Digital experience platform vendors remain bullish about generative AI’s potential to make websites and apps faster, better and more personalized. They are relying on relatively new technologies and startups to get there.
Though OpenAI unveiled ChatGPT little more than a year ago, many vendors have already deployed its generative AI features to accomplish numerous tasks, including text generation, image generation, content tagging and A/B testing. Few customers, however, use it, according to Deep Analysis founder Alan Pelz-Sharpe.
“I’ll be blunt, I don’t know that that many enterprises really have a full-steam-ahead Gen AI strategy in the first place,” Pelz-Sharpe said.
“Unfortunately, everybody got behind this hype wave. Finding anything serious at an enterprise level other than, ‘Yeah, sounds interesting. We might do something with it,’ is rare,” he said. “There’s some use cases in customer service and customer support, where there’s definite interest. But beyond that it’s wait and see.”
That said, Pelz-Sharpe believes generative AI is exciting and holds promise for the long term for applications such as marketing, content operations and customer experience, especially when paired with “good, old-fashioned predictive AI.”
Pairing those two types of AI together might be the killer app that unlocks the promise of both, which is creating personalized websites or app content spontaneously, tailored to the viewer’s interests and preferences.
But success with these AI tools requires enterprise users to feed current, clean and true data into them. Most companies are behind in that regard, Pelz-Sharpe said.
The other “but” — and it is a big one — is that the whole generative AI technology ecosystem is built on startups.
DXP companies closely watched OpenAI flounder
Lost amid the fast-moving generative AI boom of 2023 was the harsh reality that large tech companies pinned their financial hopes and dreams of growth on startups.
The OpenAI crisis late last November, which saw founder Sam Altman fired and rehired within a few days, was a brutal reminder of the volatile nature of startups.
Microsoft, which has invested heavily in OpenAI, saw its market cap fluctuate by tens of billions of dollars over a span of a week beginning when OpenAI fired Altman. Then Microsoft hired Altman and some of his colleagues. At the same time, an OpenAI employee revolt prompted mass resignations. The next week, OpenAI restored Altman as CEO.
Leaders at DXP companies such as Contentful, Optimizely and Sitecore shivered. Like many application vendors in the greater tech world, they redrew product roadmaps earlier in the year to accommodate a host of new generative AI features.
“We grabbed the popcorn and read all the crazy Twitter feeds,” said Dave O’Flanagan, Sitecore chief product officer. “Just like everybody else, I was glued to it. It was like some crazy soap-opera drama.”
Sitecore was in contact with Microsoft to ensure its services based on OpenAI technology that Sitecore had wired into its DXP would be available and accessible to Sitecore customers.
The key to trusting GenAI startups that underpin DXPs’ hottest new features? API flexibility. That way, customers can use and switch generative AI large language models (LLMs) as they please, be they from OpenAI or rivals such as Amazon, Anthropic, Cohere, Meta or new challengers likely to emerge in the coming months.
Sitecore has such an open DXP model, which lets customers swap LLMs as needed; so does Contentful. That made OpenAI’s leadership turmoil less painful to watch, said Kalvin Brite, vice president of product at Contentful.
“While we [use] OpenAI in some of our solutions, we also make Google, AWS and others available as well,” Brite said. “That’s a big part of our strategy to ensure that we don’t place too big of a bet on one vendor, which causes harm not only to our customers but [also] to our business.”
At Optimizely, the reaction was similar. The company has a load-balancing scheme to switch between AI providers when the necessity arises, said Kevin Li, Optimizely vice president of product and strategy.
“There may be a little bit of cost that we eat as a result of shifting that. There may be a little bit of performance degradation. But by and large the thing that we promised customers is not an API. We promise customers the ability to generate text from a prompt. … That it doesn’t fall down is our first priority,” Li said.
The OpenAI affair reminded DXP vendors that the generative AI tech ecosystem is still nascent, Li added. Companies that tied their generative AI roadmaps exclusively with OpenAI probably had difficult conversations with their customers as it unfolded.
Generative AI tools are being used to create content such as text, images and code.
GenAI still full speed
That said, DXP vendors are steaming full-speed ahead on generative AI-based product roadmaps. Forrester Research analyst Joe Cicman said so many tech companies in the digital experience realm remade their products with generative AI that bumps in the road with startups won’t derail its momentum.
“A lot of chief strategy officers pulled their hair out because they were locked in a room, directed by their boards to figure this thing out,” Cicman said. “And if they didn’t have long term investments, they literally had to figure it out from a blank sheet. But all of them firmly believe that they’re going to have to have GenAI as a core feature.”
AI and customer experience will become inextricably linked this coming year, said Muhi Majzoub, OpenText executive vice president and chief product officer. Businesses can curate the entire customer journey with AI assistance.
“We expect to see an explosive growth in organizations using AI and machine learning for sentiment analysis to better understand customer priorities, values, behaviors and emotions — and tailoring experiences accordingly,” Majzoub said. “An area of focus for us is to explore how generative AI can further help power better marketing and customer success.”
Last year, many tech vendors grafted GenAI tools into their platforms. This year will be about connecting it to data, Optimizely’s Li said. The GenAI arms race will be led by companies that have long data histories that can be run through AI systems to create usable results.
Because of this, companies like Salesforce and Oracle stand to benefit from generative AI, even though they haven’t created their own behemoth LLMs, such as ChatGPT or Amazon Q. Their customers can tap decades of data to sharpen their marketing and sales operations.
Contentful will concentrate on extensibility of its developer-friendly platform, Brite said. To deliver digital experiences across a multitude of channels — web, mobile apps and messaging — developers at Contentful’s customer sites need to access content across sources such as digital asset management system, legacy content management systems, the cloud and elsewhere.
Three typical uses for generative AI among Contentful customers are text generation, translation and personalization. OpenAI powers some of Contentful’s features, but developers can also bring their own custom models and build apps or connectors to them. They also can mix and match Contentful partner tools.
“We are rather unopinionated about which LLM you bring to the table,” Brite said.
There’s some use cases in customer service and customer support, where there’s definite interest. But beyond that it’s wait and see. Alan Pelz-SharpeFounder, Deep Analysis
Sitecore recently hosted an internal GenAI hackathon. Some 55 teams competed and used AI not only to create content and websites but also for research and development as well as to make it more accessible, such as right clicking a mouse to generate content in the context of whatever a user is doing. Features like these are next up for Sitecore as the hackathon projects stimulate more ideas, O’Flanagan said.
GenAI promises to help line-of-business users perform sophisticated operations. For example, they can make image edits to photos, change the background of a product image or drop in a different model image.
“This is something that you’d have to go to [Adobe] Creative Cloud and do with an expert and go through a review cycle,” he said. “Now you can do it point-and-click in a web browser. So this technology provides disruptive capabilities.”
A/B testing moves to the forefront
Marketers and website designers use A/B testing to find words and layouts that get customers to buy goods or services. Personalization engines can broadcast emails or web and app content to niche audiences, sometimes even one-to-one. But it takes a lot of design iteration to find the right approach.
Enter AI for A/B testing, where words and layouts can be tested against virtual lookalike audiences at speeds never seen. DXP vendors are at different stages of implementing A/B testing features, but most of them are investing in the technology.
Acquia engaged VWO, an A/B testing partner, to directly integrate into its DXP. Others have built their own A/B testing features. EpiServer bought into the concept so heavily that it bought an A/B testing platform, Optimizely, and renamed itself Optimizely.
Cicman said A/B testing requires a new business mindset: finding “aha” moments that can stimulate a sale for a particular audience and then building on that for other audience cohorts. While that might require a lot of trial-and-error for humans, AI can self-optimize experiences. The humans in the loop, however, need to erect guardrails to prevent the technology from getting out of hand.
“That’s experimentation, what you’re going to need for content experiences — to figure out how to train your AI to generate feasible variants for different cohorts,” Cicman said. “But they must be morally anchored to the ethics of the organization.”
Pelz-Sharpe reminded users of A/B testing platforms that they still must be good stewards of their own data to reap results that drive revenue, or it is a waste of time.
“We’ve been sold it as sort of silver bullet,” Pelz-Sharpe said. “It’s not the algorithm. It doesn’t matter if NASA built the AI. It’s the quality of the data.”
Don Fluckinger covers digital experience management, end-user computing, CPUs and assorted other topics for TechTarget Editorial. Got a tip? Email him here.
![The Calibration Gap between Model and Human Confidence in Large Language Models. (arXiv:2401.13835v1 [cs.LG])](https://aigumbo.com/wp-content/uploads/2023/12/arxiv-logo-fb-235x190.png "The Calibration Gap between Model and Human Confidence in Large Language Models. (arXiv:2401.13835v1 [cs.LG])"){kind=logo}
