‘Fail Fast and Fail Forward’ When Implementing AI into Workflows
This is the second article in a two-part series. In the first article, AlixPartners shared a framework and approach to get on track with Generative AI and drive value for retailers and brands. This article explores the keys to impactful implementation of AI in retail.
It’s tempting to jump on the buzzy Generative AI bandwagon amid fears of getting left behind, but in today’s competitive fashion and retail climate, diving in without proper preparation can compromise a company’s ability to effectively integrate the technology. New technical solutions often prove to be a waste of time and money due to flawed or misguided implementation, poor adoption, and/or the lack of integration with business processes.
More from Sourcing Journal
Knitting Sustainability, Tracing Excellence: Forging Value Through Innovative Textile Initiatives
How Chubbies Used AI to - Literally - Lengthen its Product Line
LENZING™ ECOVERO™ Black Viscose Offers Reduced Carbon Footprint and Durable Color
Global management consulting firm AlixPartners has devised a self-assessment framework to help retailers and brands look before they leap into AI, however, accurately assessing needs and abilities is just the first step. Implementation is often the hardest part. Companies must begin with a well-defined strategy and clear objectives, actively measure AI’s efficacy with KPIs, then quickly and constantly course correct as needed. In other words, “fail fast and fail forward,” according to Angela Zutavern, partner and managing director and AI expert in AlixPartners’ Digital Practice.
Here, we explore basic steps companies must take while applying AI into their workflows, all with the objective to drive conversions, reduce costs and increase profits, as well as boost average customer value, acquire new customers and gain market share.
Know your AI goals
AI can be applied narrowly or broadly, but many companies make the mistake of putting the tech before their business concerns. “Experimentation is absolutely critical for any AI-generated initiative, because not all of them will work the first time. But initiatives must be goal-oriented,” said Zutavern.
Companies must first have a clearly defined business case to know how the AI initiative can impact the business results. Once they have those financial metrics up front, they can pilot with a minimum viable product.
“If you expect a particular AI initiative to reduce customer churn or increase customer lifetime value, for example, you’ve got to be measuring that along the way,” she said. “You need that historical baseline where you start and then you need to measure it throughout the pilot phase and the implementation phase, to make sure that you are making progress in those metrics that you’ve defined up front.”
As part of its self-defined digital transformation, Spanish retail chain Mango identified a broad range of touchpoints across its value chain that could benefit from AI—from pricing to personalization—and since 2018, it rolled out 15 different platforms to address them. Midas, for example, is used for pricing policy, while Gaudí recommends products online. Mindful of its massive international clientele, Mango programmed its AI customer service bot Iris to speak to customers in 20 languages in over 60 countries.
For its most recent initiative, Mango’s goal was to help employees and partners create—or rather, co-create. The GenAI images on the Inspire platform help design and product teams envision new prints, fabrics, garments and more, and Mango already has 20 garments in the market co-created by AI. On the conversational side, newly launched employee-facing AI platform Lisa generates text a la ChatGPT but is trained specifically for Mango. Developed in just nine months, Lisa utilizes both private and open-source models.
While Mango’s customers already communicate with its conversational agent, Iris, the company wanted to improve its capacities by equipping it with “interactive conversational generative AI, in order to allow a more fluid interaction with our consumers,” said Jordi Alex Moreno, Mango director of technology, data, privacy and security. “In other words, we wanted to go from a conversational assistant for specific applications to an interactive conversational assistant that can deal with multiple applications in our commercial and social media channels.”
For multi-brand beauty retailer Ulta Beauty, their AI strategy has always centered on the customer.
“The speed in which we adopt AI technologies reflects our guests’ evolving needs and interests, which has only accelerated over the years as beauty enthusiasts turn to digital tools to explore the category,” said Prama Bhatt, chief digital officer at Ulta Beauty. Although the chain was already working on digital innovation, the pandemic accelerated these efforts while in-store product testing was inaccessible. In addition to guest-facing solutions like a Virtual Beauty Advisor and GLAMlab virtual try-on, Ulta uses AI internally to “help accelerate the speed of asset creation as well as help cut down time spent on laborious tasks.”
Looking ahead, Bhatt also sees the potential for AI to support store employee training, noting, “In beauty, it can be difficult to be an expert on all categories and products but with GenAI we may be able to put the level of expertise at their fingertips to educate both guest and associates seamlessly.”
Build a skilled team
Artificial intelligence can learn, but it must be taught carefully and managed with skillful oversight. This is particularly crucial with consumer-facing initiatives where an offensive chatbot or social media post can sink a company PR-wise.
To preserve brand integrity across text or images, AI must learn the brand voice for generative marketing pieces and the brand aesthetic for creative outputs, a fact that can cause concern. In Jasper’s 2023 AI in Business Trend Report, a study of 500 professionals found that “factual inaccuracy” (36 percent) was the largest concern for generative AI, followed by “generic outputs” (35.1 percent) and “outputs that lack correct tone” (26.4 percent).
Smart output comes from smart input, and the growing expertise of “prompt engineering” helps users write prompts for the most meaningful GenAI results. Fashion, which has notoriously lagged in technology, can look outside the industry to the tech industry or college grads skilled in this area.
Once teams are in place to teach and monitor AI, others need to stay abreast of AI’s fast-moving advances so systems can be adapted as needed. “In all the projects created in Mango’s Technology Department, we [always took] into account the constant innovations being developed in the sector as well as the speed at which this area is advancing,” said Moreno. “Specifically, within the sphere of GenAI, new implementations appear very quickly, and we have to develop solutions that are scalable to the value chain, flexible, and allow us to adapt to potential improvements.”
To make this possible, Mango “assembled a multi-disciplinary team with different profiles of user experience, architecture, back/front-end development and data science, in addition to the involvement of functional areas of the company.”
Bhatt also noted the need for the right talent. “To replace an experience that is so uniquely personal to beauty enthusiasts like physically trying on makeup with a digital app on your phone was a challenge in itself, but being able to staff such an undertaking would be another hurdle,” she said. Ulta gathered the teams it needed through its acquisitions of GlamSt and QST, which gave it “access to the technical expertise required along with the talented data scientists.”
All this, however, requires buy-in from the top. Zutavern said she often gets asked by data science teams how they can have more influence on the business. “I think it’s just making sure that the business and technology initiatives are really integrated and that we don’t have a disconnect between what’s happening in technology and what’s happening in the day-to-day business,” she said.
Make processes AI-ready
Maher Masri, president of AI software company NAX Group, noted that one of the biggest challenges in implementing AI into workflows is the speed of experimentation. “Innovation in AI will be measured in weeks and months, not years. Corporates must create cultures that are maniacally focused on rapid experimentation at-scale to unlock data in new ways,” he said.
So while AI has great capabilities, companies must often re-engineer existing systems and protocols to take advantage of them. “It’s like in the early days of automation,” said Zutavern. “If you just automate an old process, you’ll get some but not all the benefits you would if you rethink the entire process. Same with AI.”
Without a blank check, however, fashion companies must be strategic. So where to begin?
Masri suggests starting with a view of impact potential across the value chain and key processes. “The largest economic and strategic opportunities should be prioritized by where proprietary data sets exist that provide an unfair competitive advantage,” he said. “Enterprise valuations will increasingly be centered around a company’s data story and it is important to demonstrate growth and productivity impact across a range of areas across the enterprise,” he said. Data can be used to drive value across several areas such as supply chain, product development, merchandising, service, etc.
Experimentation is good, but the key is to measure results along the way. “Some companies have hundreds of AI initiatives in flight but aren’t really sure where the benefits are coming from,” said Zutavern, who recommends quarterly portfolio reviews/adjustments to analyze value and gaps for companies spending millions of dollars on AI initiatives, and annual reviews for the rest.
Mango looked at its 15 AI platforms through multiple impact lenses, depending on each use case, and created “financial KPIs, commercial KPIs, customer experience KPIs and adoption KPIs, among other parameters.”
To measure its efforts, Ulta uses an Innovation Success Experimentation framework that looks at both quantitative and qualitative factors. “For us, success has many faces. We know this is not a race to the finish line, rather a marathon that will have plenty of learnings along the way,” explained Bhatt. “Currently, we’re focused on understanding how GenAI technologies can help us successfully deliver value in supporting our aims to speed time-consuming tasks, accelerate the asset creation process, unleash greater power of data, strengthen beauty advisory, and personalize digital experiences.” Among Ulta’s efficiency boosters has been the low-code AI environment Interplay, which improved development speed.
Communicate, Learn, Iterate
When it comes to integrating AI into workflows, companies must realize it won’t be a one-and-done scenario. The cycle of assessment and implementation should repeat often in a series of continuous improvements, sharing wins along the way and never allowing perfectionism to block progress.
And until AI is part of every company’s vernacular, companies must stress how the tech will enhance employees, not replace them. “Generative artificial intelligence is an extended intelligence, in other words, a technology that will act as a co-pilot for our employees and stakeholders, and help us extend our capabilities,” said Mango’s Moreno. “Technology will either make us more human or be of no use.”
AlixPartners will be speaking on this topic at WWD’s Apparel & Retail CEO Summit on October 24. “Retail Disrupted: Unlock AI Value,” will feature speakers RJ Cilley, chief operating officer, Saks; Danielle Schmelkin, chief information officer, J.Crew Group; Maher Masri, president, NAX Group with Sonia Lapinsky, partner & managing director, AlixPartners.
