Stay organized with collections Save and categorize content based on your preferences.

Deepdesk: Enabling support agents to improve the customer experience with AI technology

About Deepdesk

Deepdesk is a startup that provides AI assistance to contact center support agents to make their jobs easier. Its Agent Assist platform is trained with historical data from previous conversations to provide tailored, automated suggestions and the best answers to real-time customer queries on digital channels.

Industries: Technology
Location: The Netherlands

Tell us your challenge. We're here to help.

Contact us

Deepdesk was able to productionize its powerful AI tools to help its support agents have an improved work experience, helping both agents and customers alike.

Google Cloud results

  • Fully automated machine learning in production using Kubeflow Pipelines
  • Increases model training speed by up to 10x using GPUs
  • Offers autocomplete suggestions in under 100 milliseconds
  • Reduces time to onboard new clients from weeks to just two hours
  • Meets ISO27001 compliance powered by Vanta's Google Cloud integration

Deploys first Dutch GPT-2 model for agent assistance

When customers reach out to a company’s support staff to answer a pressing issue, mainly in times of distress or dissatisfaction, they expect helpful and immediate answers. This puts a lot of a brand’s value in the hands of its support agents. Imagine if these agents had a tool at their fingertips that could help them find tried and tested answers to pressing questions in real time. That’s where Deepdesk can help.

Giving people a better customer experience with Google Cloud

Deepdesk enables support agents to improve assistance by providing real-time solutions to customers' questions with its AI-powered platform, Agent Assist. It helps reduce the time agents spend searching for the best response to a customer query, with autocomplete suggestions appearing in under 100 milliseconds.

"Deepdesk learns from past conversations and processes this data with machine learning to predict what the agent can say to the customer during the entire conversation in real time," says Robbert Dijkstra, Chief Executive Officer, Deepdesk.

The platform is being further developed to process even more enterprise data in knowledge bases, FAQs, ERP, and CRM systems to supply agents with relevant answers, and relieve them from repetitive searching. As a result, Deepdesk’s platform elevates agent happiness and retention, taking it from being just an operational efficiency solution to one that benefits both agents and customers. But while Deepdesk has made inroads to revitalizing one of the most important communication channels for enterprises, its AI systems still had room for improvement, which is why it turned to Google Cloud for solutions.

"When we realized we needed more control over our architecture, Google Cloud was our first choice because of its developer friendliness and focus. Being a tech company, we needed technology that helps us deliver software."

Lukas Batteau, Co-Founder and Chief Technology Officer, Deepdesk

Creating the first Dutch GPT-2 model

When Deepdesk began operations in April 2019, it started off with a platform-as-a-service vendor that offered a simple and cost-effective solution, but lacked control. At a certain point, however, language modeling demands meant that Deepdesk engineers needed more control over its machine learning model production and deployments.

"When we realized we needed more control over our architecture, Google Cloud was our first choice because of its developer friendliness and focus. Being a tech company, we needed technology that helps us deliver software," says Lukas Batteau, Co-Founder and Chief Technology Officer, Deepdesk.

The company participated in the Google Cloud for Startups program in the fall of 2019, while also receiving $100,000 worth of credits for Google Cloud, which helped it get started.

Since Deepdesk’s architecture relies heavily on Kubernetes, it made sense for the startup to migrate to Google Kubernetes Engine (GKE). As the company increased automation during the migration process, its DevOps specialists have said that the intuitive interface of GKE made it easy to find everything they need. Batteau reveals that the company was able to be self-reliant during migration because many GKE components and services are well documented.

Furthermore, personal support from Google Cloud helped the Deepdesk team better leverage Google Cloud tools. "The personal relationship with the Google Cloud support team has been really good," Batteau says. "To think we were able to build a company without any hardware assets, and yet we managed to create the first Generative Pre-trained Transformer 2 (GPT-2) Dutch language model. This couldn't have been achieved without Google Cloud and the starter credits."

For their workflow orchestration, Deepdesk's engineers used the AI tools on Google Cloud's AI Platform to help scale its machine learning. The engineering team specifically used Kubeflow Pipelines, a Kubernetes-native tool that brings orderliness to machine learning code development and cloud scaling for automated model training and deployment. As a result, Deepdesk's AI tools could be made both language-and platform-agnostic to deliver real-time assistance to support agents.

"That's the power of Google Cloud, to have control over so much with just a small team of developers, while being able to deliver enterprise-level software security, data compliance, and isolated tenant environments."

Lukas Batteau, Co-Founder and Chief Technology Officer, Deepdesk

Tailoring the best support assistance for global clients

The migration to Google Cloud has helped the company accelerate technology deployment for all its customers. Previously, it used to take Deepdesk weeks to launch a new customer. Today, Google Cloud's automatic provisioning capabilities help to reduce deployment time to just two hours.

"The fact that we're provisioning, deploying, and maintaining software for these large enterprise customers with a relatively small team of dedicated developers is incredible. If you just look at the number of components being rolled out every day and autoscaling everything that's happening, it's just a Milky Way of components that are working together," Batteau explains. "That’s the power of Google Cloud, to have control over so much with just a few people, while being able to deliver enterprise-level software security, data compliance, and isolated tenant environments."

"To think we were able to build a company without any hardware assets, and yet we managed to create the first Generative Pre-trained Transformer 2 (GPT-2) Dutch language model. This couldn't have been achieved without Google Cloud and the starter credits."

Robbert Dijkstra, Chief Executive Officer, Deepdesk

Every Deepdesk deployment is unique and private, with each customer implementation starting with a new AI model that is trained with Deepdesk's own Machine Learning and AI technology using the client's specific data. Deepdesk does not share any information between models, given that data governance is central to many customers' considerations.

"Having a data center in the Netherlands was another reason we chose Google Cloud, because we needed to assure our local customers that their data doesn’t leave the country. We can flexibly extend this worldwide, localizing data storage and using it in applications as we provide our infrastructure to our global clients," Dijkstra says.

As the Deepdesk team looks to reimagine the future of customer experience, it is developing ways to move from text-based to speech-based assistance, which are substantially more compute-intensive. Tools like Looker and Vertex AI are currently being evaluated for future implementation.

"With Google Cloud technology on hand, you can ramp it up and go ahead and build great stuff. Then creativity can just flow freely," Dijkstra says.

Tell us your challenge. We're here to help.

Contact us

About Deepdesk

Deepdesk is a startup that provides AI assistance to contact center support agents to make their jobs easier. Its Agent Assist platform is trained with historical data from previous conversations to provide tailored, automated suggestions and the best answers to real-time customer queries on digital channels.

Industries: Technology
Location: The Netherlands