The Unsexy Future of Generative AI Is Enterprise Apps

Keith Peiris says he started to see the generative AI writing on the wall six months ago.

Peiris is the cofounder and chief executive of Tome, a San Francisco startup that makes presentation software juiced with generative AI. The company launched its product in early 2022 with a healthy cushion of $32 million in venture capital funding, and successfully surfed the ChatGPT hype wave after that, raising even more funding in early 2023. Venture capitalist and LinkedIn cofounder Reid Hoffman, former Google CEO and chairman Eric Schmidt, and Stability.ai’s then CEO Emad Mostaque were all backing Tome.

Tome had one problem, though: It wasn’t generating meaningful revenue. And AI startups like Tome, which build their services on top of both open source and proprietary language models, pay significant fees to companies like OpenAI to power their apps. Some kind of action was needed if Tome was to keep the lights on.

Peiris and his cofounder Henri Liriani ended up laying off 20 percent of their 59-person staff last month. They also announced a new focus: Their app, which is often described as PowerPoint-on-GenAI, would be aimed squarely at enterprise customers. They would now charge three times what they were charging premium users.

“We realized we were going to run out of time if we needed to teach Tome’s AI models how to do high school homework, how to write post-surgery guides, how to craft marketing briefs and sales briefs,” Peiris said in an interview with WIRED. “We said, let’s pick a segment of customers that not only have a lot of presentations to build but also have clear outcomes, like whether they closed a deal or not. And that is salespeople.”

Tome’s homepage used to call the app a “collaborative AI partner,” an all-singing, all-dancing helper. Now the company markets itself as “the leading AI-native research and presentation platform for sales and marketing teams.”

Pivotal Moment

Tome’s turn toward enterprise customers falls in line with other AI startups—and some billion-dollar unicorns—that launched gen AI apps into the intoxicating buzz sparked by ChatGPT. As the cloud API bills keep piling up, they’re selling more narrow and premium products, experimenting with different pricing models to juice revenue.

Last week, Perplexity, another buzzy startup that offers an AI-powered search engine, announced Perplexity Enterprise Pro (along with another massive round of funding). Perplexity says its early customers include Zoom, HP, payments startup Stripe, and the Cleveland Cavaliers. Some are using it to drum up sales pitches; others are using it for research. Perplexity is charging “Pro” customers $40 per month, or $400 per year, per employee.

Sierra’s Taylor is also the chairman of the board at OpenAI, which, with its $80 billion or more valuation and more than $10 billion in support from Microsoft, is not a conventional startup. But OpenAI has been ramping up its enterprise sales business, too.

James Dyett, head of platform sales at OpenAI, says the company has spent the 18 months since ChatGPT launched in 2022 building out its software sales and go-to-market team, growing it from 15 people to 200 employees. The group now makes up one-fifth of OpenAI’s current workforce.

OpenAI has two enterprise products in addition to its free and $20-per-month versions of ChatGPT. Last August it rolled out ChatGPT Enterprise, a version with enhanced security features that costs $60 per “seat” or employee. The company also sells cloud access to APIs that developers and companies can use to build AI products on top of OpenAI’s language models. API pricing is per “token,” a term for a chunk of text output comprising a word or piece of a word. For access to OpenAI’s most powerful model, companies pay up to $0.12 per 1,000 tokens.

“In the early days of ChatGPT there were a lot of poems and rhyming schemes that blew people away, but I suspect what happened is a lot of executives saw the technology behind the chat interface and immediately recognized this wasn’t just a toy,” Dyett says. “They could ask legal questions. They could ask marketing questions. They could summarize documents.” The upshot was a vast new business opportunity for OpenAI, which also needs big revenues to offset the huge costs of training its AI models.

“It kind of clicked for the C-suites of the Fortune 500 companies that there’s a real place for this in their businesses,” Dyett added.

Bills, Bills, Bills

AI startups are expensive companies to set up and operate because of their need for computing power, Peiris says. “If you’re a VC looking at these businesses, you might not believe a company that says it’s going to scale up to 50 million or 100 million users and figure out how to monetize later. It’s different from the mobile or social era.”

AI startups are operating right now in a tighter market for investment funding, where high interest rates have made investors skittish. So, to Peiris’s point, creating opportunities for steady recurring revenue and not being subject to the whims of consumers can help them craft a more compelling pitch to VCs. It’s a page borrowed directly from Big Tech companies like Microsoft, Google, and Salesforce, which have long capitalized on traditional software-as-a-service models and are now infusing their own products with heavy doses of AI. But unlike many early-stage AI startups, those businesses are self-sustaining.

Pitchbook, which tracks venture capital and private equity, has been logging investments in generative AI startups since 2021. By late 2023 companies in that category had raised $23.2 billion, up 250 percent over 2022’s total, its figures show.

However, that amount includes massive funding from corporate backers, like Microsoft’s infusion of capital into OpenAI and Amazon’s funding of Anthropic. Stripped down to conventional VC investments, funding in 2023 for AI startups was much smaller, and only on pace to match the total amount raised in 2021.

Pitchbook senior analyst Brendan Burke noted in a report that venture capital funding was increasingly being funneled towards “underlying core AI technologies and their ultimate vertical applications, instead of general-purpose middleware across audio, language, images, and video.”

In other words: A GenAI app that helps a company generate ecommerce sales, parse legal documents, or maintain SOC2 compliance is probably a surer bet than one that drums up a clever video or photo once in a while.

Clay Bavor, the cofounder of Sierra, says he believes it’s not necessarily computing or cloud API costs driving AI startups towards B2B models, but more likely the benefits of targeting a specific customer and iterating on a product based on their feedback. “I think everyone, myself included, is fairly optimistic that the capabilities of these AI models are going to go up while costs come down,” Bavor says.

“There’s just something really powerful about having a clear problem to solve for a particular customer,” he says. “And then you can get feedback on, ‘Is this working? Is this solving a problem?’ And if you build a business with that, it’s very powerful.”

Although ChatGPT triggered an AI boom in part because it can nimbly generate code one second and sonnets the next, Arvind Jain, the chief executive of AI startup Glean, says the nature of technology still favors narrow tools. On average a large company uses more than a thousand different technical systems to store company data and information, he says, creating an opportunity for a lot of smaller companies to sell their tech to these corporations.

“We are in this world where there are basically a bunch of functional tools, each solving a very specific need. That’s the way of the future,” says Jain, who spent more than a decade working on search at Google. Glean powers a workplace search engine by plugging into various corporate apps. It was founded in 2019 and has raised over $200 million in venture capital funding from Kleiner Perkins, Sequoia Capital, Coatue, and others.

Error Checking

Tuning a generative AI product to serve business customers has its challenges. The errors and “hallucinations” of systems like ChatGPT can be more consequential in a corporate, legal, or medical environment. Selling gen AI tools to other businesses also means meeting their privacy and security standards, and potentially the legal and regulatory requirements of their sector.

“It’s one thing for ChatGPT or Midjourney to get creative for an end user,” Bavor says. “It’s quite another thing for AI to get creative in the context of business applications.”

Bavor says Sierra has dedicated “a huge amount of effort investment” to establishing safeguards and parameters so it can meet security and compliance standards. This includes using … more AI to tune Sierra’s AI. If you’re using an AI model that generates correct responses 90 percent of the time, but then layer in additional technology that can catch and correct some of the errors, you can achieve a much higher level of accuracy, he explains.

“You really have to ground your AI systems for enterprise use cases,” says Jain, the CEO of Glean. “Imagine a nurse in a hospital system using AI to make some decision about patient care—you simply can’t be wrong.”

A less predictable threat to smaller AI companies selling their wares to enterprise customers: What if a giant gen AI unicorn like OpenAI, with its burgeoning sales team, decides to roll out the exact tool that a singular startup has been building?

Many of the AI startups WIRED spoke with are trying to move away from depending entirely on OpenAI’s technology by using alternatives like Anthropic’s Claude or open-source large language models like Meta’s Llama 3. Some startups are even intent on eventually building their own AI technology. But many AI entrepreneurs are stuck paying for access to OpenAI’s tech while potentially competing with it in the future.

Peiris, of Tome, considered the question, then said that he’s singularly focused on sales and marketing use cases now and “being amazing at high-quality generation for these folks.”

Source : Wired