Retail reimagined: Adore Me accelerates time to market with Writer AI Studio

Listen to the Adore Me story or read the edited story below

Writer is the full-stack generative AI platform for enterprises. We make it easy for organizations to adopt AI apps and workflows that deliver quantifiable ROI quickly.

Writer helps organizations build highly-customized AI apps that compress entire business processes, support complex use cases, and infuse work with company intelligence. Our enterprise-grade platform can be deployed flexibly, keeps your data private, and adheres to global privacy laws and security standards. Leading enterprises choose Writer, including Vanguard, Intuit, L’Oréal, Accenture, and Kenvue.

Tell us about yourself and what you do at Adore Me.

I’m Ranjan Roy, the SVP of Strategy at Adore Me, a direct-to-consumer intimate apparel company based in the US, founded in 2011. We were the first intimate apparel company to bring extended sizing to the market all the way back in 2013. We’re a certified B corporation, so sustainability plays a role in all parts of our business. In addition, we also focus on technology, innovation, and discovering new ways to serve our customers.

At Adore Me, we’re about 500 people, with 140 on the tech side, so technology is really at the core of what we do. I ran a tech startup focused on content personalization for a number of years, and we tried experimenting with natural language generation back in 2013. It wasn’t very good at the time, but it started this interest in what became generative AI. Now, as generative AI use cases have exploded within the company, I sit between the marketing, content, and technology side of things.

What led you to pursue generative AI, and why did you choose to work with Writer?

We’d become a certified B corporation, and compliance around sustainability was becoming an important part of the way we produced content. So we realized we needed to come up with some tool or other. We realized we needed to work with a partner who could help us create style guides that could be used across the company.

Writer was the clear solution once we started researching different products. But for us, the partnership has grown in this very interesting way, because as generative AI and large language models have exploded, we’ve been able to work together, build together, and have proudly been the alpha testers for several new products that made a lot of peoples’ lives a lot easier.

An important aspect of Writer that’s impressed us these last few years is the fact that Writer doesn’t follow ‌trends that are happening in the industry, but actually creates products and infrastructure that lead the way. As generative AI is developing so quickly, many products in the market are not tailored for business use cases, especially taking into account concerns surrounding security and privacy. Writer AI Studio is a great example of how Writer products and infrastructure have evolved faster than the market‌.

How is generative AI used within your organization?

We see limitless potential with generative AI. We’re using generative AI for any kind of content creation, things that show up in front of the customer. We also envision using generative AI to create completely new experiences for customers in hyper-personalized ways, like the way products are presented on the website, product recommendations, and the language around those products. Ultimately, we see generative AI in every part of the business, but we’re only starting to scratch the surface. We can take processes that used to be writing long Excel formulas or writing lines of code and instead, with generative AI, use language to create solutions to answer specific business problems.

To start prioritizing use cases, we’d create a two-by-two matrix. One axis reflects how unstructured versus structured the content is, so something like an email subject line or a product description is highly structured, whereas completely original web copy for a marketing campaign is unstructured. On the other axis, we have ‌revenue risk‌ — ‌how high versus low is it? We really wanted to start tackling content opportunities that are highly structured but represent a relatively low-revenue risk.

People need to incorporate thinking from the perspective of a risk manager for every piece of content that gets generated. The perfect places to start are pieces of content that won’t have a huge negative impact if it’s not perfect. The most important part of any kind of generative AI strategy is getting started, experimenting, and learning, then moving forward with those learnings.

What were some of the initial use cases that were most impactful for your team?

Product descriptions were our first use case, and it was a big success. It was quick and easy, and it aligned with the larger business strategy at the time. It was summer of 2022, and as folks in direct-to-consumer commerce or online advertising know, iOS 14 disrupted online acquisition, so SEO became a priority. We created an AI app with Writer to generate the product-recommended product descriptions, which were highly SEO-optimized. We sell everything in sets, but “bra and panty set” isn’t the most natural thing to write; however, it’s actually what people search for. With the Writer app, we increased our search volume for that specific term immensely because we were able to train the model to answer that specific question. We quickly saw results — a 40% increase in non-branded search.

We also have a service called Dailylook, which is a home try-on service. You get a box with four to eight items, and you keep what you want, sending the rest back. In the process, there is a human stylist who’s picking those items for you and including a handwritten note. We see a significant uplift in the number of items people keep when a note is sent, saying, “I think you’re going to love this product because of this reason, this product goes really well with this other product, this is how you can style them together.”

We thought this would be a good use case to test a generative AI solution. In these notes, the stylists were writing 400 to 500 words over one thousand times a day. It’s repetitive and painful, and the perfect example of how we used an AI app that’d create a first draft with the correct language and style. The stylist was still in the loop, but there were initial worries about how this workflow transformation would be received. But the stylists were very happy with the first drafts of the notes that Writer produced. This was the perfect use case of AI getting you 95% of the way there, but the user adding their personal flair for the last 5%. The time that it takes the stylist to write the note is already down 36%. As people get more comfortable with this new workflow, we see the time savings improving even more.

Writer AI Studio is our suite of development tools to quickly build AI apps. How has AI Studio impacted your process of building apps?

The very first AI app we created before AI Studio was for product descriptions. When we created this app, there were only a few of us who were more focused on generative AI working on this. Now, we can bring the end user into the process using the no-code element of AI Studio and have them work on prompts.

We created a new role called prompt coordinator, who is someone in the company interested in prompting. This is less of a technical engineering role, but more to show how prompts work to the end user and how to get the desired output. We’ve seen multiple use cases go much faster. What used to take four months now can take ten days, one week, or less, depending on the use case.

AI Studio comes with both no-code tools and Writer Framework, an open-source Python framework to build feature-rich apps. Now my business and engineering teams can seamlessly collaborate on AI apps. Business users can use no-code tools to spin up new apps and adjust prompts, while engineers can use Writer Framework to build sophisticated UIs and embed apps in our system. We have a very flexible and easy way to create structured inputs and prompts with the no-code element, and you can see the output right away and instantly address any issues.

We recently launched Adore Me Mexico, and we had a big task for us around translation and very quickly we realized that this was something AI Studio could help us with. We put together a small team of a prompt coordinator, a merchandiser, and a native Mexican Spanish speaker who went into the no-code platform to work on this. They started with the English input, trying to get a direct translation into Spanish, but then realized they could add another layer to improve the translation.

They added another prompt to take the outputted translation and ask for suggestions that could improve it for a consumer in Mexico looking for fashion products from an intimate apparel company. This allowed us to get into the nuances focused on selling to consumers in Mexican Spanish. So it’s not just a translation. Our team is going over and taking these suggestions we think are valid, and incorporating them back into the prompt.

All of this is being done without a developer, and we’re able to get a very good result and prompt within the no-code side that we can be confident and proud enough to put it on the website. With the Writer Framework, we’re able to turn this into an ongoing app where the end user doesn’t even see the prompt. Anyone in marketing or copywriting can enter some English language and have it translated.

Because of the no-code platform, the prompts we generate are available via the API and we can run batch processes. We could run 2,900 product descriptions at once, get the output in a csv, and then in our own backend upload those to our website. The whole process of solving both the one-off use cases and these batch processes has all been solved with AI Studio.

Another one of our early AI Studio initiatives was to work on Adore Me’s expansion into more marketplaces, which we call Adore Me Everywhere. The goal is to enable people to purchase Adore Me products on macy.com, target.com, Walmart, Amazon, and victoriasecret.com, our parent company. Each one of these marketplaces has very specific rules and requirements around their product description. Before Writer, someone on the merchandising team would have to take the product descriptions and edit them manually for each channel. This was a major limiting factor in terms of speed to market.

This was another example where the prompt coordinator and merchandiser worked together in the no-code side in AI Studio until we got to a place where the prompt was good and everyone was happy with the output for each marketplace. Then, we pushed that to a Writer Framework user-facing app so now anyone can go to the app, put in a product ID, and it automatically populates product descriptions for each of these channels.

Beyond that, you can publish to the Adore Me website directly from the Writer Framework app. The entire manual process now lives in a very clean, simple app where the prompt is a living, breathing thing that can be updated if something changes. The Writer Framework has a fully-customizable UI, so we can make it look however we want without even needing a front-end coder. That flexibility and level of customization is a huge plus.

We’ve also been focusing on creating quick apps that can help our teams with the more tedious, day-to-day work. One example of this is speaker bios for conferences, which as anyone who’s ever spoken or attended a conference knows, is a standard thing to request of the copywriting team. But for the copywriting team, it’s not necessarily the highest priority. So as a quick challenge, we decided to create an app. In the no-code platform, we identified what goes into a speaker bio‌ — ‌education, what you do, a fun fact, etc‌ — ‌and gave ten examples of good outputs in the prompt, and then created a front-end app and gave people access to it.

In the past, we would have never bothered doing anything like that because no one could justify the amount of work it would’ve taken to create that app. But now it’s exciting. The idea that anytime you start to notice a pattern or see a repetitive piece of work you can quickly and easily create‌ an app to solve that problem is one of the more exciting parts of AI Studio for me.

What impact have you seen from generative AI?

Without AI Studio, it would’ve taken us months to launch Adore Me Mexico. We’ve seen companies with repetitive content‌ — ‌like the 2,900 product descriptions‌ — ‌leave it in English and put the rest of the site in Spanish. What could’ve taken months took only ten days. It gives us confidence and establishes a process for when we think about additional markets to expand to. In the past, expanding into a new market was a very heavy lift because the content side alone would be a major investment. Now, we can approach international expansion with a strategic lever since we know the content side is much more straightforward.

The third-party marketplace descriptions app has allowed us to go from a 20-hour a-month process down to 20 minutes.  The time savings are so dramatic because it was such a heavy manual process in the past. The app saves time and frees someone up to work on higher-value work.

No-code allows us to take a prompt and make it a living, breathing thing. In the past, you’d create a prompt, pass it off to a developer, and they’d hard-code it into some script, get the outputs, review them, give another prompt, and so on. With this method, you’re not incentivized to iterate on the prompts too much.

What no-code allows us to do is take that exact process and do it in real time: easily update, add different types of structures and linked prompts with no technical experience. The prompts created in no-code dynamically populate whatever front-end app you create for a user, and it’s easy for anyone to make updates that’ll automatically populate that front-end app. It changes the way we think about prompts: in the past, it was a relatively static, one-and-done process, but now it’s ongoing. Each prompt starts to become an important asset within the company that it’s managed in, and completely changes what prompting means within the company.

How do you think about measuring the impact of generative AI?

Companies that solely have a high-level organizational ROI approach risk missing all these smaller use cases that can give you early wins that will generate the energy and expertise within or across the company that’ll then create the next big win and improve employee satisfaction.

Relieving somebody of writing 40 hours worth of product descriptions to only 30 minutes is a nice, measurable time saving, and I’m sure there is a quality of life metric associated with that. But what’s more interesting is that person being excited about AI, and then working on creating more solutions and incorporating it into the way they do their job. To me, right now in 2024, that’s the single biggest thing every company can do: get people understanding and excited about the technology. After that, the solutions and potential are limitless.

How has AI impacted the retail and eCommerce space?

In terms of exciting trends, there are two parts that we see. There’s the internal changes to the ways we do business, content creation, and anything else within the company. But for me, the more exciting part is around the personalization opportunities for customers. We’ve already been experimenting with using aggregated reviews and summarizing them in creative ways to target customers with different language and emails. Imagine your purchase history, preferences, and then not only getting recommended products, but the language that’s used is tailored to you and your interests. The way we as a company speak to the customer could change depending on factors like your age group, marital status, or geographic location. There are so many opportunities ‌to personalize and make ‌online shopping a more fun experience for customers in the next few years.

What were some important factors in rollout and adoption?

On the rollout side, people who have not worked with generative AI before can get discouraged if the output isn’t right the first time. One thing we always try to do is push the idea that the output isn’t bad, but the prompt was bad, and that you should keep working on the prompt until you get to the output that you want. Getting people to understand that it is an iterative process, and that it’s better to start from a brainstorming spot rather than trying to solve a process on the most creative, high-level marketing campaign tagline. When people on our team started on smaller projects like product descriptions, emails, or subject lines, it helped them get over that initial resistance.

My favorite part of AI Studio is how easy the no-code platform is for the actual end user. The importance of that really is that the end user will always be the one who will come up with the best solution to their problem and the right prompt, and it’ll be much more likely that they’ll be happy with the output. At a lot of companies, the team creating a prompt is separate from the end user, and it’s overall less investment and understanding of the problem. To have the end user involved in the entire process with the no-code side of AI Studio which makes it easy and less intimidating for people is incredibly powerful.

What advice do you have for other AI leaders who are earlier in their journey?

To get over fear and doubt surrounding generative AI, it’s crucial to have cultural and change management processes in place. This is why finding really specific use cases that people are happy to have solved should be the biggest priority if you’re at the early stage in your AI journey. Getting people to see the value that it will provide them so they can do more interesting, smart, and creative things is the biggest win you can get at an early stage.

What’s next for your team?

It’s been hard to contain our excitement since we’ve started working with AI Studio and have seen how in a matter of days or hours, depending on the complexity, we can generate and put into production an AI app that solves a specific problem. The roadmap that previously would have been outlined over months can now be outlined over weeks, and so it’s really a matter of digging into the problems we want to solve.

But even bigger than that is bringing more end users into the process, the same way we saw with translation and marketplace descriptions. We need to educate people on the power of prompting and how to use the no-code side, and how that can help create a whole front-end app that others in the company can use. The more people see the power of that, the more empowered they feel to take on even more problems. That will be the most fun and exciting part over the next few months.