Being Human at Scale

When I said that the goal of an email campaign should be to appear as a "human at scale," this meant more than just using a conversational tone. What would you really say if you weren't automating this email and it wasn't part of an automated system? 

Sometimes, we'll have our clients show us what an email they already sent looks like. If you can replicate the tone and intention of that message, it'll be much easier for that kind of response to be automated. The offer matters, and appearing human matters more. Inject personality. A touch of humour or a relevant personal anecdote can work wonders. Write as you talk. If it sounds like a marketing brochure, you've failed.

Another good way to avoid seeming like a robot is to use news data to inject relevant industry news into your messages. Predictleads is gold for doing this. We also use online agents (like Claygent on Clay) to find mentions of the target company or relevant, industry-specific news. You might not be an expert at everything happening in the industries you're targeting, but if you can create some standard way to interpret things happening in the world and then find a way to fit that into your messaging, you'll be able to make much more convincing pitches. Set up alerts for your target accounts and reference major company events. However, if it feels forced, it'll backfire. Authenticity trumps automation.

Develop distinct message tracks for different buyer personas. The language that resonates with a CEO is different from what works for a CTO. The tone that they expect, the timing of the messages, and the kind of decisions they'd expect to make vary a lot. Sometimes, we send messages to CEOs and Owners over the weekend because they're definitely going to still be checking their emails (and all of the other automated outbound campaigns they're getting are probably paused). You can create similar strategic plans for different roles across your target companies. Customise their pain points and value propositions based on role and seniority. 

As already mentioned, at least 45% of your email should be AI-generated and unique to the receiver. We are not yet at the level where AI can write 100% of emails without hallucinations. You must, therefore, add a clear email structure and BLEND the personalization very carefully. The language and punctuation matter! Prompting is a skill that any good outbound marketer needs to know. Tell your LLMs exactly the words you want them to use and which words they should avoid. Explicitly tell it the goals you have – not to sound like a marketing email, to use a conversational tone, and to match the authority of the person who is receiving the email. 

For example, if you want AI to output the ICP of the person you're reaching out to, you should recognize that the idea customer for the head of the sales team (probably the purchaser at a company) might be different from the ICP of the heads of partnerships (who is probably looking for other heads of partnerships). Again, your strategy should be to use AI to do the quick, 10-minute research that you'd do yourself at scale. 

ABM + Steroids

Typical ABM is dead. Any book on sales would preach a strategy of creating a "market of one" around a select group of high-value accounts. With each of these targets, you'd create custom marketing plans, content, and strategies all centred around building relationships and driving a sale with that one person. The beauty of AI-powered outreach today is that every customer now has the potential to be that "high-value account ."If you can find the right workflow for automatically researching and profiling your leads, you now have the potential to build these kinds of customised marketing strategies at scale. Put more directly, it's now possible for each account to feel like you've been stalking them for months. (Because you have, just ethically.)

AI is steroids for your traditional ABM strategies. As these tools get better, you should try to integrate them into more and more thorough kinds of marketing campaigns. Instead of just scraping their websites and looking for their target audiences, why not use SimilarWeb to find out what their growth trajectory has been over the last few months? Then, you can offer a customised version of your product to those clients, offering some way to accelerate that growth using the exact numbers they discussed at a team meeting earlier in the day. AI should be like an intern on your team whose job it is to learn everything it can about your clients. 

Coordinate your outreach across email, LinkedIn, phone, and even physical mail. Make each touchpoint unique and valuable. For example, you could pre-warm with ads →, send an email →, send a LinkedIn connection request →, post organic content →, send a LinkedIn voicenote →, send another email →, make a call (Nooks/JustCall - for power diallers) →, send a WhatsApp or SMS. If you do this right, you can build a system of outreach that feels organic and gets you on the radar of potential clients. 

ABM isn't just marketing's job. Get sales, products, and even your C-suite involved for high-value accounts. Remember, if you've done the right research on your targets, you should have something that they might actually want. Your job is just to make them aware of this. Create custom microsites, personalised demos, and the works. Make them feel like they're already your best customer. Get everyone in the company to connect with 100 target accounts on LinkedIn weekly (it takes 15 mins to hit the weekly limit manually). Consistency and vision are the most important things here. 

The Art of Scaling

Automation is your friend; blind automation is your enemy. Automate the grunt work, not the brain work. Here is a brief overview of the state of AI right now (late 2024): 

We're in the later part of the first stage of LLM development which started with the release of ChatGPT in 2022. That was defined by systems (most chatbots or wrappers on chatbots) which could produce text based on prompts. That was quickly integrated into automated systems that wrote blogs or helped answer research questions. The biggest bottlenecks there were figuring out what prompts to use and overcoming things like context windows. These tools also got good at more complicated tasks like coding or developing strategies.

The next stage of AI tooling is probably going to be based on AI agents. These are tools that take in some kind of task and independently complete them over several steps. Now, instead of answering a question, LLMs can generate the next steps that they execute until some task is done. This opens up the door for much more complex relationships with AI tools and means that the things you use them for also need to change. Many companies already have these kinds of "agentic" abilities. 

Here's what a workflow that uses agents might look like: first, you set up systems to flag anomalies and opportunities. This could be based on things like new blogs or social media posts. Once these have been identified, you can trigger some kind of workflow that generates text or strategies based on what happened. You still need to have humans make the big decisions, but the overall strategy can be set in motion by one of these agents. 

Then, you should use these AI systems to test if they are working. If you're not testing, you're guessing. And guessing is for losers. Test everything. Subject lines, send times, even your signature. Try to test one variable at a time. With enough people in your pipeline, this kind of testing should be easy. The goal should be to figure out some way to build an outbound system that can improve itself over time. 

Your tools should grow with you. If they can't handle 10x your current volume, ditch them now. But don't just add tools for its sake. Each addition should solve a specific problem or unlock a new capability.