narratic guide

Personalization at Scale for B2B Sales

Finding B2B leads

Arne Wolfewicz
Co-Founder & CEO

There are numerous ways to find B2B leads. Since we are focusing on strategies to directly connect with suitable candidates, we must start our search in the right place. 

We are going to use LinkedIn because unless you target a very narrow niche where your targets are alien to networking, this is where you should start. Besides, most platforms nowadays feed off similar databases and it’s safe to assume that all of the big ones have some anchor on LinkedIn.

The more important question for finding leads is how to find the right ones. Most people get this wrong by a lot because ad platforms have been teaching us to cast a wide net, suggesting that a bigger audience is better. If you have billions of data points for optimizing your delivery, this is the right approach.

The picture looks very different when it comes to directly targeting individuals. Regardless of how well your message is written: You will not only be wasting your recipients’ time but also your own.

There are two scalable approaches we found to be useful: Using boolean search (basic) or similarity search (advanced).

Creating your Lead List

Basic: Finding the Right Leads Using Boolean Search

You are probably familiar with typical filters all modern tools have: Company size, industry, job title, has changed jobs in the last 3 months, etc. In case you haven’t used them, it’s time to shed light on a technique fewer people know about: Boolean search. 

We won’t go deep into theory since we wrote a blog post on the subject and pasted the ChatGPT prompt you need below. But let’s examine an example in practice:

Current job title: "Revenue Operations" OR "RevOps" NOT ("expert" OR "founder" OR "cofounder" OR "manager")

Even if you are seeing this for the first time you can probably tell what we are looking for: Either you or the RevOps person sitting across the room. In addition, we excluded all consultants or founders of companies that support RevOps. Nothing against them – they are just not our target audience.

This can already get you quite far. However, the real fun starts once you use this technique on a second field:

Past job title: "SDR" OR "sales development representative" OR "BDR" or "business development representative"

By adding a filter for the past job, we went from 9,000+ RevOps practitioners to 241 experienced RevOps practitioners who also know what it means to work as an SDR. This list is way more valuable than the first. Here is why:

  1. Adding another set of conditions for past positions gives us a pretty sharp audience of people who are probably interested in what we are doing.
  2. Connecting with 241 people is more than 37x faster than 9,000.
  3. We don’t even need ChatGPT to personalize our message – it’s sufficient to mention the fact we just learned about them.
  4. If we use ChatGPT, we can feed this information as additional context it can actually make sense of.

Here is the ChatGPT prompt that typically gets you strong results – refine as needed:

Situation: I want to use the boolean search on Sales Navigator on the job title field. Give me a boolean expression I can copy paste. Make sure to exclude non-decision makers from the search.

Task: Act as a revenue operations expert. Your job is to craft the perfect boolean search to find decision makers working as [role] in the [industry] industry.

Rules: Use AND, OR, NOT, parentheses, quotes as needed. No wrapper.

Advanced: Finding the Right Leads Using Similarity Search

There are great ways to hack searches with good keyword combinations and you will quickly improve the quality of your lists. But there are a few problems with the conventional way of using keyword and categorical filters that limit how powerful your system can become:

  1. Your information is only as good as the data of your platform
  2. Every person in the world has access to the exact same information
  3. You will exhaust your search eventually

We found that an entirely different approach worked much better than everything we did previously: Similarity search. If you are familiar with how lookalike audiences work on ad platforms, that’s what we are trying to achieve – find people similar to the customers you already have or that fit your ideal customer profile.

The remainder of this section gets pretty technical but it’s worth thinking through if you want to level up your game.

A Quick Introduction to Similarity Search

Imagine you're at a massive global conference with thousands of attendees. You've just had an engaging conversation with someone who's an ideal lead for your business. Now, wouldn't it be great if you could instantly identify other attendees who are similar to this person?

This is where similarity search comes into play, but in the digital realm.

At its core, similarity search is like a digital magnifying glass. Given a sample lead or a set of criteria that represent your ideal lead, similarity search scours through a vast database to find other leads that "look" or "behave" similarly. Instead of manually sifting through thousands of records, this method pinpoints potential leads based on a pattern of likeness.

In technical terms, every lead in your database can be represented as a data point. Similarity search then measures how close or distant other data points are from your sample. The closer they are, the more similar they're deemed to be.

Why Similarity Search is Powerful for B2B Sales

Even if you don’t have hundreds of customers, many of your buyers will share certain properties – some you know about, some you don’t. Even known properties can be hard to identify as they may be hidden in someone’s job description or expressed in the way they post and interact with other people’s content.

Where things get really tricky is when the leading cause for someone to be a “great customer” is hidden in subtle nuances or combinations of certain properties. We previously examined the combination of “the person used to work in Sales and is now running RevOps” but, compared to what similarity search is capable of, this is still only scratching the surface.

The biggest advantages of searching for leads using similarity search can be summarized as follows:

  • Precision: Instead of broad strokes, you're targeting leads with a higher likelihood of interest or relevance to your offering.
  • Efficiency: No more wading through irrelevant contacts. Similarity search drastically reduces the time spent in lead qualification.
  • Scalability: Especially in large databases, this method ensures you can find and prioritize high-value leads consistently.

In short, similarity search is like having a superpower. From a single ideal lead, you can unearth a goldmine of potential clients, all thanks to the magic of data comparison.

How to Get Started with Similarity Search

As you might have guessed, such a system is not available off the shelf. But to give you some guidance in case you are considering such a move, here are the high-level steps we took:

  1. Purchase a large dataset of LinkedIn profiles (e.g. Proxycurl’s LinkDB)
  2. Consolidate all lead and customer data into a database (e.g. BigQuery)
  3. Choose an embedding algorithm that turns your profile data into individual vectors (e.g. OpenAI’s text embedding endpoint handily comes with similarity measures built-in)
  4. Rank the list of prospects according to similarity
  5. Contact in that order

With such a system, you can do a lot more than just finding similar people – it allows you to match any kind of data with your database of leads, such as specific offers, value propositions, or content they are (mathematically) interested in. We get back to that in our chapter on using embeddings for advanced targeting and messaging.

Other methods

Besides these two, there are other approaches to finding suitable candidates for your outbound game. The possibilities are endless – to name the major ones:

  • Alumni networks
  • Networking events, trade shows, conferences
  • Industry forums and groups
  • Team pages on websites

There is nothing wrong with any of them. In fact, some of them will work very well for growing your business. However, they all share the property of not being very scalable which is what we are after.

Three Iterations to Improve your Targeting

There are essentially two platforms to do proper cold outreach: Email and directly on LinkedIn. In both cases you would have to extract your lead list you just created. We don’t advertise doing this but if you wanted to extract contacts from LinkedIn, here are a few ways you could do this.

There are three distinct phases that we have seen ourselves and our clients go through time and time again. 

Phase 1: Getting Started

Tools: LinkedIn Sales Navigator + Phantombuster

Every experienced B2B growth person will have heard the word Phantombuster at some point. We’ve been using the tool for many years and it is perfectly suited to get things done.

To extract your list, you need to do two things:

  1. Scrape all profiles for your search using the Search Export Phantom
  2. Enrich all profiles using the Profile Scraper Phantom

Be sure to closely follow their suggested rate limits for LinkedIn. For detailed information on rate limits, read this article.

This system will fill your supply side for a couple weeks: At the time of writing this, the daily limit for scraped profiles will be at around 150 per account, meaning that your list will grow by 150 x 7 x 4 = 4,200 contacts each month.

Important note: Before we start discussing tools to add more capacity to your system, remember that the point of scalable personalization is to reduce the number of contacts you need to make. If done right, you shouldn’t have to contact thousands of leads every week for a handful of qualified leads. Instead, you will only contact the most relevant people and start the relationship on the right foot.

Phase 2: Removing Bottlenecks

Tools: Same as above + pen + paper + your brain

Every sales process, no matter how refined, inevitably encounters bottlenecks. Addressing these bottlenecks is not just a remedy; it's an opportunity to exponentially (!) improve your funnel's throughput.

Bottlenecks are the points where the smallest improvements create the biggest boost in results. That's why they're so important. We get the biggest bang for our buck. 

For illustration, imagine you have three steps in your process: 

30% Open

5% Reply ← The constraint since it has the biggest drop-off 

50% Schedule 

Imagine we improve each step by 5% by itself.

30 + 5% → 35% Open = 16% Increase in leads (1.16x) 

5 + 5% → 10% Reply = 100% Increase in leads (2x) 

50 + 5% → 55% Schedule = 10% Increase in leads (1.1x)

Given that all enhancements are similarly hard to implement, it is clear what you should be focusing on. In most cases concerning outreach, you will probably need to work on what and whom you write, not how many.

Phase 3: Pushing the Boundaries

Tools: Same as above + data API; optional: Backend Developer

Once your system works and you are happy with your funnel metrics, you will inevitably want to increase your capacity. When that happens, you have two options: Upgrade a second LinkedIn account to Sales Navigator or use a data vendor to purchase profile information from.

The first option is viable as long as you are able to control the complexity. But in any case, it requires additional work and you will ultimately hit the next ceiling.

If you are in it for the long run, you might want to consider buying data from services like Bright Data or Proxycurl. You could still call these APIs with no-code tools like Zapier but at this point your work should be so valuable that you can afford a Backend Developer to build a proper system.

The benefit of using such services is that you don’t risk your whole process being terminated by LinkedIn blocking your account(s). As such, it is future-proof to the extent that these services remain online – which they probably will.

That being said, these services typically have drawbacks when it comes to data quality and you (or your developer) need to build in checks accordingly.

In terms of what profiles to query, you probably want to keep extracting your searches as you did in the beginning as long as possible. Filtering through those vendors’ endpoints is everything but practical and errors always come at a price.

We will take a closer look at crafting personalized emails with ChatGPT in the next part.