AiSDR Product Spotlight #6: Pick the Best AI Model for Your Sales Outreach

One of the best things about AI is how fast it’s evolving.

(Now that I’ve said it, it’s also one of the most frightening things, but let’s put that aside for now). 

To date, AI has already turned from a standalone technology to an entire market, complete with core software and offsprings (such as AiSDR). 

The good thing is that you now have a choice between AI models or AI vendors. Just like any good set of market conditions would suggest.

The not-so-good thing is that sometimes there’s so much choice that you get decision paralysis, or you aren’t able to understand the differences well enough. Leading you to this 👇

And yet, the right directions and a good selection of AI options will put you in the position to grow faster. 

And that’s why starting this week, you now have a new option for experimenting with emails: The ability to pick from the 3 best AI models for generating your emails. 

So here goes. 

Choose between GPT-4o, GPT-o1, & Claude 3.5 Sonnet 

Each of these AI models have their fortes. 

And each can be used to achieve your outreach goals. 

That’s why we added the option to select the model for each sales persona or campaign you want to use inside AiSDR. 

This way, while building a new campaign, you can pick between these LLMs, preview messages that AiSDR generates, and compare results.

Similarly, when testing a persona or messages for a specific lead, you have a similar option that lets you see multiple variations so you can pick the best one.

What is each model for? 

While each LLM can essentially tackle the same task, the WAY it does and the QUALITY of the output is what sets each one apart. 

Here’s a brief overview of the strengths and weaknesses of each model, and why we settled on these three, despite the many LLM options available.

1. GPT-4o is a pumped-up version of GPT-4 that nails real-time interactions like no other. It turns out the fastest outputs, which makes it a great fit for chatbots, live AI agents, and such. 
2. GPT-o1 is all about advanced reasoning. Yet since it follows the ‘think before you say’ pattern, it is slower than the others. It’s awesome for complex analytical problems, programming, math, and anything where you might expect a more scientific approach. 
3. Claude 3.5 Sonnet (developed by Anthropic) can handle more context and provide more ethical and reliable outputs in a professional tone, though its reasoning is also quite strong. This LLM can be used for a wide range of tasks from coding to writing, math, analysis, and more. 

Though it is recommended to use tailored prompting to get the best of each model, you can also write model-agnostic prompts and test the outputs. 

Just be mindful of the amount of context you can feed into each. With GPT, you get 128K tokens, while Claude supports 200K. 

Why these three? 

Easy. They’re at the top when it comes to creating sales copy. 

As of now, at least. 

So here’s the big question… Which one should you choose? 

For starters, here’s a short side-by-side comparison.

On top of that, here are a few extra details for you to keep in mind when testing the LLMs for your outreach goals: 

• If you want large-scale campaigns to a huge lead list and you want the AI to not stray away from your key messaging, GPT-4o is your bestest bet.
• GPT-o1 is best for dealing with objections in follow-ups and analyzing pain points.
• GPT-o1 is also great for analyzing patterns in a prospect’s behavior, building personalization using datasets, and suggesting the most-valid next step in the sales funnel (which makes it great for nurturing too).
• With Claude 3.5 Sonnet, you are most compliant with anti-spam laws such as GDPR, CAN-SPAM, etc.
• Claude 3.5 Sonnet is also most capable of handling multi-threaded convos without letting anything slip and providing a thorough and relevant response.
• If you need to be extra careful with the tone (e.g. when reaching out to legal, healthcare, and similar industries), Claude 3.5 Sonnet would turn out the best results.

Naturally, there are many more differences, but the best way to spot them is to test! 

A quick internal experiment

To get you started, here’s a glimpse of the experiment our CEO Yuriy Zaremba ran and the results he got switching between these 3 models, along with the insights gained from running campaigns in his name with AiSDR.

As with many other things, the findings made their way to a LinkedIn post (make sure to follow him for more 🙂 )

And here are the emails Yuriy got with each of the LLMs he tested. 

GPT-4o

Claude 3.5 Sonnet

And also: 

GPT-o1 

As well as: 

Why is this cool? 

Who doesn’t like more control over AI-generated emails without having to make any changes?

With a variety of models, you can align each campaign to the goal of your outreach, and use the best AI LMM to support your growth and tone to translate it into your sales performance. 

And while before we were picking the right model for you based on our experience, now you’re in the driver’s seat to make the right call on your own, based on the previews of the emails generated with each supported AI model.

Also, this empowers you to:

• Use more data in your outreach
• Run a larger number of sales plays aimed at different goals (apart from booking demos)
• Have the AI better understand the target audience and adjust to their expectations accordingly

🤫 Next week’s sneak peek

We are firm believers that the sky is the limit when it comes to integrations with CRMs, so more cool updates are coming up on the HubSpot front. Stay tuned!