In this week’s No Jitter Midroll (NJM), we focus on product news from Salesforce: Einstein SDR Agent and Einstein Sales Coach Agent. In a prebriefing call, Ketan Karkhanis, EVP & GM, Sales Cloud said that pricing for the new Agents will be announced at Dreamforce (in early September 2024) and that Salesforce intends to “make the agents very accessible and available” to even the smallest SMBs and mid-market customers. Organizations must have Sales Cloud to use either Agent. Both Agents will be generally available in October 2024.
Einstein Sales Development Rep (SDR) Agent uses generative AI to have personalized conversations with inbound prospects; Einstein SDR Agent can answer questions, handle objections, qualify leads, and book meetings. Karkhanis stressed that unlike traditional chatbots that must be programmed to answer specific questions with canned responses, Einstein acts more like a human sales assistant might.
The conversations Einstein has with those inbound leads are grounded (via Salesforce’s retrieval augmented generation (RAG) service) on a company’s business data, including Salesforce CRM data. Using Data Cloud, companies can upload existing sales materials like product FAQs, sales plays, and case studies for Einstein to use when interacting with leads.
Sellers can customize the language and style Einstein SDR agent uses with prospects, as well as set guardrails for how often, what times of day and on what topics Einstein is allowed to engage. Additionally, the human ‘manager’ of the AI agent can customize which leads they deal with first versus those that are routed to Einstein – for example, the human manager can tell the Agent that if a lead is over a certain dollar value, the prospect goes directly to human interaction while any leads below a certain dollar value initially interact with Einstein.
When prospects in the sales pipeline decide they want to schedule a meeting, Einstein handles the work around the scheduling. And then, the Agent will notify the human rep and brief them with curated meeting preparation that summarizes lead information and previous interactions.
“Einstein Sales Agent, makes every AI conversation a ROI conversation, because you can directly connect AI to additional qualified leads and correlate AI usage to the win rate of the sales reps,” Karkhanis said.
Currently, Einstein SDR primarily interacts via email. It works on mobile devices and is available in multiple languages. “Over time, it is definitely our intention that it will engage with buyers on WhatsApp, too,” Karkhanis said. This social platform is particularly important for non-U.S. markets.
This AI agent will coach human sellers by facilitating generative AI-powered, text-to-speech role-plays, tailored to each deal. During the role-play, Einstein simulates a buyer by grounding its responses (again via RAG) on information about the deal, account, and previous correspondence with that customer, as well as any external files to which it is granted access.
During the role-play, Einstein generates contextual responses to what the seller is saying in the buyer’s tone. Einstein uses Salesforce’s speech-to-text service to speak with the human seller. That service uses OpenAI’s Whisper models. Salesforce said that the Agent uses a fabricated voice when interacting with the seller during the role-play; it does not use an audio sample of the given buyer.
After the coaching session (which can be repeated ad infinitum), Einstein provides personalized feedback to the seller based on the company’s selling strategy, techniques, best practices, qualification checklists, etc. During the live sales meeting, Einstein helps during the call (the experience is similar to contact center implementations of ‘agent assist’ solutions).
Karkhanis noted that both new AI agents are built on the Einstein 1 Agentforce Platform (details on that platform are forthcoming at Dreamforce 2024). Both of the new sales agents can be configured using no-code actions, workflows, and pre-built templates.
By using Salesforce’s Data Cloud, customers can further enhance generative models by uploading relevant external information such as existing selling and training documents. Both Agents leverage the Einstein Trust Layer to safeguard that data and constrain their output. Karkhanis said that Salesforce has an open ecosystem model with respect to large language models (LLMs) and that it works with all of the major foundation model providers (Open AI, Anthropic, etc.), and that it has been working on its own family of multimodal LLMs called XGen, which was recently open-sourced. The Salesforce AI Research team published a paper on their models here.
Check out these other NJ articles about Salesforce products and news:
