Brainstorm Tech 2024: Salesforce CEO Clara Shai on building trust in AI and advancing CRM solutions
Brainstorm Tech 2024: Salesforce CEO Clara Shih on building trust in AI and advancing CRM solutions
Category
🤖
TechTranscript
00:00So, Clara, I want to start out by talking a little bit about trust.
00:05Trust has become a really big issue for companies that are hoping to deploy particularly generative
00:10AI solutions.
00:12A lot of companies are complaining that they're finding the technology very unreliable.
00:16They're getting frustrated because of that.
00:18They're afraid to deploy it at scale.
00:21And earlier when we were speaking, you said that you felt a lot of AI solutions have yet
00:25to kind of find product market fit.
00:27I wonder if you could expand a bit on that and also tell us how Salesforce is trying
00:31to overcome this problem in its own AI solutions.
00:33Sure.
00:34First of all, thank you so much for having me.
00:36It's great to be back at Fortune Brainstorm Tech.
00:40So trust is a big topic.
00:42I'm really proud to share that it's the number one value of Salesforce and has been since
00:46the company started 25 years ago.
00:49And there's trust in the sense of what you describe in terms of, you know, AI hallucinations
00:53and toxic outputs.
00:56And that's why the first thing that we built when we were creating our generative AI products
01:00was our Einstein trust layer that has data security, data privacy, ethical guardrails,
01:06everything from data masking and citations and an audit trail to our toxicity filters,
01:13et cetera.
01:14But there's this broader question of do I trust AI to drive business outcomes, even
01:19if I can mitigate the data security, data privacy types of questions?
01:23And I think that's where I've seen a lot of companies and customers struggle.
01:27And so they come to us and they want to know how they can really deploy these solutions
01:32in a way that actually moves the needle for their business.
01:35Because, you know, these efforts, they're expensive.
01:38And so I think, you know, where Salesforce is, we're really well positioned to do this,
01:43is our customers, you know, many of you in the audience, already have their trusted data
01:48and their trusted business processes that they've built into Salesforce over the years.
01:54And so it provides this ideal grounding for the AI to really be, to inject in the context
02:02that's needed for the models really to be able to perform well, whether it's using it
02:07for simple retrieval, augment and generation question answering, or as we move into some
02:11of these more agentic use cases.
02:13Interesting.
02:14And I want to go to questions for the audience, sort of try to make this session interactive.
02:18So think of your questions, and if you put your hand up, I'll try to come to you.
02:22But I'm first going to ask Clara, we spoke a little bit about how sometimes like narrowing
02:26the use case for the technology is a way that you can actually build up more trust in what
02:31the technology can do.
02:33And the fact that, you know, for Salesforce, if you're using Einstein for a particular
02:36sales case, you don't have to worry about the fact that it's, you know, it's not going
02:39to plan your kid's birthday party, it's not going to write a letter.
02:41But can you talk a little bit more about that?
02:43And what we sort of meant by the idea that if you can pick a more narrow kind of vertical
02:47within your business to deploy the technology into and kind of constrain it to that, that
02:53can help with trust.
02:54Yeah, I mean, trust is contextual, right?
02:56There's a question of do I trust the AI to help me figure out which customers to call
03:02on this quarter to hit my number?
03:04Do I trust the AI to help me answer customers' questions through an Einstein service agent
03:10so that I can deflect the easier cases and keep my contact center reps for the more complex
03:16cases?
03:18Those are very specific, outcome-oriented trust questions.
03:21I think it's easier, frankly, than creating an AI that can do everything, absolutely everything.
03:28And I know there's companies doing that.
03:29I commend their efforts.
03:31But when I talk to businesses, what they want are those specific outcomes, right?
03:35How do I really get all of my sellers performing like my very top 5% of President's Club sellers?
03:42How do I get all of my marketing campaigns to perform like the best ever marketing campaign
03:46that I did?
03:47And with that context and with the examples in the past, right, because Salesforce isn't
03:53just a database of customer data, it's a database of customer outcomes, so we know what worked
03:59in the past, what didn't.
04:00And you can use those examples for both few-shot examples, you can use it to fine-tune models,
04:07and that's really how we're able to unlock these tremendous outcomes that we're seeing
04:11at everyone from AAA, from insurance, to ADP payroll, biggest payroll processor in
04:18the world, to Heathrow Airport, to OpenTable, across a wide variety of industries, geographies,
04:25and customer segments.
04:26Interesting.
04:27And at Salesforce, I know you've been eating your own dog food.
04:30You've been deploying these AI solutions internally, as well as selling them to customers.
04:34What's the experience been like internally?
04:37It's been, you know, we like to drink our own martinis, as I like to say, but it's bumpy
04:41at times.
04:42I really like that accountability of us being customer zero, because that's how we are able
04:47to sell with confidence, because we have deployed it internally.
04:50So we use everything Einstein internally, from deploying self-service search answers,
04:55so when any of our employees, I had a question the other day about parental leave benefits
05:00for someone I'm trying to hire, rather than logging a ticket, I just went to our employee
05:05portal, I typed in my natural language question, it was automatically grounded in who I was,
05:12which, you know, I'm in the state of California, so is my employee, and it gave me the answer.
05:16All of that powered by the Einstein trust layer, and our entire Einstein platform, all
05:21the way to Slack AI, right, we, you know, a bunch of us took some time off over the
05:264th of July, we got back to Slack, and we had to catch up on all these channels, conversations,
05:32different deals we're working on, Slack AI summarizes it in an instant, right, and so
05:37all in, we've saved over 50,000 person hours in productivity by deploying our own AI solutions
05:44and other AI solutions to our employees.
05:46Of those various use cases, what do you think has been the most impactful?
05:50It depends on the role that someone is in.
05:52I mean, you know, our salespeople, they're using Einstein Copilot to do their account
05:56planning and their reach out to their customers, our customer service team, they're getting
06:01assistance in the moment to resolve customer issues faster, instead of having to look up
06:06every knowledge article and piece it to the customer situation, we're able to use Data
06:10Cloud and Einstein to automate a lot of that.
06:13For me personally, just given the nature of my work, probably Slack AI is what I use every
06:18single day.
06:19Interesting.
06:20I want to get questions, who's got questions for Clara?
06:22Please raise your hand, and I'll come to you.
06:25We've got a question over here, down in front.
06:31You can just stand up.
06:34I'm just curious, because the way you describe how AI is going to be used for customer service
06:38sounds a lot like marketing, and marketing begins to sound a lot more like advertising,
06:43and you describe a lot of things that you see the major consulting firms do, like McKinsey's
06:47and Accenture's, there's some reports that 20% of their revenue is from digital transformation
06:52to AI.
06:53Do you think using AI is going to kind of collapse the roles of those different organizations
06:57and how they serve the Fortune 500?
07:00I do.
07:01I think that AI plus data will allow us to actualize a vision that predates me at Salesforce,
07:08but it's what we call the customer 360, where instead of shipping our org charts and being
07:14limited by what the sales team knows versus what the customer service team knows versus
07:18what the marketing team knows, we're able to really be customer centric, and across
07:23every customer touchpoint, pre-sale, at the top of the funnel, post-sale, actually transacting
07:29the sale, loyalty, all of that becomes much more seamless, and we've seen this with some
07:35of our customers.
07:36I'll just share an example.
07:37We work with a global apparel retailer, and during COVID, like many companies, they had
07:43trouble hiring and retaining enough contact center reps, but they really wanted to, they
07:49weren't able to keep up with the volume of requests that were coming in from customers,
07:53and so the customer hold times started expanding, and so they came to us, started using our
07:58Einstein AI platform, and something amazing happened, which is not only did the resolution
08:04time on customer support issues decrease by 15%, but they actually started seeing the
08:10customer support reps start to play a bigger and broader role.
08:14They started to be able to answer sales questions and marketing questions.
08:18They actually became brand storytellers, and it kind of reminded me of how in The Matrix
08:23and the movie, Keanu Reeves gets downloaded, new skills, and all of a sudden, he knows
08:29Kung Fu.
08:30Like, that's what we witnessed, and when you talk to these customer service reps, a lot
08:34of them feel like they're doing the best work of their careers, because they now feel empowered
08:38by AI.
08:39Other questions for Clara.
08:41I see one back there.
08:45If you can stand up and let us know who you are.
08:47Yes, Sterling Addy with Barclays.
08:50So as AI starts to incorporate more data from more disparate sources, some of those
08:56sources are not necessarily just in, you know, your cloud.
09:01How are you expanding or implementing the trust model to be able to control that as
09:06those different data sources are implemented?
09:08That is such an important question.
09:10I mean, data is at the foundation of all AI, right?
09:14Good AI depends on good data, and something that my colleagues did, again, predates me
09:20rejoining the company, but they built something called the Salesforce Data Cloud a few years
09:25ago, and what the data cloud does is it unifies and harmonizes all of a company's structured
09:32and unstructured data across ERP systems, our CRM, of course, other CRMs, data lakes,
09:41data warehouses, like Snowflake, Databricks, BigQuery, and what we realized was that every
09:47company has multiple of these data silos and applications, many multiple of them, and
09:54within these data silos are data that's actually highly relevant to salespeople and highly
09:59relevant to service people and marketers.
10:01So for example, almost every modern company today has a data lake that logs website and
10:07mobile activity.
10:08So if a customer or prospect goes to your website, views product pages, that gets logged
10:13in a data lake, right?
10:14Probably everybody does that.
10:16Legacy companies are doing that now, but the problem is your salespeople aren't logging
10:21into your Snowflake, right?
10:22They don't even know what Snowflake is or what data lake that company usually has, but
10:27they actually should know, because that's a timely signal for them to prioritize and
10:31reach out to that particular prospect, and so that's what data cloud does, is it unifies
10:36and harmonizes these real-time signals for use in CRM and by Einstein AI.
10:43Other questions?
10:44Down here in front, there's a question.
10:45Let's get a mic.
10:46No, no, wait for the mic.
10:47Thanks so much.
10:48How's it going?
10:49Healy Seifer, CEO of a company called BoomPop, we're an AI travel company.
10:56One of the things you have in the room is a bunch of founders who are trying to build
10:59AI applications, because foundational stuff is kind of like out of reach for us, you know?
11:04And so I've been thinking about it a bunch and argued about it with a bunch of friends.
11:08It feels like there's kind of three things you can do to have a good AI strategy, building
11:12an application, proprietary data, just your data, something that's a learning platform,
11:18or a great UI UX.
11:20How does that resonate with you?
11:21What are we missing?
11:22Like, if you were us sitting here trying to compete on building an application that doesn't
11:25get steamrolled by open AI, like, what do you do?
11:30Great question, especially as a former founder myself.
11:32I think there's so many pockets of possibility to use AI to completely rethink traditional
11:38industries from health care to wealth management to banking, for example.
11:43And I do think that when you think about the traditional model view controller, right?
11:48It's the data model.
11:49And so structured, unstructured data, what's proprietary?
11:53What are the pockets of unstructured data that traditionally haven't been mined and
11:57haven't been used because we didn't have LLMs to be able to help us analyze it and generate
12:02metadata to organize it and understand it?
12:05Definitely.
12:06And then being able to create that learning loop, right, the reinforcement learning, both
12:11from human feedback, expert feedback, as well as from objective outcomes.
12:15As I mentioned earlier, you can only get so far with RLHF, but if you have customer outcomes,
12:21like sales data, conversion data, now you can start to create a really powerful flywheel
12:26over time.
12:27And that needs to be built for every single domain and every single segment.
12:32And that's all greenfield for the types of companies that I would build if I were going
12:35out to start today.
12:37From a UX, UI standpoint, I'm not so sure that that is so defensible.
12:40I mean, certainly amassing a big user base is important so that you can collect more
12:44of that human feedback and expert feedback and outcome feedback.
12:48But I think a lot of UXs going forward are going to be generated themselves.
12:52So the companies themselves will design their own UX or, yeah, you don't think it's like
12:56a niche that you could have an outside startup design a copilot for?
13:01I think that, well, so first of all, a couple of thoughts.
13:03So regarding copilots, I think that that conversational UI that we are familiar with,
13:08that is but one of many UIs that are possible with AI, and we're going to see more coming
13:13just as a teaser.
13:15Slack is a phenomenal place for a team of human workers to have conversations with and
13:20collaborate with a team of AIs.
13:23And you can build applications on top of Slack that take advantage of that.
13:26And then I think secondarily is that AI is able to generate pretty compelling user experiences.
13:33You mentioned sort of more agentic AI coming along.
13:36In terms of Salesforce, when are we going to see a Salesforce agentic system debut?
13:41Thank you for teeing me up.
13:43We weren't going to announce this until Wednesday, but Mark tweeted it.
13:45So we've announced Einstein Service Agent.
13:49It's a continuation of our copilot platform, going from single turn, one chain, to multiple
13:54chain orchestration, and initially for customer service, but we're going to open it up to
14:00all types of CRM applications in the near future.
14:03Do you have a timeline on when it might be available?
14:07It will be available in the coming months.
14:08It is in active pilot right now with a number of customers, including some of the organizations
14:12here, and we're seeing really promising results so far.
14:15Great.
14:17We're about out of time, but thank you very much, Clara, for joining us.
14:19I really appreciate it, and thank you all for listening.
14:20It's been fantastic.