At Dreamforce 2023, Enterprise Times spoke with Bill Patterson after he had been on stage during the Service Cloud keynote. During the service Cloud keynote, as with other keynotes, the emphasis was on Einstein One and demonstrating what the new AI features would deliver to customers.
Is the human still in the loop?
The example shown during the keynotes often seemed to take the human out of the loop, with complete automation. Is this the direction that Salesforce is going?
“I don’t think we’re taking the human out of the loop at all. We are bringing humans into places where humans are needed to do their best work. The reality today in customer service is that humans still do 40% of work that is mundane, manual tasks. I know this, I’ve been a customer service agent myself. When you have this sort of work that’s on auto drive, it actually doesn’t allow humanity to offer creativity or empathy or compassion, in ways that I think are lacking in our domain and discipline.
“Our goal is threefold. We want to utilise AI to not deflect but to triage a lot of the customer service cases that are routine or more mundane in nature so that we can create enough capacity for humans to do higher-quality interactions. When we sit and think about humans in the loop, we think it is a human loop. it really is an opportunity for you to spend more time, personalising, and tailoring services.
“It varies by industry, really. When you think about the story of AAA, a lot of their routine interactions, like address changes, can be offloaded. Because when you’re actually having an accident, I need to make sure you’re okay and how to best help you. What we really want to do is help service teams keep it in balance. That’s how we expressed it yesterday.”
It sounds when you talk about triage, are you using AI and where you’ve got a new situation, would that break out the human.
“Yeah, encountering a first-time incident, we don’t have the data, to really tell you, how this is going to pass through?”
Generative AI is often called out as helping people learn. However, these are often textual, and people have diverse learning styles. How can organisations and generative AI overcome that challenge?
“I’m not a read-it learner, I’m a visual learner. For myself, that would not work. I think it can assist, but it doesn’t solve the task. What we’re doing with our technologies, like enablement technologies, which is more about video, visual, pictorial or gamification, as we do at Trailhead, that’s the more all-encompassing pathway to on-ramp to train representatives.
“Some of those moments of recommendation are just there for assistance. It’s not there to complete the task. You’ll find a little bit more around the direction of adherence, maybe around the direction of compliance, more than you’ll find it on enablement or training.”
To summarise, enterprises need to be wary just to relying on generative AI to train their staff.
“Yeah, it’s a tool. I have lots of tools in my shed. That doesn’t mean I’m an artisan with them.”
Where next for generative AI?
“Today in this progression of AI, predictive was the first wave, generative is the current wave that we’re in. I do think autonomous use cases are going to become routine. We’re a long way off AI for artificial general intelligence. I hope to see it in my lifetime, but I’m not sure it’s really gonna become mainstream like that.
“These autonomous use cases will take shape. Even today. I’m starting to see this enter into the consumer world already. I was using a restaurant booking service app. I needed to make a change and the app asked me, “we can’t make this change, but we use an AI engine, would you like us to call and confirm that, and we’ll update you when it happens?” And I was like, “Heck, yeah”. Because I didn’t want to do that, so I think these autonomous functions or tasks are probably the very next phase that we’ll get to. Then some of those tasks may start to complete more functions.
“I do have a vision for customer service. I started my career in customer services. I was the young guy in the contact centre, I got to answer things like social responses or emails. I hated that, I just felt like I was answering the same thing all the time. If you can automate a lot of those routine interactions, using an autonomous agent, a lot of organisations would love that. It frees you up for more human-to-human kind of connection.”
The attitude to this automation varies by age group and by country, the younger generation seems more accepting. What is your view?
“Absolutely, it does. When you say the younger generation, you have to segment that. Its a problem that keeps getting different as we age. There are a lot of studies actually that show they want to reject technology, they actually want to be more face to face. And so I think we’ll just see these societal shifts as certain things become mainstream.
“You find counterculture, enters into the mould, which is, I’m going to choose to do business with those that handwrite letters to me. I think we’ve seen this constant cycle evolve. And those leaders that you know, really stand out in service today, they offer great frictionless experiences, so that they can create more human connections with you.”
What can Einstein One deliver?
Which Salesforce cloud has the most potential for Einstein One?
“It’s a really hard question to answer from most potential, I think they all have a lot of potential. In sales use cases, I still think sales teams today spend a lot of their time in the data entry world, and not really in the connection world. What I love about Einstein One is some of the work that we’re doing around meeting wrap-ups and on-call interaction, is very powerful.
“Today, as the world has moved to a lot of digital interaction, especially on video, some of the work that I’m really excited Einstein One and the Co-Pilot experiences will help us with. It’s not just capturing what is said, but the way in which it’s said. The way in which things are expressed and the pauses in the meeting. If I throw an offer out to you, and it gets no pickup, well, that probably has some signal of loss of value there.
“I really am excited about using AI to piece together how to make a sales interaction better. And that’s one of the areas I’m excited about in the sales demand.
“In Service, taking a lot of the volume and load off of service teams is probably the biggest area of opportunity. When you’re on a call, it still takes 20% longer to close the case today, because what gets through (to them) is more complicated. I want to use Einstein One and Co-Pilot to truly enrich that call so I can not just solve an issue, but make sure that there are questions you didn’t ask me that can offer some value to you proactively.”
On the ecosystem support
How are you supporting ISPs going forward? Many I have spoken to seem unsure about where they go next to build in Gen AI, they have been bitten before around licensing on Wave.
“We are, I think, becoming more mature as a platform organisation. It used to be that the ecosystem was built, where this is AppExchange, and you play by the rules of the AppExchange. A lot of businesses curated and conformed to just that one way of working.
“What you find now is our platform is becoming more open. The ecosystem is giving a lot of opportunities and choices. When we talked about the Einstein One Co-Pilot and Co-Pilot studio, it’s an open ecosystem of LLMs that you can utilise. Not just the most popular ones with Open AI, Anthropic or Cohere. As an ISV, you can build your own LLM for industry or domain-specific tasks. This is a moment for the ISVs to understand that our strategy for our ecosystem, while it was just about AppExchange in the past, now there are many more choices and options, and is welcoming more people into the ecosystem, which I think is very powerful.
“Does mean that certain partners have to evolve, they have to think about how do I modernise my relationship with Salesforce to take advantage of these opportunities. That’s part of our mission here, is to make them aware of it and help them achieve it.”
Are LLMs the future?
Everyone talks about very large language models, but there is a growing interest in small LLMs, where do you see things going?
“It’s evolved. Eight months ago, it was all about these giant models. Now what we’re learning and seeing through performance is small models actually outperform the large language models because they become so task, domain or industry-specific in the value that they offer.
“That’s why I believe what we’ve done with Co-Pilot, and the Co-Pilot studio, to allow our partners to build models that tie to specific tasks, you saw this in the service keynote yesterday. When we were creating our bot that had a case task or a response task, you can actually insert a task that, might be a creditworthiness task that has an LLM built alongside of it. Your small LLM that becomes task specific gets utilised in those kinds of areas, I think we will see that the world will actually become very small around the LLMs.”
So specific small LLMs, with fewer hallucinations, one assumes.
“It’s not just hallucinations that are an artefact of today. What we want to see is tasks that become so great at what they do, that you can trust that, that gives you the authentic and qualified response.”
The Opportunity of AI
You said recently about generative AI, the most profound opportunity of our lifetime around enterprise software. What is the opportunity?
“If you imagine what the world of service experiences looks like, for any product that you represent today. It is a help centre page. It is a search for your own answer. If you can’t find the answer, it’s a talk to a bot. If you can’t get to the bot, it’s then call customer service. I really believe that, that is a miserable experience.
“As a customer, it’s so frustrating to be offered these choices where you think you may have an answer, but the actual result is – I had to call you anyway. That’s very frustrating to me as a consumer, it’s frustrating to me as an innovator. One of the things that I’ve really been taking a lot of time working with companies on is, how can we make this better?
“Forget the canvas of what we have today. How do we make this better? I was with the Disney company yesterday; they want to create a truly magical experience every time you come to one of their properties. If you’re putting the search query in any way, just give you the answer. That’s the opportunity.
“When I say profound opportunity, it’s about really rethinking all of the world of technology just to get to better outcomes for our customers, and I’m really excited about that. it’s good for your end customers and is also good for companies. When you look at Enterprise Architecture today, it’s a mess for all these different systems, platforms, and capabilities but don’t talk to those things that work over here. We can advance this to a whole new future that’s more integrated.”