Recipes for Building Conversational AI Teams

Building AI assistants requires a blend of skills: software engineering, deep knowledge of the user, writing and UX design, and data science.  Product teams that build conversational AI software share some similarities with traditional software teams, but there are also important differences. You’ll need a solid background in backend development and DevOps, but also expertise in conversation design and machine learning.

Whether your organization is large or small, building an AI assistant for the first time or maintaining an existing assistant, you want to assemble a product team with the right mix of skills and the right size for the project.

But there’s a remarkable amount of variety when it comes to conversational AI teams. Finding the right formula comes down to an organization’s priorities, stage of development, and size. In this blog post, we’ll start by breaking down a few of the most common roles that make up conversational AI teams. Then, we’ll consider a few example team structures:

  • A lean team, fast time to market
  • A medium team that prioritizes media-rich customer experience
  • A large team, post-launch

Roles and Skill Sets

Let’s start by looking at seven roles commonly found on conversational AI teams. Keep in mind that conversational AI is a new and rapidly growing field, and this doesn’t represent every possible title. There are specialized roles like AI Innovation Leads and computational linguists, as well as smaller teams where a single developer might take on both data science and conversation design responsibilities.

Backend Developer

On nearly every conversational AI team you’ll find at least one backend developer. Backend developers are responsible for building the application around a conversational AI framework, including integrations with backend systems and databases. They may also build out APIs to allow the assistant to communicate with existing internal applications and data sources.

Frontend Developer

Frontend developers integrate the AI assistant application with a user interface, like a website or mobile app. Many conversational AI teams bring on a frontend developer during the initial buildout, but don’t maintain a full-time front end person after launch, as most ongoing development and maintenance happens on the back end.

Data Scientist

Data scientists are responsible for training and optimizing machine learning models. On a team building AI assistants, a data scientist needs a deep understanding of algorithms and natural language processing. Their primary touchpoints on the project will be training data, NLU pipeline configuration and hyperparameter tuning, and testing tools to measure model accuracy. In some organizations, data scientists may do data analysis as well.

Data Analyst/Data Engineer

Teams that place a high value on measuring an AI assistant’s impact on the business may include a data analyst or data engineer. A data analyst is responsible for analyzing usage metrics to uncover patterns and inform business and development decisions. They’re often in charge of structuring data and creating reports to identify where the AI assistant can be improved. A data engineer handles the more technical aspects of setting up and maintaining this data pipeline.

Conversation Designer/Copywriter

A conversation designer is like a UX designer for conversational interfaces. They use their insight into the user and business needs to create conversation flows for the tasks a user wants to complete. In some cases, they may also be responsible for the look and feel of an assistant, including crafting a personality for the assistant and writing conversation scripts. In other cases, creating the assistant’s voice may be supported by a dedicated copywriter.

Product Manager

Product managers act as a bridge between business stakeholders and the development team. They’re responsible for translating business requirements into features and prioritizing where feature development sits on the roadmap. On smaller teams, the product manager may also act as a project manager, managing the backlog and flow of tickets to the development team, or a conversation designer, mapping out conversation flows.

DevOps Engineer

The rest of the team’s hard work doesn’t mean much if it can’t be reliably deployed to production. A DevOps engineer is responsible for setting up the AI assistant’s hosting infrastructure, including making sure the architecture plays well with an organization’s existing security and network requirements. DevOps engineers are also often in charge of building automated processes to test and ship code updates to production.

Example Team Structures

We’ve looked at a few of the individual roles that make up conversational AI teams; now let’s see how they fit together. In this section, we’ll walk through a few variations of what a conversational team might look like, based on criteria like project priorities and development stage.

Lean team, fast time-to-market.

For our first team, let’s imagine that we have a short launch deadline and limited headcount to get the AI assistant off the ground. We’re assuming that the project is in early stages and will move quickly from POC to production. We also expect the assistant to see heavy traffic soon after launch, and high availability is required.

In this case, the team is likely to prioritize the technical aspects of the implementation, making sure the infrastructure is sound, rather than focusing on copywriting and experience design. We’ll call this our lean team: the minimum team needed to get a functioning AI assistant into the hands of users.

  • 1 Backend Developer
  • 1 Data Scientist
  • 1 Product Manager
  • 1 DevOps Engineer

Here we’ve added one backend developer and one data scientist, who would work together to build the AI assistant as well as the application that surrounds it. We’re skipping a dedicated frontend developer in this case, under the assumption that the front end requirements are not complex. Our product manager is likely multitasking: collecting business requirements as well as designing conversation flows and analyzing the assistant’s performance. And lastly, we’ve added a DevOps engineer to manage infrastructure and uptime.

Medium team, creating a rich customer experience.

Our second team has more headcount but different success criteria for their assistant. Here, the organization has built their brand around media-rich storytelling, which extends to the digital experiences they create for customers. This conversational AI team is focused on crafting the voice and tone of the assistant, and they have a longer timeline from POC to launch.

  • 1 Product Manager
  • 1 Copywriter
  • 2 Conversation Designers
  • 2 Backend Engineers
  • 1 Data Scientist
  • 1 DevOps Engineer

This team is heavily weighted toward content creators—only about half the team is working on the technical implementation. The other half is focused on designing the experience of the assistant. We’re still including a DevOps engineer to oversee the infrastructure and deployment.

Large team, post-launch.

Teams within a large organization often evolve depending on the assistant’s development phase. Some teams bring on extra members leading up to launch and scale back afterwards during the maintenance phase. Others start with a lean team early on and add team members as development progresses.

Here, we’re imagining a team within a large organization who have launched their assistant within the last six months. Although maintenance takes up a large portion of the team’s attention, they are also working on a second iteration of the assistant, which will include new features. Because of this, this team has kept many of the members they brought on leading up to launch, and have added a few more team members focused on understanding usage patterns and business metrics.

  • 1 Product Manager
  • 1 Conversation Designer
  • 4 Backend Developers
  • 1 Data Scientist
  • 2 Data Analysts
  • 1 DevOps Engineer

Conclusion

Assembling a team for building AI assistants can look very different depending on the organization and project. In our experience at Rasa, the most successful teams reflect the project’s goals and priorities, whether it’s creating a distinctive brand experience or going to market quickly on a solid technical foundation. What they have in common, however, is a blend of skills that include software engineering, machine learning, UX, and DevOps.

What team structures have you seen in your experience? Share your thoughts and continue the conversation in the Rasa forum.