• Depop's new gen-AI feature creates item descriptions based on photos that users upload.
• The tool has boosted the number of listings on the company's website and saves sellers time.
• This article is part of "CXO AI Playbook" — straight talk from business leaders on how they're testing and using AI.

Depop is an online fashion marketplace where users can buy and sell secondhand clothing, accessories, and other products. Founded in 2011, the company is headquartered in London and has 35 million registered users. It was acquired by Etsy, an online marketplace, in 2021.

Situation analysis: What problem was the company trying to solve?

Depop's business model encourages consumers to "participate in the circular economy rather than buying new," Rafe Colburn, its chief product and technology officer, told Business Insider. However, listing items to sell on the website and finding products to buy take time and effort, which he said can be a barrier to using Depop.

"By reducing that effort, we can make resale more accessible to busy people," he said.

To improve user experience, Depop has unveiled several features powered by artificial intelligence and machine learning, including pricing guidance to help sellers list items more quickly and personalized algorithms to help buyers identify trends and receive product recommendations.

In September, Depop launched a description-generation feature using image recognition and generative AI. The tool automatically creates a description for an item once sellers upload a product image to the platform.

"What we've tried to do is make it so that once people have photographed and uploaded their items, very little effort is required to complete their listing," Colburn said. He added that the AI description generator is especially useful for new sellers who aren't as familiar with listing on Depop.

Key staff and stakeholders

The AI description-generation feature was developed in-house by Depop's data science team, which trained large language models to create it. The team worked closely with product managers.

Colburn said that in 2022, the company moved its data science team from the engineering group to the product side of the business, which has enabled Depop to release features more quickly.

AI in action

To use the description generator, sellers upload an image of the item they want to list to the Depop platform and click a "generate description" button. Using image recognition and gen AI, the system generates a product description and populates item-attribute fields on the listing page, including category, subcategory, color, and brand.

The technology incorporates relevant hashtags and colloquial language to appeal to buyers, Colburn said. "We've done a lot of prompt engineering and fine-tuning to make sure that the tone and style of the descriptions that are generated really fit the norms of Depop," he added.

Sellers can use the generated description as is or adjust it. Even if they modify descriptions, sellers still save time compared to starting with "an empty box to work with," Colburn said.

Did it work, and how did leaders know?

Depop has about 180,000 new listings every day. Since rolling out the AI-powered description generation in September, the company has seen "a real uplift in listings created, listing time, and completeness of listings," Colburn said. However, as the tool was launched recently, a company spokesperson said that specific data was not yet available.

"Aside from the direct user benefits in terms of efficiency and listing quality, we have also really demonstrated to ourselves that users value features that use generative AI to reduce effort on their end," Colburn said.

Ultimately, Depop wants sellers to list more items, and the company's goal is to make it easier to do so, he added. Automating the process with AI means sellers can list items quicker, which Colburn said would create a more robust inventory on the platform, lead to more sales, and boost the secondhand market.

What's next?

Colburn said Depop continues to look for ways to apply AI to address users' needs.

For example, taking high-quality photos of items is another challenge for sellers. It's labor-intensive but important, as listings with multiple high-quality photos of garments are more likely to sell. He said Depop was exploring ways to make this easier and enhance image quality with AI.

A challenge for buyers is sometimes finding items that fit. Depop is also looking into how AI can help shoppers feel more confident that the clothing they purchase will fit so that their overall satisfaction with the platform will be enhanced, Colburn said.

• Siemens uses AI to tackle industrial challenges like safety and workforce shortages.
• Siemens says its AI tools, such as Senseye, boost productivity and reduce costs for global clients.
• This article is part of "CXO AI Playbook" — straight talk from business leaders on how they're testing and using AI.

Siemens is a German technology company that operates in many sectors, including industry, infrastructure, transportation, and healthcare. It has about 320,000 employees worldwide.

Situation analysis: What problem was the company trying to solve?

The industrial sector faces several challenges, including security and safety regulations, environmental sustainability, and a shortage of skilled experts. Peter Koerte, Siemens' chief technology officer and chief strategy officer, said the company aims to solve many of these issues with artificial intelligence.

"What's most important for AI is that in the industrial context, it needs to be safe, it needs to be reliable, and it needs to be trustworthy," he told Business Insider. Siemens, which has been investing in AI for about 50 years, offers several industrial AI products that help manufacturers across industries, such as automotive and aerospace, to predict maintenance issues and improve worker productivity using data.

"We believe if we can take data from the real world, simulate it, understand it in the digital world, we can be much faster for our customers, and our customers can be more competitive, more resilient, and more sustainable," Koerte said.

Key staff and stakeholders

Koerte said Siemens works with a number of tech partners on its industrial AI products and services, including Google, Microsoft, Nvidia, Amazon Web Services, and Meta. The company has about 1,500 employees with AI expertise who work closely with these tech companies, and Siemens' internal product development teams are also involved.

AI in action

Siemens' industrial AI work focuses on predictive maintenance, technology to assist workers, and generative product design.

One product is Senseye Predictive Maintenance, a tool that integrates with a manufacturer's data sources and uses AI to analyze the information. The company said the platform provides insights into how well machinery, tools, and other infrastructure are running. The tech can also help predict maintenance issues, which increases productivity and helps companies speed up the adoption of technology across their businesses.

Recently, Siemens debuted Industrial Copilot, a generative AI-powered assistant for engineers in industrial environments. The assistant can generate code automatically, identify problems quickly, and provide advice to support engineering tasks, such as troubleshooting equipment maintenance. The company said the tool can boost "human-machine collaboration" and enable companies to address workforce shortages while staying competitive.

Koerte said that when Industrial Copilot notifies a worker of an issue with equipment or software, that employee can use verbal commands in any language to create a work order, which is automatically sent to a team in a different country to take action to solve the issue. "AI breaks down barriers and democratizes many of the technologies because we take the complexity out of them," he said.

Did it work, and how did leaders know?

Siemens found that companies using Senseye Predictive Maintenance have reduced maintenance costs by 40%, increased maintenance staff productivity by 55%, and decreased the amount of time a machine is unavailable for maintenance by 50%.

The Australian steel company BlueScope implemented the predictive maintenance platform in 2021 to minimize downtime across its plants, increase operating time, improve the rate at which it can produce products, and lower costs. Together, Senseye and BlueScope's IoT sensors can detect abnormal vibrations in equipment early, preventing maintenance problems and saving the company money.

Schaeffler Group, a German automotive and industrial supplier, augmented a production machine with Industrial Copilot. Its engineers are now able to generate code faster for programmable logic controllers, the devices that control machines in factories. Siemens said the technology is helping Schaeffler Group automate repetitive tasks, reduce errors, and free up engineers for "higher-value work."

What's next?

Koerte said Siemens continues to research and develop new use cases for AI.

The company is working on a project that feeds computer-aided design data, such as models and digital drawings, into large language models and prompts it to create products.

The project is still in the early stages of development, but Koerte said it could enable design engineers, particularly in the automotive sector, to create more product variations and produce higher-quality items faster.

• Black Duck Software uses AI to speed up sending security advisories to customers.
• It says that with AI it can send out about five times its usual number of notifications a month.
• This article is part of "CXO AI Playbook" — straight talk from business leaders on how they're testing and using AI.

For "CXO AI Playbook," Business Insider takes a look at mini case studies about AI adoption across industries, company sizes, and technology DNA. We've asked each of the featured companies to tell us about the problems they're trying to solve with AI, who's making these decisions internally, and their vision for using AI in the future.

Black Duck Software, formerly Synopsys Software Integrity Group, offers security products and services — including security testing, audits, and risk assessments — to help companies protect their software. Black Duck is headquartered in Burlington, Massachusetts, and has about 2,000 employees.

Situation analysis: What problem was the company trying to solve?

Beth Linker, a senior director of product management for AI and static application security testing at Black Duck, said the company had been using artificial intelligence internally for several years but recently began developing the tech for its customers.

The company sends Black Duck Security Advisories, or BDSAs, to notify users that their software is at risk and potentially exploitable. Linker said this spring Black Duck started using generative AI to send BDSAs faster so that customers could act swiftly to address issues.

The need for speedier BDSAs arose after the National Vulnerability Database, a government cybersecurity resource that provides information on data threats, started publishing fewer vulnerability reports because of a backlog. At the same time, Linker said, the Linux kernel, an open-source operating system, began flagging more risks, significantly increasing the number of vulnerabilities it disclosed.

"The net effect was that all of a sudden you had a much larger number of vulnerabilities and less support from the National Vulnerability Database," Linker said. "This is something that was making things a lot harder for our customers because they were not able to get all the info that they were used to receiving."

Key staff and partners

Linker said Black Duck's engineering and research teams were involved in integrating gen AI with BDSAs. The system also uses some commercially available large language models.

AI in action

Linker said that accelerating BDSA delivery with gen AI was an opportunity to provide customers with a "timely and comprehensive feed of data that they need to make decisions."

To speed up BDSAs, Black Duck developed prompts, which they input into commercial LLMs, to query their internal data. This information is used to compile the advisory reports. Previously, this process was done manually.

A researcher reviews each AI-produced report before it's sent to customers. "Hallucinations are a risk," Linker said, "and everything we put in front of our customers has to meet a certain standard of quality."

Once BDSAs are created, the research teams review the reports and provide analysis and context about the seriousness of an identified vulnerability. This helps customers make decisions about the risk: Some vulnerabilities may need immediate attention, while others are less serious and could be fixed during a planned software update.

Did it work, and how did leaders know?

Linker said that more than 5,200 BDSAs were created with AI from March to October and that the company could now send out about five times the number of notifications each month that it could send before the tech was rolled out.

"We've been able to really scale this up to meet the need," they said.

What's next?

Black Duck recently unveiled Polaris Assist, an AI-powered security assistant. This new addition to the platform will help customers' security and development teams work more efficiently. It combines the company's existing application security tools with LLMs to give automated summaries of detected vulnerabilities and suggestions for how to fix the code.

"It's still a work in progress," Linker said. Polaris Assist is in beta testing, which is likely to wrap up by the end of the year.

They added that Black Duck continues to invest in AI to serve its customers. "A lot of that boils down to how can we make application security testing and remediation easier, faster, and more scalable?" they said.