With Gemini 1.5 Pro from Google Cloud, the accuracy of Gazelle’s output increased from less than 95% accuracy to 99.9%, while customer feedback also improved. Gazelle’s customers can now extract key information from lengthy documents, such as building-inspection reports and electricity reports, and summarize it in their preferred format. When performed manually, this process took up to four hours. Gazelle reduced this to around a minute with its previous gen AI provider. “Now, with Gemini, we can deliver new content in just 10 seconds,” says Gaddy. “And Gemini is around a fifth of the price of alternatives. Gemini wins across the board.”

Gemini has also helped to reduce Gazelle’s product development time, because the accuracy of its output means the company needs to spend less time controlling and correcting it. “With Gemini, we really amped up the speed and the frequency at which we can deliver a new product, because the quality-control cycle to make sure we hit our accuracy targets has shortened a lot,” explains Gindele. “Since we switched to Gemini last year, we’ve already launched four new products.”

Gazelle also uses BigQuery to analyze its tools’ performance and keep track of how customers are using them, helping to inform the development of new products. For example, Gazelle is developing a product that will use Gemini to allow customers to create area descriptions for the properties they’re selling. Brokers will enter the address of a given property, along with any relevant information about its location, and Gazelle will use Google Maps Platform to build a clear picture of the property’s surrounding area. This information will be sent to Gemini, which will ground its responses using Google Search to ensure all the data is correct. It will then automatically generate an accurate text description of the area. 

This tool will also use Gemini’s multimodal capabilities to analyze images of the property and the surrounding area and generate descriptive captions for those images to be used in ads to help sell the property. What’s more, Gemini also factors in text-based descriptions of the area to bring further context to the image captions it generates. The whole process of generating area descriptions and image captions will take just 20 seconds, saving brokers hours of research and writing time.

Gazelle’s use of Gemini is already having a profound impact on its customers’ processes. “Gazelle is helping us to work in a more efficient and productive way,” says Michelle Grennes Andreassen, System Manager at EiendomMegler 1 Oslo Akershus AS, a Gazelle customer. “We get tailor-made solutions to automate time-consuming tasks and this gives us greater precision and quality in our work.”

Gazelle’s growing reputation for saving businesses hours of labor is helping to open doors with some of Norway’s biggest realtors. “Previously, our gen AI output lacked accuracy,” says Gaddy. “Now, with the increased quality Gemini gives us, customers feel confident to introduce us to other regions and other businesses. These conversations wouldn’t be happening without the improved output from Gemini.”

Gazelle is now looking forward to the day when it can give real estate agents a complete suite of tools to automate every piece of content they need. “We plan to use Gemini to the maximum in the next 12 months to develop tools that cover every step of the real estate agent’s workflow, so that when someone even thinks about selling their property, the broker already has everything ready for them,” says Gindele. “Gemini allows us to think differently about the problems we need to solve. We’re confident it will continue to allow us to develop the innovative new products our customers need.”