1. Get organized and plan for the long-term

The first critical step to ensuring the success of any project—especially a voice project—begins with organization and planning. Good planning includes determining parameters around how internal and external teams will collaborate, and setting schedules with goals and deadlines for each phase of the project. 

I recommend planning regular check in meetings, setting up sprint planning meetings, and developing milestones that have some fluidity to them. The more planning and effort you put into communication internally and with external teams, the better the expected outcomes will be.

If you use agile, it’s important to get onto a sprint cycle so that you can develop a cadence and understand that this is not going to be a short term endeavor. Most voice service projects take a year or more and shouldn’t be considered in weeks. Although there are some voice services that you might be able to get done in a couple months, those would be very limited in scope and maturity in terms of user experience.

As companies start the conversation with their internal engineering teams, it’s important to realize that putting together a proper voice assistant takes a tremendous amount of time. It’s taken several years, if not decades, for voice technology platform providers to get to the point where they’re able to offer voice AI services to other companies. Thinking that your internal teams can accomplish the same goal in a matter of months is not realistic.

Before you embark on any engineering effort, take the time to look closely at what is going to be involved and what resources and outside support you’ll need to develop a robust voice assistant. Spend a month or so in the planning stage. Determine what sort of engineering efforts are going to be needed internally and how your teams will interact with the tools available from the voice AI platform provider.

2. Understand your customers and use cases

When first deciding on the voice user interface (VUI) for your products, I think it’s important to understand the use cases for the voice assistant. Begin by understanding how your customers interact with the product, service, or app you intend to voice-enable, and drill down as to the behavior they currently exhibit when they’re addressing your product.

It’s important to understand your core use cases, build the voice user interface around them, and then use them to set goals and identify requirements. Ultimately, a KPI report will be developed from the use cases to track the iteration and improvement cycles of the voice assistant.

Of course, when you first expose the voice user interface to users, you’ll also need to provide guidance and ongoing education. User education is a key element to the initial adoption and continued engagement. Afterall, if your users don’t know what your product can or can’t do, they may abandon the voice interface and your product in frustration.

A lot of people are still getting used to using voice assistants and the more guidance you can give them, the happier they’re going to be interacting with that interface. Once your voice assistant has been in the market for a while, you will want to go back and look at your user data to understand what functions your customers are and aren’t using. Then, you can make improvements and continue to iterate.

3. Make improvements and learn from regressions

Developing a voice assistant is never a one and done project. Even voice user interfaces that have been in the market for the last couple of years are going through regular rounds of iterations based on user data. While many of those iterations will happen during the development phase, teams need to plan for ongoing maintenance and changes to meet user needs. 

It’s also important to note that during the development phase, not every change is going to result in an improvement. Oftentimes. injecting new data into the models for training purposes produces regressions. It’s important to understand anytime a regression occurs and to understand how the data is affecting the training of the models.

Engineering teams are often surprised that new data doesn’t always result in improvements. Make sure your teams understand that you will see regressions and accept that it’s part of the learning and development process. Over time, regressions get smaller and smaller as teams move towards a specific goal. However, the process can be a bit of a bumpy road, so be prepared to roll with the ups and downs of voice development.

Putting time into planning and establishing clear lines of communication will significantly mitigate misunderstandings around the development process and keep the project moving forward. Setting expectations that there will be setbacks and regressions up front will help teams to communicate these events as normal stages in the development process.

4. Use data not opinion to guide voice assistant development

It’s also important to understand that new processes are going to have to be created in order to support your voice project. It’s important to gather and use the most accurate and detailed information about user behavior and product functionality. 

If you begin with information that is imperfect, it makes the process of creating an effective voice assistant a lot more difficult. If, for example, you are creating a voice user interface that relies on movie titles and you start with movie titles that are spelled incorrectly or were abbreviated, the voice user interface will never meet customer expectations. Any information that isn’t high quality will affect the output and result in inaccurate responses.

Collecting accurate and timely data is key to every stage of building and delivering a custom voice assistant. Brands can get the most value out of their data by understanding how users are interacting with the voice assistant, identifying gaps, and then making the modifications that will improve the voice interface over time.