Table of Contents
AI is one of the innovative ingredients in the secret sauce that makes second nature awesome. Second Nature is a unique and highly innovative conversational sales training software that makes sales training soar. We take the burden off sales managers, provide on-demand training round the clock for sellers through AI-simulated conversations, deliver targeted feedback to help sales reps improve, and enable realistic conversations that empower salespeople to grow in skills and confidence.
At the center of the platform is a dialogue system that simulates a buyer in a buyer- seller roleplay. Here’s a short example:
One of the many tools used to conduct such simulated conversations is generative AI.
Generative AI is a new name for a collection of artificial intelligence tools that generate new content, rather than categorizing information or predicting outcomes. It’s a step ahead of the use cases that people are most familiar with, and moves artificial intelligence from the era of interpretation to the era of generation.
If you have used ChatGPT to produce original text, you’ve already interacted with generative AI. DALL-E is another generative AI model that operates in a similar way to ChatGPT, but it produces high-quality images instead of text.
We’ve been using generative AI for two years to help power the sales training conversations that Jenny, our AI role play partner, holds with users. We started using Open AI’s GPT3 when it was in closed beta, and since then we have been optimizing it specifically for the sales conversations our customers need. Read on to learn more about how generative AI makes Second Nature a sales training powerhouse.
Many of our product features are powered by AI in various forms. We use AI to create a realistic-feeling conversation that users feel comfortable with, to evaluate our users’ talking style, and to produce timely scores.
A typical dialogue system may match the user’s current sentence to a node in a predefined dialogue tree that was produced beforehand in the lab. This may be useful in closed, target oriented conversations like those done with Siri or Alexa where your goal is clear, but free form conversations like those done by sales professionals don’t follow such a scripted path, so we had to build other tools to support those use cases.
Over the past several years, we’ve been increasingly using generative AI to produce the responses given by the AI role play partner to the salesperson trainee, delivering a richer and more fulfilling user experience. For example, with generative AI, the sales training role play partner can answer any question, even if it was not thought of in the lab ahead of time.
The GPT3 family of models was pretrained on enormous datasets, giving them a deep understanding of how real people write and talk so they can generate texts that sound authentic to human ears. At Second Nature we invest a lot of our R&D in training them more specifically for sales situations, however they do start out with the basics of generic language from GPT3, so they already know how to “talk.”
In regular conversations, we don’t usually notice who is taking the initiative – that is, leading the conversation – and who is responding. But it makes all the difference in a simulated conversation – the burden on a digital system to answer a complicated question is higher than what it takes to ask it.
This difference maps further into the semantics of sales conversations – in a “discovery” conversation the salesperson typically asks the prospect to answer questions, while in a “pitching” conversation the salesperson is the one who is answering the prospect’s questions.
Building a dialogue system to answer these needs is challenging, and generative AI proves to be a great tool to solve this, especially when the initiative is at the buyer’s side.
Although generative AI has impressive capabilities, like all other tech it needs to be integrated in the right way.
Prompt engineering: a generative AI model needs a prompt in order to generate useful content. A prompt is a textual context that guides the model to the right topic. The generated content on the other hand is non-deterministic, that is the same prompt may generate different results at different times. This attribute is problematic when one wants the model to help achieve a specific learning objective. For example, I may want to train a salesperson to speak with a new prospect who knows nothing about the product I’m selling, but the model may generate content where the prospect admits he already uses my product for several months. “Taming” the model to produce good results through prompt engineering is a challenging task.
Answer latency: Top quality generative AI algorithms can take time to produce a response, sometimes up to a few seconds of delay. But such a lag is noticeable in a voice conversation and drags down the experience. Incorporating generative AI in a voice-based dialogue system therefore requires novel ideas around the models’ usage.
Content filtering: A generative AI model can produce content which is relevant, but not necessarily appropriate, especially in an enterprise environment. Often, extra training needs to be carried out for the model and filters need to be applied to prevent offensive or racist content from getting through.
Generative AI is a paradigm shift for dialogue systems, and Second Nature utilizes its potential in the unique space of audio conversations. Today’s implementations are mostly text based, but we’re already working on some projects using video and audio cues to enhance the outcomes of our generative AI models. When this happens, our system could ask a user why they seem down today, based solely on their body language.
Even though we’ve been using generative artificial intelligence for sales training for the past two years, we’re still at the beginning of the journey. There are new techniques, models and capabilities being released on a monthly basis. We’re thrilled to be part of such a dynamic environment, sharing our experience in the space and learning from others. Every new innovation helps us give salespeople a better experience with our product and enhances their practice of sales conversations.
To learn more about Second Nature, request a demo.
We have positions available in our Tel-Aviv and New York offices and remote/hybrid.
With over 50K happy trainees