Brian Kolodny, a seasoned business consultant with deep expertise in AI solutions, has led the development of new customer experiences for Toyota, Universal Theme Parks, Jack-in-the-Box quick-serve restaurants, and even the Royal Family’s mental health foundation.
All in all, he’s led 37 engagements for Global 100 companies – across automotive, financial services, energy, entertainment, hospitality, pharmaceuticals, telecommunications, and planned communities – and foundations. With thousands of companies looking to AI virtual assistants for efficiency, customer service, and sales, we sat down with him to hear his best tips for creating one.
Building Toyota’s first chatbot – customer service and cost reduction
Brian’s Toyota chatbot is a good example of one way to get started with AI automation: Choose one, manually-intensive process to focus on. Not only does this serve as a pilot that demonstrates the value of chatbots to the business, it serves as the foundation for further automation.
"Our project with Toyota was their initial entry into the world of virtual assistants or chatbots.”
In Brian’s words, Toyota was spending too much time and valuable company money having humans answer the same questions regarding basic things – like scheduling service. The problem is that FAQs are broad, they can’t answer more specific questions, which ultimately drives customers to call a live agent.
A chatbot is the perfect go-between – it reduces the time and costs of having a person answer the same questions over and over again while allowing a customer to get as granular as they need to.
To start, Brian’s team focused on defining the vision for the Toyota virtual assistant. “We had to look at the entirety of being a prospect, becoming an owner, and living with a Toyota vehicle to understand the breadth of use cases the solution would have to manage.” This essentially future-proofed the chatbot, allowing it to one day become omni-channel (aka a chatbot in your car).
The next step is to define the phases. To get an initial chatbot off the ground, your team needs to crawl/walk/run towards the final vision. Figuring out what’s in each phase, comes down to mapping out the high-level functional and technical architectures and what data sets and capabilities are included in each.
Together, they built the logic tree of customer experiences a chatbot should be able to handle without the need for human intervention. If a customer needs to book an oil change, the chatbot needs to be able to ask if they’ve been to a certain location before, if they’ve been serviced in the past etc.
"You have to lay out the questions, and the questions that follow those questions,” he said.
Building the ‘happy path’ logic is the easy step, because that’s what website FAQs do: question, answer, question, answer, etc. The next step is to train the chatbot to handle every variation of question a customer might ask, not just the predictable ones.
The first part of that is extending the chatbot's vocabulary so it understands that there are multiple ways to ask the same question – and therefore multiple ways to get to the same endpoint. “It’s generally pretty iterative because you have to interact with it and see where it answered awkwardly or clearly didn't understand the input.”
The second part of this step is seeing where it only answered part of a question. These are the places where users will get stuck and a person will have to be looped in.
“You're trying to automate this series of interactions so that you don't need the human in the loop. Otherwise, you really didn't create the efficiency that you're trying to create.”
*As a note here, this chatbot was developed in 2016 - 2017 and may look different today or include new features.
Delivering an end-to-end experience for Universal Theme Parks
The other way companies approach AI automation is by looking at the customer journey as a whole and implementing a solution that cuts across every major phase of the customer experience.
“With Universal, it was done at the entire experience level,” Brian said. "We focused on the overall experience – from booking, to planning the vacation, to being in the park, and after the visit. Think Waze combined with Jarvis from Ironman."
Universal has multiple worlds within their parks, all with rides, shows, dining options, and gift stores – which means you can’t just show up and experience everything. It requires a lot of planning, which is the customer pain point Brian’s team solved for.
They started with scoping the entertainment options available. “There are 10 lands, and each world has several different things you can do in it. If there are four people in the family, there are hundreds of different things that people could want to do. The critical question is how do you combine all these preferences into an easy to navigate path that spans multiple days.”
After laying out every activity, store, and dining option within the logic tree, Brian’s team built a chatbot that could act as an agent and gather the preferences of an entire family.
Based on the length of someone’s stay, their preferences, and the locations of each activity in the park, the chatbot puts together a streamlined itinerary – then uses your geolocation data and real-time park data – such as volume of customers on each path, ride wait times, weather etc. – to create a custom itinerary.
As you're walking through the park, your AI agent is notifying you:
- 'Hey, you wanted to go on the Transformers ride. It's currently an hour and a half wait. You should head over now, if you want to stick to your current itinerary'
- ‘It’s 11:30 am and you haven’t eaten yet. Your family is heading to the Wizarding World of Harry Potter, do you want me to order you the following items from your list of preferences so that you can pick it up when you arrive?’
- ‘Your kids wanted to buy these three things from Nintendo World. If you come to the central store in the next 30 minutes, we will have those items ready for you to review.’
The real power of the solution is in real-time alternate pathway development. If you choose to deviate from your itinerary, it can use your preferences and location to suggest alternatives and then, just like Waze, create a new path for you.
*As a note here, this chatbot was demonstrated in the park in 2017 for Universal executives and then handed over to their existing development team to implement.
Brian’s top 3 takeaways for chatbot design
1. Start with the end goal in mind
The difference between Toyota and Universal’s chatbot design demonstrates the importance of deciding how large or small scale you want it to be. Are you automating one frustrating process, or are you overhauling the entire customer experience?
AI chatbots are a good fit for “end goals” that:
- Automate an existing manual process
- Are currently a friction point for customers
- Will allow your team to serve more customers
"Vision is key because whatever else you build into the chatbot needs to be supported by the framework you initially built,” Brian explained. “How are you building things out, in what order, and what’s the priority?”
Brian explained this phasing process in the way you peel an onion: Start with what it could eventually be, and then peel back the layers to get to what you build first. Crawl, walk, run.
2. Chatbots should be living things
Iterative development is key in ensuring a chatbot continues to improve customer experience.
"Building a chatbot involves continuous testing, feedback, and refinement to ensure it meets user needs effectively,” Brian said.
This means your chatbot should:
- Be fed new data regularly
- Continue to be tested with edge cases
- Have its logic updated routinely
In testing and iterating, Brian encourages you to focus on places where users get “stuck” with the chatbot and have to resort to a human, where the chatbot provides awkward answers, or areas where the chatbot doesn’t seem to fully understand the input.
3. Don’t erase people from the equation
As Brian describes it, chatbots have “walls”. There’s a reasonable limit of what a chatbot is going to be able to do before a human needs to get involved.
So design with intention when considering when a human needs to step in. And beyond that, be mindful of the fact that freeing up capacity could mean entire jobs.
“You don’t just need to ask ‘what will the chatbot do?’ but also ‘what happens with the people?’”
Want to hear more about how Brian built personalized ordering for Jack in the Box and a mental health bot for the Royal Family? Subscribe to our Strategy Brief for part 2.
