Salesforce introduces artificial intelligence assistant for all CRM applications 

Organizations using Salesforce can also use the new Einstein Copilot Studio to customize how Einstein Copilot works, how to access and reference proprietary data, and leverage third-party large language models (LLMs) such as OpenAI GPT-3.5.

Salesforce got in on the enterprise AI race early when it launched Einstein in 2016 with the goal of helping with customer outreach, search, segmentation and product recommendations.  In May this year, they launched a conversational CRM tool, Einstein GPT, and now they've gone a step further with Einstein Copilot, which offers an artificial intelligence function, also known as an AI agent,  which can perform a range of CRM and application-specific tasks autonomously.

- said Ágnes File, head of Deloitte Digital's Salesforce business line.

High degree of configurability

The biggest advantage of the solution is that each organization or business can fully customize it to its own needs.  Here are just a few of the domain-specific tasks that the new solution can handle:  

• Sales: automatic updates of customer data in Salesforce, preparation of meetings, automatic generation of sales emails tailored to the customer.
• Service: streamline customer service by providing personalized, relevant responses to customers through different channels: texts, email, social media, live chat.  In addition, it suggests relevant answers to customer service agents during their workflow, using real-time customer data.
• Marketing: generating email texts, and creating landing pages based on customer preferences.
• Commerce: helps you create digital stores and automate complex tasks such as catalog management.
• Development: converts natural language requests into Apex code and suggests coding improvements.  
   

Einstein Copilot Studio also provides the configurability to make Einstein Copilot available on consumer channels such as real-time chat on websites, Slack, WhatsApp or text messages. 

Help with good prompt design

For users, the most important element of generative AI is the so-called prompt, or input channel, if you like, the interface where the instruction is written.  It is often found that generative models require multiple retries, multiple variations of the prompt, to yield exactly the desired answer.  In many cases, it is the insufficiently precise prompts that lead to incorrect answers and misleading output.  

Einstein Copilot Studio includes a "Prompt Builder" that allows non-technical users to simply describe what they want to achieve, and the Prompt Builder takes this ordinary wording and creates the correct prompt to produce the desired result. 

For example, a marketing professional could ask the Prompt Builder to generate a personalized message and discount for a new product based on a customer's purchase history and location. Einstein Copilot then automatically generates personalized messages that are tailored to the customer's individual
preferences based on previous purchases and demographic information.

- said Ágnes File.

Safety first

Much of the solution's customizability depends on the data that companies using Salesforce provide to Einstein Copilot.  This data is stored securely in the Salesforce Sales Cloud, does not travel over the public internet, and third-party LLMs have "zero retention", storing no data after giving the response. 

Behind Einstein GPT and Einstein Copilot, an additional secure AI architecture layer has been added. The Salesforce Einstein Trust Layer runs continuously and scans all AI responses and warns of inappropriate responses, such as those generated by biased AI.  The layer also captures all AI interactions for record keeping, compliance and auditing.

Demand for AI is growing dynamically across all sectors, with 45% of executives saying they are increasing their investment in the technology, according to a Gartner survey.

We see a lot of value in implementing Salesforce's new solution, for example in
the area of achieving greater customer engagement. Although Einstein Copilot is still in its early stages, it seems ready for widespread adoption, as evidenced by the fact that it is already used successfully by organizations such as Heathrow Airport.

- said Tamás Schenk, partner at Deloitte
Hungary's technology consulting business.

If you are interested in more details, Deloitte Digital's online event on AI and Salesforce will answer your questions on the topic: https://www2.deloitte.com/hu/hu/events/esemenyek/2023/deloitte-digital-dreamforce-2023-webinar.html