The current run rate of AI development “is not sustainable” according to Salesforce’s global chief marketing officer, Ariel Kelman.
Lead image L-R: Ainsley Harris, senior writer, Fast Company; Ariel Kelman, president and CMO, Salesforce; Greg Howell, president and co-owner, Flexo Concepts; Scott Morris, CMO, Sprout Social; Robert Marusi, chief customer officer, Turtle Bay.
Speaking at the company’s Connections conference in Chicago, Kelman said that the frenzy we’re seeing for AI development with “every enterprise IT department training models” cannot be sustained due to the sheer cost of AI development.
A recent Morgan Stanley report predicted that global capex spending on cloud infrastructure — essential for powering the creation, training and development of AI models — would reach $US300 billion (just over $AU450 billion) by the end of the decade. That figure shows a 44 per cent year-on-year increase in spending, up from the 26 per cent the investment bank predicted a few months before.
However, while the cost implications (not to mention the environmental ones) of AI development are vast, Kelman believes that the massive increase in development cost does not necessarily mean that businesses will be left behind.
“It’s not required any more, given how smart these [AI] models are. If you look at GPT-4 that just came out, it’s better, faster and cheaper and it’s the second release they’ve done in the past six months where they delivered order of magnitude cost savings,” he said.
“There is relief in sight for corporate America and corporations around the world to not have to spend such massive amounts of money on training.”
During the rest of the Connections conference, Salesforce took great pride in telling attendees of its agnostic stance on AI model integration, as well as its Einstein Trust Layer, which provides something of a firewall between the AI models and companies’ private customer data. Kelman also took pains to show that all of the AI tools within Salesforce’s suite of Clouds require a human sign-off before pushing out comms to Joe Public.
“We have a product called Einstein Conversation Insights. It records the conversations that sales and service people have and one of the interesting features we’ve added to it is that at the end of a call that a sales or service person has, it can summarise the call and give an email that they could send back,” he explained.
“That human aspect is absolutely critical. It’s not writing and sending the email for the salesperson, it’s writing the first 80-90 per cent of the email for them to go and edit and make sure it’s great. It literally becomes an 80-90 per cent time saving for them.”
Kelman also that, despite much chatter to the contrary, Salesforce’s approach to AI means that marketing communications or sales staff do not need to be retrained as “prompt engineers” to make the most of AI. This means that customers should have better interactions with brands because their communications are more targeted and personalised. During a keynote, the tech provider reaffirmed that the number one reason for customers to stop purchasing with a brand was “poor service.”
“We built our AI platform in a way to not make that happen [surfacing of incorrect information through AI-generated data fields],” he said.
“You could have ten emails prepared to use in reply to customers. But let’s say, for example, it’s how to reply to a customer email about status on a shipment. You can embed in there a piece of data that the customer is going to ask for and you do a callout to get it from this place.
“Basically the large language model (LLM) was told, ‘Don’t try and figure this out, use this data in your answer’. That’s the approach that we’ve taken to be explicit about it. It also goes to why the vast majority of employees don’t need to be retrained to become prompt engineers. You have to retrain the people who set up the systems and do the administration to put the intelligence and guardrails in. But then everyone else should just be able to use these things in a natural way.”
While many companies in the tech sector, such as Google, Meta, Adobe and even Elon Musk’s X Corp, have invested huge amounts of time and money — as well as marketing and comms professionals’ time — in telling the world the wonders of their AI tech, Salesforce’s approach does seem different. It is firmly grounded in what is actually useful for its customers, rather than generating pie-in-the-sky images and videos for the public to swallow up and wring their hands about.
Fellow panellist Scott Morris, the CMO of Sprout Social, perhaps summed it up best.
“We spend a lot of time thinking about [the human side of AI]. People want that human interaction when it’s appropriate. But also, and we’ve done a lot of research on this, folks know when you’re using an AI and they completely welcome it when it’s doing something that’s making their experience better or more productive,” he said.
Kelman may have called time on the AI arms race but companies are still finding their feet with the tech. We’ll dig into that on B&T tomorrow.
