
AI: Microsoft & Salesforce Sweating Enterprise AI Adoption. RTZ #847
Regular readers here at AI: RTZ know I’ve been beating the drum about tech adoption generally lagging tech innovation and deployment. Both on the consumer and the enterprise side.
And this AI Tech Wave is no different. It’s somewhat masked today given the unprecedented Scale of LLM AI research and compute deployment, both at the Boxes 1 & 2 infrastructure and Boxes 3-5 software layers of the AI tech stack below. Regardless of the extraordinary amounts expended for AI Talent and Chips.
But there are almost always real world friction points that don’t track at the same speed.
Even with technologies like AI that are seeing the fastest growth to big numbers vs prior tech waves. We saw that of course post OpenAI’s ChatGPT moment November 30, 2022, with its adoption by over a 100 million folks globally two months. And of course AI Coding firm Anysphere Cursor’s financial growth to over $100 million ARR, also in record time measured.
Those are always impressive. But the hard task is scaling the numbers far beyond them, again in near record or record time.
This week saw that reality underlined on the Enterprise side of AI, with both Microsoft and Salesforce showing issues with pushing mainstream adoption. We’re seeing that with Microsoft Copilot sales on top of its $87 billion+ annual Office 365 productivity global empire. And Salesforce’s efforts to push its AI Agentforce product line on top of its multi-billion enterprise SaaS software empire.
Despite the best efforts by their CEOs Satya Nadella and Marc Benioff respectively rallying the entrenched troops.
The Information outlines the former in “Microsoft Hopes Hastened AI Rollout, Price Discounts Can Fuel Office 365 Growth”:
“CEO Satya Nadella reorganized who’s in charge of buffing up the crown jewel software, which risks falling behind in the AI boom.”
“For nearly 30 years, Office has been one of Microsoft’s most prized crown jewels. The software accounts for roughly 30% of the company’s overall revenue and it still sees considerable growth—some 13% last year, to $87 billion. But Nadella has lately become increasingly concerned that Office’s growth could slow if Microsoft doesn’t add better AI to the software and convince customers the AI feature is worth it. Right now, 365 Copilot routinely struggles with basic functions, like accurately filling in what comes next in an Excel spreadsheet.”
“In response, Nadella has in recent months made improving 365 Copilot’s quality a top priority, executing a series of sweeping organizational and product changes meant to speed up development and rollout of the AI features. Meanwhile, Microsoft has been gradually reducing the software’s price with more generous discounts on the AI features, according to customers and salespeople.”
“And Microsoft has fallen behind parts of Nadella’s original timeline and is still working out kinks in 365 Copilot’s more advanced AI features that it promised would soon be ready when it launched the product in 2023. Some features that it initially planned to ship earlier this year have been postponed.”
This despite Microsoft taking the unprecedented step of diversifying beyond core AI Partner OpenAI’s models to its rival Anthropic:
“At Nadella’s direction, those executives have been moving hastily in recent months to make the Office AI features more valuable to customers. In one sign of that push, Lamanna has in recent weeks spearheaded an effort to swap out OpenAI’s models for Anthropic’s to power some of the more aspirational features in Excel and Powerpoint, according to two people involved in the effort. The move is noteworthy because it will be more costly to Microsoft, whose massive investment in OpenAI grants it the rights to reuse the startup’s models. By contrast, Microsoft will pay for Anthropic’s models via its cloud archrival Amazon Web Services.”
“At Microsoft, engineering leaders working on AI features in Office have complained that OpenAI’s models weren’t reliable enough to handle large amounts of data without making mistakes or hallucinating, according to two people involved in the effort.”
“Encouraged by success with Anthropic’s models, Microsoft is preparing a new wave of Copilot updates in Office before year’s end to improve tasks such as complex Excel calculations and designing visually appealing PowerPoint presentations, according to two people involved in the effort. Some of those announcements could come as soon as Microsoft’s Ignite conference in November, one of the people said.”
Similar herculean efforts have been underway at SaaS leader Salesforce, led by the AI enthusiast founder/CEO Marc Benioff.
The Information also goes on to illustrate Salesforce’s Enterprise headwinds in “Marc Benioff Saie AI Was Easy. A ‘Crazy’ Team at Salesforce Proved Him Wrong”:
“Salesforce’s CEO said this would be the ‘absolute year of Agentforce,’ the name of his company’s AI product. It hasn’t turned out that way, as the enterprise software pioneer faces skeptical customers and greater competition from its peers as well as AI startups.”
“Last fall, one of his technical teams told large Salesforce customers that using Agentforce, the software firm’s new artificial intelligence for automating customer service and other functions, would require extensive planning. The product information the Salesforce team shared with those customers acknowledged the complexity of making the AI perform well.”
“That message contradicted Benioff, a charismatic salesman who had been telling customers that using the AI was a cinch and they could set up AI agents to handle customer service discussions and other tasks in minutes.”
“A senior Salesforce leader said that when Benioff noticed the disconnect, he complained to colleagues that the technical team advising large customers, known as Well-Architected, was acting “crazy” and should be fired.”
“Two months later, after Agentforce became available to all customers, the Well-Architected team was quietly disbanded, with some members taking severance packages and others finding new roles in the company, a former Salesforce manager said. But the team’s insistence that setting up agents was harder than it looked ended up being right, and Benioff was proven wrong: The company has struggled to sell Agentforce in the past year, in part because of the extensive prep work customers need to do to make it work right. And Salesforce in August said it would relaunch the Well-Architected program.”
Of course, both companies are also competing against each other for the same enterprise customers in most cases.
Both pieces are worth reading in full for the details and nuance of the AI adoption headwinds in two key Enterprise AI market categories. But the reality is worth registering that AI Adoption does not Scale at the same pace as AI hardware/software and Infrastructure in this AI Tech Wave.
And private and public investors need to keep that in mind regardless of how fast the ‘flashing green’ the current lights seem to be now. We’re not there yet, and will have to wait for it. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)