My last Salesforce Connections event was 2018 - and it seems like everything has changed. "Great to be back together again" - sure.
But right now, it feels more like: figure out how to make AI work for you, before it takes away your market share - or your livelihood.
To be fair to Salesforce, the vision/demos of generative AI presented at this show are resolutely about data privacy, trust, and human augmentation, not job replacement - an approach I happen to agree with.
Generative AI - the professional stakes are high
But can we also agree on this? The professional stakes are high. A DJ might be jamming out in the background, but there is a jugular energy here - a serious type of engagement that feels categorically different than the pre-pandemic Connections vibe.
Perhaps the same is true for Salesforce also? Time to articulate a compelling version of what's next, beyond the focus on profitability that quieted activist investors. Flesh out the new operating model that my colleague Stuart Lauchlan laid out in Salesforce COO Brian Millham on how the firm is transforming to rise above the noise. Add more organizational shifts to the mix: on the day that Sarah Franklin delivered the Connections 2023 keynote, Franklin also announced she was moving on as Salesforce CMO, into a role as Chairwoman and President of Salesforce Advisory.
Franklin made the keynote count, doubling down on a memorable line my colleague Stuart Lauchlan also put out in Salesforce Connections - if AI is the new UI, what does that mean for marketing and commerce? I question whether 'AI is the new UI,' but that goes along with a undeniable level of grouchiness about what I view as the technical limitations of generative AI, versus the PR hype balloons that threaten to carry my inbox away. But what do customers think? Check this quip from a customer panelist:
No offense to crypto fans, but this is real. This is not crypto. There are real use cases. This is a priority for us.
We cannot expect customers to be in lockstep on their generative AI plans - especially when you consider how AI's regulatory implications vary by region/industry - but one thing we do know: they all need answers, and pronto. In his last Salesforce Connections piece, Lauchlan laid out some revealing stats on AI and marketing:
- 53% reckon generative AI is a “game-changer” for marketers.
- 71% believe it will eliminate busy work.
- 71% say using generative AI will allow them to focus on more strategic work.
What do customers need from generative AI - and Salesforce?
On the marketing side, the appeal of a deeper level of personalization is a big Salesforce generative AI talking point. David Schmaier, Chief Product Officer, cited a McKinsey stat during a media roundtable:
Seventy percent of the companies they surveyed expect a personalized experience, 76 percent of consumers are unhappy if they don't get that - that's the expectation.
In conjunction with the newly-released 2023 Holiday Guide for Retailers. Salesforce's Rob Garf, Director, Strategy and Insights, Retail and Consumer Goods, revealed a few holiday retail predictions to attendees. Such as: $194 billion in online sales will be influenced by both predictive and generative AI. "That's across commerce, marketing and service," he told us.
As retailers face tough pricing margins and inventory/supply chain pressures, they need to retain those 20% of customers that deliver 150% of the profits. Salesforce argues that AI is the way to pull that off:
By diving deep into your data and using next-generation tools like generative artificial intelligence (AI), you can deliver experiences that show you recognize, understand, and appreciate your loyal customers.
Salesforce customers echoed what I've heard throughout the spring event circuit: customers expect their vendor/platform of choice enable their AI needs. Yes, they may want tooling choices, but the vast majority of customers are simply not equipped to build their own generative AI solutions or large language models. During a GPT 101 for Retail panel, Mandeep Bhatia, VP of Global Digital Product & Omnichannel Innovation at Tapestry, hit this theme:
Most retailers are not technology shops. If the underlying technology makes something like that possible, those platforms are going to win.
This is not something, unless you have mass investment, you can do in-house, so you'll have to rely on these enabling platforms to make it happen. It comes down to how the Salesforces of the world, and other platforms, make this real.
So how will Salesforce enable this? Well, that's what the flurry of GPT branding announcements is about. Salesforce made a slew of GPT-branded annnouncements this year, from Einstein GPT to Tableau GPT, Slack GPT - and yesterday, Commerce GPT and Marketing GPT. I think we can see where this is headed. But some may not realize that Salesforce has already built its own LLM (Large Language Model).
One thing worth noting: Salesforce is not relying solely on its own LLM for this. Customers can choose to access a range of LLMs and AI models, including OpenAI, Anthropic, co:here, AWS SageMaker and, as of June 7, Google Vertex, via an expanded Google partnership (Salesforce, Google Cloud expand collaboration on AI, data).
We've all heard the stories of companies having proprietary data compromised by employees using large language models outside company walls. So do external LLM partnerships challenge Salesforce's commitment to customer data privacy? During our interview, Schmaier pushed back against that, offering this clarification on Salesforce's LLM gateway: customer data is never co-mingled with the bigger generalized training data set of the LLM provider.
We do this in a trusted way. That's the key. Nobody wants to send signals back to the LLM provider from the customer data. That doesn't work; that's why we have multi-tenant SaaS. We've never done that in the history of Salesforce... So we know exactly how to partition the data.
Retail GPT 101 - Salesforce customers weigh in on generative AI
Salesforce customers have had time to assess the impact of AI. Generative AI might be relatively new on the radar, but Salesforce announced Einstein AI in 2015. Perhaps that's why the customers on the Retail GPT 101 panel seemed ready to take on AI, despite being in the early stages when it comes to generative AI rollouts.
When asked about early generative AI use cases, Bhatia told attendees he would be sticking with internal-facing use cases in the early stages, citing concerns on how generative AI's occasional factual missteps could impact external customer relationships (though one of Marketing GPT's keynote use cases was email subject header comparisons/analysis, something Bhatia says they already do via an external partner).
As for early stage use cases, he is looking at generative AI for product maps, content/SEO, and image creation. Creating videos and images is a big expense - can this be addressed with AI? "Will it be on brand; will it be good?" Bhatia acknowledged there is some internal skepticism. "It's a new skill for all of us," he says. Then he added a crucial point:
Do we have all the right data? Yes. Do we have it where we are ready to use it? No. This is step one for us.
But Bhatia's team is going to press ahead - you ignore this at your own peril, he cautioned attendees. Another panelist, Samantha McCandless, Chief Merchandising Officer with luxury re-sale retailer TheRealReal, told attendees about her firm's considerable experience with both personalization through Salesforce, and AI. TheRealReal uploads 20,000 SKUs to their web site each day, and "There's no way to scale that with a human being," says McCandless.
For next steps with generative AI, McCandless is also interested in the SEO aspects - "you can never have enough content," she counsels - content throughout the customer journey, content that is difficult to manage and update. She sees potential in everything from email subject header suggestions/comparisons to speaking into a prompt and generating SQL. High end store associates could get a big information/training boost from generative AI - especially, McCandless thinks, in an "appointment selling" scenario, where you could cater advice to a particular known customer.
But McCandless points out that these scenarios are about helping humans to scale - and augmenting what they do, not replacing them. "They don't have the same judgement as a human being would," she says about LLM tools. She also sees concerns if store associates end up distracted by GPT prompts rather than the customers in the store. For McCandless, early use cases are about personalization, creating content, visual search, testing and learning how your customers are using your products. Using AI to match consigners with buyers is an intriguing use case, but there are data privacy concerns she'd want to cover off carefully.
Bhatia added a cautionary note about the enabling partner: you must still upskill your internal teams.
Train your product people; train your engineers. Even if you are going to use a platform, you're going to fall behind. You need to invest in it.
The use cases will mature, but Bhatia says that the tech is there now, so it's time to run tests with it, and show skeptical engineers and product people what it can do. In the end, it's about execution - reward and incentivize those who get it done.
McCandless reminded the audience: it comes down to the quality of your data. She advises creating a task force across the business to understand where generative AI is most applicable. "And then it becomes a prioritization exercise," she says. 1. Get people on board, 2. Give them a voice, and 3. Make sure your company is prioritizing. Focus on what will move the needle the most first - the level of effort for each use case is an important variable also.
I hit the Salesforce Connections show floor with two AI chips on my shoulder (yes, it's possible to have two). The first? I felt the glitz of generative AI threatened to overshadow the heavy lifting of data quality and integration. That chip got knocked off pretty quickly. Plenty of customers shared their data challenges. A "Leverage your customer data with a single source of truth" session, called a "Circle of Success," was very well attended. Hearing customers swap data challenges/stories, and seeing how Salesforce facilitated these discussions, was worthwhile.
Salesforce executives rang this bell loudly as well, albeit often in the context of Data Cloud, the "fastest growing cloud in Salesforce history." During my final interview of the day, Stephen Hammond, GM of Marketing Cloud, hit this topic head-on:
When it comes to using AI to really do advanced segmentation, or to go in and use it for propensity modeling, and other things that go along with it as well, that requires a foundation of good data. If you don't have good data going in, it's going to give you garbage out. And so we are absolutely in full agreement about what has to happen first with the organization: they really need to go through and evaluate what problems they're trying to solve.
Salesforce makes the case that Data Cloud can address these data issues more effectively than the arduous ETL efforts of the past. That's outside the scope of this piece, but it was good to see customers having candid talks on this all over the show floor.
The second AI chip on my proverbial shoulder? The over-hyping of generative AI. But there is a difference between over-emphasis and over-hype. Did Salesforce over-emphasize generative AI today? That's for customers to decide, not me.
As for over-hype, the "AI is the new UI" idea is a provocative point, and we can debate from there. My quick rebuttal: AI is not really a new UI in the way we think of UI. I see AI evolving into a personal assistant of sorts, one that will render all kinds of UIs irrelevant, or at least less popular. A general purpose AI assistant won't happen anytime soon - but it will happen.
If so, won't that turn enterprise software vendors on their heads as well? Do you want your AI assistant to tell you, "I can't complete that task, because you lack the necessary product licenses for this cloud or that cloud"? If AI is that disruptive to UIs, then won't enterprise software product categories be one of the casualties? Maybe I can get someone to joust with me on that, as Salesforce Connections day two kicks in. Let's find out...
Updated, 9:30am US PT, June 8, with a number of smaller tweaks for readability.