In Europe alone, there are 135 startups working with Generative AI (GenAI). The technology is a subset of deep learning that can generate images, video, audio and other content in response to text prompts — and according to BCG analysis is expected to achieve ~30% share of the total addressable market by 2025.
Much of the current focus on GenAI lies with the capabilities of foundation models like large language models (such as GPT-4, which powers ChatGPT) and those powering art applications such as Midjourney. But the tech is also starting to integrate with everyday tools — for example helping to write code, enhance game experiences, optimise ad targeting and create voiceovers — transforming the way people interact with software, and therefore organisations and businesses.
GenAI is exciting and has the potential to make companies more money, but to ensure greater product adoption and commercial success, user experience (UX) will be key. Humans are the end-users so GenAI solutions need to work for them and should encourage them to use the solution repeatedly.
We dove into the opportunities, challenges and best design practices to keep in mind when approaching UX in a GenAI world.
Where to start
“GenAI will create interfaces on the fly, custom to what users request from it,” says Phil Gerrard, managing director and partner and design lead at BCG X, the tech build and design unit of Boston Consulting Group. “This has profound implications for how you approach experience design, as historically the designer controls this — and tests this — in the new paradigm that control is lost to the AI engine.”
Companies should start integrating something off-the-shelf like ChatGPT to get their feet wet in GenAI
How then, can companies curate UX without knowing each outcome? Gerrard recommends building a sandbox — a design system that’s plugged into the relevant GenAI models and tools (with data) so the designer can get a feel for how the UX will behave before launching it to customers.
A sandbox provides “the confidence and security to not launch completely into the unknown. It gives businesses the ability to control and minimise risk,” Gerrard says, although he notes it’s still early days and is still conceptual, but he expects to see this approach more as GenAI UX matures.
Also because GenAI is still in its infancy, the landscape of tools for design in the space is constantly changing, but Gerrard notes that there are a few “no-regret moves” for trying out GenAI. “Use ChatGPT for research and ideation, Copilot from Github to write code automatically and LangChain to create custom applications and agents,” he says.
Matt Lehmann, COO of Aflorithmic, a company using GenAI to produce synthetic voiceovers at scale, agrees: “Companies should start integrating something off the shelf like ChatGPT to get their feet wet in GenAI.”
He adds that the initial goal should be to “establish UX that lets users interact with AI as if they were talking to a human. You should make the technology and complexity [of GenAI] disappear”.
Double down on prompt engineering
GenAI can result in fast, high-quality results for users, but much of its insights rely on the strength of the initial input from the user. That’s why prompt engineering is so crucial for successful GenAI — how a question or instruction is entered into a machine learning model will determine (in part) the accuracy of the output. It’s also why the role of prompt engineer is seeing more interest.
If your GenAI always delivers the same thing no matter what the input is, users will quickly be disappointed
“Companies need to guide users on how to best use GenAI,” says Alan Sternberg, founder and CEO of Beams, an AI insights platform for highly regulated industries. He says that poor prompts get poor results, so companies need to take care not to make users feel confused and give them parameters to submit concise, effective prompts.
For Lehmann, prompt engineering is directly related to user engagement. “Basic prompts are easy to build and reasonably fast. But be aware: AI is ultimately pattern recognition and humans are very good at figuring out patterns,” he says. “If your GenAI always delivers the same thing no matter what the input is, users will quickly be disappointed or they won’t be able to use your product as much as they’d like to.”
To improve prompt engineering, Lehmann advises utilising multilayered frameworks for GenAI or bolting on other services that gives users options to clarify their prompts. For example, Midjourney has a number of free text modifiers that users can add to their prompt to determine the output qualities of an image.
However, for Sternberg, there’s a fine line between good prompt engineering and simply offloading the burden of meaningful UX to users. “In my opinion,” he says, “there has to be a bigger focus on moulding GenAI to limit the surface area of the interface and guide users with smarter choices to achieve their goals.”
Remember that GenAI doesn’t exist in a silo
Ultimately, Gerrard says that from a user-experience perspective, GenAI should not be considered in isolation. “We need to consider how [GenAI] connects and combines to the wider user experience to deliver more value,” he says.
It’s difficult to predict how the UX with GenAI will evolve
In particular, Gerrard highlights the potential of GenAI and AI in the retail sector, where users could potentially ask a GenAI agent to purchase items for them from multiple retailers, and AI carries out the task. While the process is streamlined, it cuts out direct interactions between brands and users, and so retailers have fewer opportunities to commercialise the experience.
“The direct-to-consumer movement dis-intermediated many industries by cutting out the middle-man and connecting consumers directly to suppliers. The GenAI and the agent model will re-intermediate,” Gerrard says, adding that we should expect a lot more innovation.
This demands creativity from retailers who will need to re-think how they own and differentiate the customer-experience in this new paradigm.
When investing in GenAI, the BCG formula is 10% goes toward building algorithms, 20% on the foundational tech, and the remaining majority for innovating processes, team structures, and operating models in GenAI.
“It’s difficult to predict how the UX with GenAI will evolve,” Sternberg says. “The technology is growing exponentially, unlike anything before it.”