# Jag Jag (Senior Solutions Engineer, Enterprise, Salesforce (data and AI specialist)) was a guest on AI with Sach, a podcast hosted by Sacha Gucciardo in Montréal. ## Episode 3: How Salesforce is Using AI to Transform Customer Relationships (July 2024) A Salesforce solutions engineer explains how AI is added to CRM, why trust and humans in the loop matter, and why he expects a return to human connection. Page: https://aiwithsach.com/episodes/03-how-salesforce-is-using-ai-to-transform-customer-r/ > if we go back to circa 2000 when Napster came out, right? And everybody was like, this is the next big thing. We're just not going to pay for music anymore, right? And what happened out of that? We had unlimited, abundant access to music. But it also made us less excited and less happy. Every time we heard a new song, we valued each song less. > — at 4:08 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=248s) > So we were able to unlock access to all that wealth of information in those previous RFIs using RAG because we're feeding that data back into the vector database so that when a new RFI comes in, it automatically looks at the questions in this new RFI and finds the closest match to the answers from previous RFIs. So something that used to take 175 days to fill out now takes less than 30. > — at 11:55 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=715s) > the biggest winner out of this whole AI boom, has been technology that's decades old, which is like machine learning. So customers that weren't taking machine learning seriously a few years ago are using AI as a scapegoat to do something that's been around for a long time. > — at 18:37 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=1117s) > I would love to see like an AI email spam filter so that we go back to just drafting real emails. And this is kind of odd as me saying as a Salesforce solution engineer, but I think it will help the people that desire human connection stand out versus the ones that may have taken a shortcut. > — at 20:15 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=1215s) > I think we're jumping to solutions before listening. And I find myself doing this when a customer is asking me a question. I got ChatGPT open and I'm already trying to understand the answer to it rather than trying to understand the question in much more depth than I used to. > — at 20:48 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=1248s) > think about which parts of your life you want automated in AI in two years, not just what you want automated tomorrow. And think about the psychological impact every time you want to use AI versus just intuition or using human interaction. And desire to be perfectly imperfect rather than be perfect. > — at 22:26 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=1346s) > Why? Because at the end of the day, anything that gets too popular too quick hits diminishing returns. And we go back to something that's more natural to us. > — at 4:38 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=278s) > we built out our own trust layer, as we call it, and essentially a firewall between any LLM and Salesforce. And it has a couple of main tasks. One is to make sure that the LLM does not get trained on customer data. So that retention stays on the Salesforce side boundary. > — at 13:33 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=813s) > Because we're dealing with B2B customers. It's not like a Spotify recommendation. This is really serious shit, right? Like, am I recommending a promotion to a customer that probably isn't entitled to one? Could that be bleeding dollars out of the business? > — at 14:42 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=882s) > there was a few articles about is AI, does AI mean the death of structured relational databases? I don't think so. I think we desire structure and we desire the visibility to the structure. > — at 17:05 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=1025s) > I think in the healthcare and life sciences industry is going to be disrupted hardcore for good reason. Like a radiologist may be able to use AI to detect cancer much, much earlier than they used to before. They'll still be a human in the loop, but they'll be able to get some assistance from AI. > — at 19:27 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=1167s) > quantifying human behavior can mean like, if you are a musician, how do you read the crowd? That's quantifying human behavior, which is using your natural senses rather than using an algorithm. So essentially, my goal was to kind of learn how to build an API into someone's mind. > — at 3:09 (https://www.youtube.com/watch?v=8-UnJLq6NYA&t=189s)