Gelato, a Norwegian software company that brings local production to global ecommerce, saved time and improved performance with Gemini 1.5 Pro, and increased scalability and product quality.
Gelato, a Norwegian software company that brings local production to global ecommerce, saved time and improved performance with Gemini 1.5 Pro, and increased scalability and product quality.
Gelato brings local production to global ecommerce. Through its network of more than 140 printers and logistics partners in 32 countries, Gelato allows creators to reach customers anywhere. At the same time, its end-to-end production software enables local print producers to capture the growth opportunities that ecommerce presents and ensure efficiency and profitability in a new era of print production.
The Gelato model reduces cost, transportation distances, carbon emissions, and waste caused by overproduction, while empowering both makers and creators to scale their businesses. The result is a new way of manufacturing and distributing products that is better for people and the planet.
Gelato invests in the latest cloud technologies, as well as AI and ML, to optimize every area of its business. However, Gelato’s previous ML stack was time-consuming to maintain and needed considerable expertise to operate. This limited the ability of Gelato’s small ML team to experiment quickly with ML to improve internal operations.
As extensive users of BigQuery and Looker, the Gelato team spoke to Google Cloud, which recommended using Vertex AI. “As a fast-moving company with a small team, long ML development times are unacceptable,” explains Nicola Croon, Senior Machine Learning Engineer at Gelato. “We quickly understood the power of Vertex AI. It's easy to manage and really streamlined our process. Previously, a customer-lifetime-value prediction model took us two weeks to put into production. With Vertex AI, it would take us just one or two days.”
We quickly understood the power of Vertex AI. It's easy to manage and really streamlined our process. Previously, a customer-lifetime-value prediction model took us two weeks to put into production. With Vertex AI, it would take us just one or two days.
Nicola Croon
Senior Machine Learning Engineer, Gelato
Saving 120 hours a week triaging support tickets with generative AI
With Vertex AI, Gelato gained access to Gemini models including Gemini 1.5 Pro. With all its ML and AI capabilities in one place, Gelato worked closely with the Google Cloud team to define areas that would benefit from Gemini.
Gemini 1.5 Pro stood out for its uniquely large context window of up to 2 million tokens. That's huge, and it was critical for us, given the complexity of our triage process. Gemini is also extremely fast and I love working with it. It was an obvious choice.
Nicola Croon
Senior Machine Learning Engineer, Gelato
A key area Gelato identified was its engineering-support ticket-triage process. Gelato’s engineering department consists of 15 teams, with many overlapping responsibilities. Every time a bug is reported, Gelato needs to determine which engineering team is responsible. Previously, this triage process took over 100 hours weekly. This process limited Gelato’s scalability and led to inaccuracies, with just 60% of tickets correctly assigned. During periods of peak demand, it also acted as a bottleneck, slowing the resolution of tickets, with potentially damaging consequences for creators, such as lost revenue.
To solve this issue, Gelato trained Gemini 1.5 Pro to understand the triage process. Now, as soon as a ticket is reported, Gemini automatically assigns the ticket to the correct engineering team, allowing it to begin resolving the issue immediately.
Gelato has saved 120 hours of weekly labor as a result, meaning it no longer needs to assign dedicated resources to triage. Accuracy has increased too, with 90% of tickets now assigned to the correct team.
“As a fast-growing company, addressing customer problems effectively is critical,” says Croon. “With Gemini we can resolve tickets more quickly and increase customer satisfaction. This helps us win more customers, further reducing emissions and waste from overproduction.”
For Gelato, Gemini was the natural choice due to its large context window, which allowed the team to train the model on a huge volume of data. “Gemini 1.5 Pro stood out for its uniquely large context window of up to 2 million tokens. That's huge, and it was critical for us, given the complexity of our triage process,” Croon explains. “Gemini is also extremely fast and I love working with it. It was an obvious choice.”
Improving performance across the platform with accurate error categorization
Gelato also uses Gemini as part of its error-categorization process. When an end consumer raises an issue with Gelato, the customer support team needs to categorize that error accurately, both to resolve it quickly and to give the company visibility of its performance. However, with more than 120 categories, Gelato needed to spend a lot of time training its customer support team to categorize errors accurately.
To solve this issue, Gelato embedded Gemini in its customer-support systems. Now, when an error is reported, Gemini classifies it instantly. This allows errors to be resolved more quickly, while reducing the time required to train new customer-support agents. Categorization accuracy has also increased, allowing Gelato to understand exactly how its platform and network are performing.
“Quality is non-negotiable for Gelato, and we never forget the customer eagerly awaiting their product,” says Croon. “That’s why error categorization’s so important. With Gemini we have better visibility of issues and can work proactively with our partners to bring error rates down, while understanding where we can improve internally.”
Simplifying future challenges with AI and ML
Gelato now plans to use Gemini 2.0 to further increase efficiency and scalability as it pursues its mission of redefining global production to benefit customers and the planet alike. In particular, Croon intends to use the image-recognition capabilities of Gemini 2.0 to control product quality before items are shipped, helping to raise standards along with customer satisfaction.
For Croon, this is yet another milestone along Gelato’s journey with Google Cloud and AI. “I used to be narrow-minded about what AI can do,” Croon recalls. “Google Cloud pushed us to be the better, more innovative version of ourselves. Now, we see every problem as a challenge to solve with Google Cloud AI and ML.”
I used to be narrow minded about what AI can do. Google Cloud pushed us to be the better, more innovative version of ourselves. Now, we see every problem as a challenge to solve with Google Cloud AI and ML.
Nicola Croon
Senior Machine Learning Engineer, Gelato
Gelato is the fast, smart, green one-stop-shop for customized print products on demand. Its solution enables entrepreneurs, creators, and global brands to sell their products globally and produce them locally in 30 countries, reaching up to 5 billion potential consumers overnight.
Industry: Technology
Location: Norway
