#5-Bohemian RhapsAI: ReplayzIQ's AI Sales Coaching Revolution, GTM AI Tools Library Launch, Claude 3.0 vs ChatGPT vs Gemini, and Navigating AI

Every time I think nothing crazy will happen to write about in 7 days, something inevitably happens, which I am excited to tell/show you!

As always, this newsletter and podcast is sponsored by GTM AI Academy which we are about education, training, and AI Enablement for GTM teams, individuals, and leaders.

As a thank you from me for being a part of this journey, if you want to join in the Academy, you can get 30% off any bundle or course using the code: podcast30 over on https://www.gtmaiacademy.com/courses Right now, we have Generative AI Foundations course, which is a great place to start for most of you who are new to AI and then Enablement AI Powered which is specifically for L&D, Enablement, or even Revops people who want to dive into AI workflows specifically for sales enablement, revenue enablement, etc.

Today we will be going over:

1. Podcast #5 summary with Dave Kennett of ReplayzIQ (Joined forces with Highspot) which is an amazing AI call scoring tool.
2. GTM AI TOOLS LIBRARY LAUNCH!
3. Claude New Release, better than ChatGPT?
4. How do you know what tool to use or where do you even start?

Lets do this...

You can hear the podcast on Youtube, Apple, Spotify, or here on Linkedin as well as a whole other plethora of locations.

I wanted to share some of the key insights and takeaways from my fascinating discussion with Dave from ReplayzIQ. We really dove deep into how they're utilizing AI to transform sales coaching and enablement. Here are the highlights:

• ReplayzIQ is tackling the issue of sales leaders not having enough time or sometimes lacking confidence to effectively coach their reps by reviewing calls. The current tech focuses more on call recording than actual coaching.
• They've developed a proprietary AI model that's trained on a massive amount of labeled sales call data. This allows it to accurately assess calls based on over 150 sales skills, tailored to a company's unique sales process and personas.
• In Dave's words, "We had 24 amazing sales coaches coaching these fantastic companies and sales reps. And we've got this like hundreds of hours of recorded data that's now labeled data that has informed our models. That's the moat." (For those of you that do not know what 'moat' means, in AI terms, its the unique data that the tool is trained on that no one else has access to, thus making it unique and building a moat around the solution)
• The AI doesn't just score calls but provides actionable recommendations on the most important skills to coach reps on at each stage of the sales process to boost performance. This has led to eye-opening discoveries for customers. As Dave mentioned, "Every single time ReplayzIQ delivers that for a customer, it's a surprise to them. And it's a surprise to me because I would expect to see the things that you and I might think... And every time it's different things because it's based on that data of that customer."
• ReplayzIQ is developing an AI sales assistant to aid reps on calls in real-time and will be conducting "human vs AI" coaching contests with sales experts.
• For trusting AI tools, Dave recommends confirming the AI is trained on high-quality, applicable data and testing it yourself to verify accuracy and value. Don't anticipate perfection, but top AI should surpass human performance. In his words, "Once you trust the tech, the next thing you need to do is know if it's accurate, know if it's been trained on the right data that is, a deep has deep context in the subject matter that you're looking at..."

Key Takeaways:

1. AI coaching tools like ReplayzIQ can deliver scalable, data-driven sales coaching that pinpoints the highest-impact skills to focus on for each rep and stage of the sales process.
2. Surprising insights emerge when you base coaching on actual sales call data rather than assumptions.
3. AI sales assistants could be a game-changer for supporting reps in real-time on calls.
4. Verifying the quality of the AI's training data and testing it yourself are key to trusting and getting value from these tools.

The discussion really underscored the huge potential of AI to personalize and scale sales coaching in ways that drive meaningful improvements in performance. As Dave aptly put it, "I think sales reps deserve more support out there. And this is, one easy and quick way to do that."

All in all, I was struck by how thoughtful Dave and the ReplayzIQ team are being in their approach - focusing on building robust AI models trained on quality data, validating results, and truly aiming to empower sales reps and leaders. It's not about replacing humans but equipping them with powerful tools to succeed.

I'd love to hear your thoughts! What stood out to you from the conversation?

ITS TIME!

My friends it is time... The GTM AI Tools Demo Library is now live.

You will need to register for it, but just requires an email and you are in!

So over the last several months, I have been demo'ing tools both for necessity for my team and curiosity because I am loving all the developments.

I always wanted to have a central place where I could see demos because as much as I love marketing videos, (I am leading a content marketing team after all) there is a difference between professionally produced marketing videos and a real demo.

So I have split them out into the best categories I could think of in the course.

I hope you will find value in these demos and see what would be helpful for you.

With permission from each team, I have done the following:

• Trimmed down the demos (so no small talk or listening to my shananigans)
• Any talk of pricing is taken out
• Each one has their website listed for you to go to and find out more

Some of the tools are:

SellMeThisPen Momentum.io GTM Buddy Scalenut Wordly Replicate Labs bizPROFI Second Nature RNMKRS Copy.ai Regie.ai Truebase Ubique Kipsy Grain.ai CloseStrong.ai Quantified Spiky.AI Swyft AI Glyphic Uniphore WINN.AI Tough Customer AI and more

If you want to have your tool featured in the library, you must provide a demo like video that is more real and is like a conversation talking to me, like on a Loom that I can download. Sending me a marketing video that already exists on your website will not cut it, I want real, pretty please ;) Just message me.

Claude 3 vs ChatGPT vs Gemini

Anthropic who owns Claude just released their new Claude 3.0 model and let me tell you, it is impressive.

You may be asking, why would I use each of these tools over the other? Good question, but before we jump into that, lets talk about the updates to Claude 3 and why it matters to you as a GTM pro. Now there are all these items in the above list, let me break those down first:

First Shots and CoT (Chain of Thought)

The concept of 0-shot, 3-shot, 5-shot, 8-shot, and 10-shot learning in AI can be explained using a cooking analogy. Just as a chef's ability to create a dish improves with more guidance and examples, an AI model's performance enhances as it is provided with an increasing number of examples or "shots" to learn from.

Starting from 0-shot learning, where the model relies solely on its pre-existing knowledge, to 10-shot learning, where it has access to a wealth of examples, the AI's capacity to understand and generalize to new tasks grows, much like a chef's skills improve from being a novice to mastering the culinary arts.

Now What are all these tests?

Alright, let's dive into the nitty-gritty of these AI benchmarks and tests that everyone's been talking about.

MMLU (Massive Multitask Language Understanding)

First up, we've got the MMLU test. This bad boy is designed to put an AI model through the wringer, testing its understanding across a wide range of subjects you'd typically study in undergrad. We're talking everything from the artsy stuff like literature and philosophy to the hardcore sciences like physics and math. The goal? To see if the AI can match the breadth and depth of knowledge of your average college student.

GPQA (Graduate Program Questions Assessment) and Diamond

Next on the list, we've got the GPQA and Diamond tests. These are like the MMLU's older, wiser siblings. They're all about testing an AI's ability to reason at the level you'd expect from someone in a graduate program. We're not just talking about regurgitating facts here; these tests involve complex questions that require the AI to apply, analyze, and synthesize information like a boss. It's like putting the AI through a virtual Master's degree!

GSM8K (Grade School Math 8K)

But let's not forget about the basics. The GSM8K test is here to make sure our AI friends haven't forgotten their arithmetic. This test throws a bunch of math problems at the AI, ranging from simple addition to the kind of stuff you'd see in middle school pre-algebra. It's like a little refresher course to keep the AI's math skills sharp.

MATH Dataset

Speaking of math, the MATH dataset is the next level up. This one challenges the AI with high school to early college-level mathematics problems. We're talking algebra, calculus, statistics - the whole shebang. The goal here is to see if the AI can not just crunch numbers but really understand the underlying mathematical principles.

MGSM (Multilingual Grade School Math)

But what if we want to test an AI's math skills in different languages? That's where the MGSM test comes in. It's just like the GSM8K but with a linguistic twist. The AI has to solve math problems presented in multiple languages, showcasing its versatility across different tongues.

HUMANEVAL

Now, let's talk about the HUMANEVAL test. This one's for all the coding whizzes out there. HUMANEVAL puts the AI's programming chops to the test, challenging it to write snippets of code that perform specific tasks. It's like a job interview for AI, seeing if it can walk the walk when it comes to coding.

DROP (Discrete Reasoning Over the content of Paragraphs) and F1 SCORE

Next up, we've got the DROP test and the F1 SCORE. These are all about testing an AI's ability to reason over text. The DROP test focuses on discrete reasoning, like calculating dates or interpreting numerical data embedded in sentences. The F1 SCORE is used to evaluate how well the AI does on tasks like DROP, balancing precision and recall. Think of it as a reading comprehension test on steroids.

BIG-BENCH-HARD

If you really want to push an AI to its limits, you bring out the BIG-BENCH-HARD. This is a collection of tasks designed to test the boundaries of what AI can do, with challenges ranging from easy to "good luck with that." It's the ultimate benchmark for seeing where AI excels and where it needs to hit the books.

HELLASWAG

Last but not least, we've got the HELLASWAG test. This one's all about common sense and general knowledge. The AI has to complete sentences or predict what comes next in a scenario, showing off its understanding of everyday concepts and logical sequences. It's like testing if the AI could survive a dinner party conversation without embarrassing itself.

Now when comparing Gemini vs Claude vs ChatGPT, the following was found:

1. The Apple Test

In this classic reasoning test, Claude 3 Opus managed to hold its own and answer correctly, but only after it was given little pep talk with a system prompt. GPT-4 and Gemini 1.5 Pro? They aced it without breaking a sweat.

2. Calculate the Time

This one's a tricky test designed to separate the smart from the... well, not so smart. And unfortunately, Claude 3 Opus fell into the latter category, along with Gemini 1.5 Pro. GPT-4 had a bit of an identity crisis, sometimes getting it right, sometimes not.

3. Evaluate the Weight

When asked to compare the weight of a kilo of feathers and a pound of steel, Claude 3 Opus got it wrong, while GPT-4 and Gemini 1.5 Pro nailed it.

4. Solve a Maths Problem

The problem: If x and y are the tens digit and the units digit, respectively, of the product 725,278 * 67,066, what is the value of x + y. Can you explain the easiest solution without calculating the whole number?

Despite boasting an impressive 60.1% score on the MATH benchmark, Claude 3 Opus struggled with our math problem, giving wrong answers no matter how it prompted. GPT-4 and Gemini 1.5 Pro, on the other hand, solved it with ease.

5. Follow User Instructions

Now, this is where Claude 3 Opus really shines! When it comes to following user instructions to the letter, the Opus model outperforms all the others. It generated 10 perfect sentences ending with "apple", while GPT-4 managed 9, and Gemini 1.5 Pro barely squeezed out 3. If you need an AI that can follow orders like a champ, Claude 3 Opus is your guy!

6. Needle In a Haystack (NIAH) Test

Despite Anthropic's claims of Claude 3 Opus' prowess in handling long-context data, it couldn't find the needle in our 8K token haystack. GPT-4 and Gemini 1.5 Pro, however, located it with ease. More testing is needed, but so far, it's not looking great for the Opus model.

7. Guess the Movie (Vision Test)

In the image analysis test, Claude 3 Opus and GPT-4 both correctly guessed "Breakfast at Tiffany's", while Gemini 1.5 Pro missed the mark. Kudos to Anthropic for creating a model with solid image processing skills!

The Verdict

While Claude 3 Opus has its strengths (like following instructions and image analysis), it falls short in commonsense reasoning, math, and long-context data compared to GPT-4 and Gemini 1.5 Pro. However, there are specialized areas where it truly excels, such as rare language translation, quantum physics, and learning self-types annotation.

WHY DOES CLAUDE MATTER TO GTM PROS?

For sales, enablement, marketing, and customer success leaders, this new model offers exciting opportunities. Its ability to follow user instructions with precision can streamline content creation and personalization efforts.

Marketing teams can leverage its image analysis capabilities to create engaging visuals and enhance brand messaging.

Sales and customer success can utilize its language translation and specialized knowledge to better serve clients across the globe.

While Claude 3 Opus may not be the ultimate solution for every task, understanding its strengths and weaknesses can help GTM leaders make informed decisions and stay ahead of the curve in the rapidly evolving world of AI.

Now how do we pick between all 3? Let's summarize and break each one down:

Claude 3.0:

1. Instruction Following: Excels at precisely following user instructionsIdeal for content creation and personalization tasksHelps streamline workflows and maintain consistency
2. Image Analysis: Strong image processing capabilities. Useful for creating engaging visuals and enhancing brand messaging. Can assist in developing compelling marketing materials
3. Specialized Knowledge: Proficient in niche areas like rare language translation and quantum physics. Valuable for serving clients with specific needs or in specialized industriesCan provide expert-level assistance in certain domains

ChatGPT 4.0 with GPT Store:

1. Commonsense Reasoning: Outperforms Claude 3.0 in commonsense reasoning tasks. Helps in creating logical and coherent content. Useful for developing persuasive sales and marketing materials
2. Long-Context Data Handling: Efficiently processes long-context data. Valuable for analyzing extensive customer feedback or market research. Can uncover insights and trends from large datasets
3. GPT Store Integration: Access to a wide range of pre-trained models and plugins. Enables customization and extension of ChatGPT 4.0's capabilitiesAllows for seamless integration with existing tools and workflows

Gemini 1.5 Advanced:

1. Strong Overall Performance: Consistently performs well across various tasks and benchmarks. Offers a reliable and versatile solution for GTM professionalsCan handle a wide range of sales, marketing, and customer success tasks
2. Commonsense Reasoning and Math Skills: Excels in commonsense reasoning and math problems. Useful for data analysis and creating data-driven content. Can assist in making informed decisions based on quantitative insights
3. Long-Context Data Handling: Capable of efficiently processing long-context data, similar to ChatGPT 4.0 Valuable for analyzing extensive customer interactions or market trends. Can help uncover opportunities and optimize GTM strategies

How do you even know where to start with AI?

I hear this question all the time and this last week, I spoke to a group of Sales Enablement Collective members about this very thing.

Short version, instead of focusing on the shiny new AI tech, just remember, it is a tool, and tools are meant to help us do what we do or need help with obviously.

To help with this, I am going to lean on a friend of mine Mike Kunkle who talks about a concept called COIN-OP (Challenges, Opportunities, Impacts, Needs, Outcomes, Priorities) in his Modern Sales Foundations methodology. Now usually he is talking about this in reference to a sales discovery process, but it applies as a good change management, initial step for us when trying to identify what to do and where to start with AI.

Questions below should not only identify where AI can be most impactful but also align AI initiatives with the strategic goals of the GTM process. Here's how the COIN-OP model can be tailored for this purpose:

Challenges (C)

• What are the main bottlenecks in our current GTM process?Identify specific stages in the GTM process where you encounter delays, inefficiencies, or lower than expected conversion rates. Understanding these bottlenecks can highlight areas where AI might streamline operations.

Opportunities (O)

• Where can AI technologies create a competitive advantage in our GTM strategy? Look for opportunities where AI can provide insights, automation, or personalization that competitors are not utilizing. This could be through advanced customer segmentation, predictive analytics for lead scoring, AI sales coaching reports, or automating repetitive tasks to free up sales reps for more strategic activities.

Impacts (I)

• How would improving specific GTM stages with AI affect our overall business outcomes? Evaluate the potential impact of addressing identified challenges with AI solutions. Consider both direct impacts, like improved conversion rates or shorter sales cycles, and indirect impacts, such as enhanced customer satisfaction or increased upsell opportunities.

Needs (N)

• What data and infrastructure changes are necessary to support AI integration into our GTM processes? Determine the data, tools, and platform needs to implement AI solutions effectively. This includes assessing data quality, availability, and the need for data integration platforms, as well as evaluating whether your current tech stack is AI-ready.

Outcomes (O)

• What specific, measurable outcomes do we aim to achieve by integrating AI into our GTM processes? Define clear, quantifiable goals for your AI initiatives, such as reducing lead response time by 50%, increasing lead conversion rates by X%, or enhancing customer engagement scores by Y points. These outcomes should directly contribute to the overall business objectives.

Priorities (P)

• Which AI initiatives should we prioritize based on potential impact and feasibility? Prioritize AI projects based on a combination of their expected impact on the GTM process and the feasibility of implementation. This involves considering factors like cost, available resources, technical complexity, and alignment with strategic goals.

Buying Process

• How can AI influence our understanding and optimization of the customer buying process? Explore how AI can enhance insights into customer behavior, preferences, and decision-making processes. This could involve using AI to analyze customer data for patterns that predict buying behavior, personalizing customer interactions based on these insights, or automating engagement at key decision points in the buying journey.
• What AI-driven improvements are needed at each stage of the buying process to meet the exit criteria of different personas? For each stage of the customer journey, identify how AI can help meet or exceed the exit criteria that prospects need to move forward. This may involve automating personalized content delivery, offering real-time support via AI chatbots, or providing AI-driven product recommendations.

There are so many ways AI could be applied and used inside of a company, but in my opinion, unless a true GAP analysis has been done and after you have done so, put into place a change management plan. We will be diving more into this next week.

I hope you enjoyed this weeks podcast and newsletter, let me know if there is anything else you are missing or wanting to know about!