far, I know you’re eager to jump in, but stick with me just 30 seconds longer. Ranjani shared the practical tips that will take you from a prompt padawan to a marketing Jedi master.

1. You have to be the tastemaker.

“Before you use AI, do you know what a good email looks like? Don’t let AI tools decide what good looks like. You, as a tastemaker, need to express to the AI whatever you want it to do.”

Ranjani says this tip is actually more important than the data, the format, or even the prompt.

“If you don’t know what good looks like, no matter how much AI you use, it’s never going to give you the results you want.”

2. Clarity over quantity.

While thoroughness is encouraged for every section of the prompt, Ranjani cautions that it is possible to go overboard.

“A tight, unambiguous prompt is the most important decision in any AI content program. More rules don’t make it better. Clarity does.”

3. Start with only one or two data points.

“When people think of AI personalization, they want to do everything all at once. Tick all the data points. But you end up overwhelming the system and the AI starts hallucinating.”

Once you’ve lost trust with a bizarre, false, or creepy email, it’s really hard to get it back. So start small and work your way up.

As you become more comfortable with prompting, and more confident in the system you’ve created, you can gradually layer in more data.

4. Choose the right action for the right trigger point.

The two audiences you don’t want to reach are people who will never buy or people who were about to buy anyway.

“We shouldn’t be emailing people for the sake of emailing. We should be emailing them if we think that is going to give extra value.”

So choose a trigger action that would ordinarily be a clear purchasing signal, but for whatever reason has stalled out.

“It all depends on intent,” Ranjani says. “I would bucket [the signals] as low intent, medium intent, or high intent. If somebody’s in the low or medium [bucket], you should be giving them more helpful resources. But if somebody is high-intent — viewing your product page, abandoning the cart — those are signals to push them towards a conversion.”

5. Acknowledge the trigger loudly and clearly.

If I can’t tell why a company is emailing me in roughly 30 seconds, I’m deleting the email and shaking my fist in their general direction.

“When someone reads an email, they should immediately think, ‘There’s a very good reason why this company sent me this email.’ That justification has to be loud, strong, and clear.”

At minimum, you should instruct the AI to mention the trigger within the first two sentences, if not the subject line. In fact, do it again even if it’s in the subject line.

“They should know it’s a follow-up to what they just did, not a random email. That’s what differentiates an automated email versus a marketing email.”

And this doesn’t mean a generic, “You viewed our pricing page.” You can do that with boring email templates. Instead, it should be, “I noticed you viewed this specific pricing page, so here’s a very specific resource that digs deeper for your specific industry.”

6. Quality test rigorously and repeatedly.

The more content you ask AI to generate, the more likely it is to hallucinate.

“One thing I notice AI will do is make up statistics. It’ll say, ‘93% of HubSpot customers are able to close deals with HubSpot.’ So we had to put up guardrails like, ‘Don’t make up random numbers.’”

But even with strong guardrails in place, you’ll want to thoroughly quality test the output before launching, and then occasionally sample the outcome even after launching.

“Before we put anything into production, we do rigorous testing. AI would generate hundreds of copies and we would go through plenty of them manually. That’s the hard work you need to do now in order to stop AI from hallucinating in the future.”

Even though the chef is trained, and the order is clear, they’ll still taste the soup before it goes out.

Prompt without the pomp.

Finally! While HubSpot’s sensitive details have been scrubbed, here’s a real-life prompt you can copy. Leave the section headers, but swap the details to meet your needs, then feed it to any AI writing tool.

SECTION 1 — WHO YOU ARE

Your role

You are an expert B2B marketing copywriter. You write short, consultative outreach emails that feel human and earn trust quickly. You write with clarity and confidence — never corporate, never pushy. Every word earns its place.

SECTION 2 — WHO THEY ARE

Your audience

You are writing to demand generation managers at mid-market B2B SaaS companies — typically 50 to 500 employees. They own the email channel and are responsible for the pipeline. They’re busy, they’re skeptical, and they’ve seen every template going.

SECTION 3 — WHY NOW

The signal

This person has been exploring the product on the website — they visited the features page and the pricing page in the same session, but didn’t convert or reach out. They’re in an evaluation phase. That’s your opening.

SECTION 4 — THE BLUEPRINT

The structure

Write three paragraphs.

• Paragraph 1 (~35 words): Open with the signal — they explored the product on the website. Make a reasonable guess about what’s prompting that.
• Paragraph 2 (~60 words): Describe the situation they’re likely in — evaluating tools to drive more pipeline without adding complexity — and what becomes possible when they solve it.
• Paragraph 3 (~20 words): One reason this is relevant for a team at their stage. One clear CTA.

SECTION 5 — THE RULES AND THE CRAFT

How to write it

Write like a peer. Use “you” and “your” throughout. Contractions where natural. Active voice. Short sentences.

Open with the signal — not with a pleasantry or a warm-up line. Focus on what shifts for the reader, not what the product does. The reason for reaching out should be clear within the first two sentences. One CTA only. Total: 120–150 words.