May 19, 2023

Can you build a voicebot POC in 2 days?

Learn how our implementation team prepped and coached MINDD to get results in 48 hours.

You bet your bottom dollar we can! Keep reading to learn how our implementation team prepped and coached the client to get results in 48 hours.

The backstory of the 48-hour voicebot challenge

We’re not strangers to tight deadlines or mission-impossible requests, but we had to hustle when we got this particular phone call the last week of March.

Moet Ik Naar De Dokter (MINDD) is a Dutch self-triage system that helps citizens figure out in less than a minute whether they should go to a doctor (general practitioner). Their digital communication team spent the last three months working on a proof of concept (POC) voicebot, but they kept hitting roadblocks that jeopardized their June deadline.

Before getting in touch with us, MINDD already prepped the project and decided to use Chatlayer as their bot-building platform. So going into this challenge, the objective, tech stack and deadline were crystal clear. 

Getting the job done

With such a tight deadline, planning was crucial. Here’s how we organized our team to ensure that MINDD left with a working POC at the end of the second day.

A calendar showing how Campfire and MINDD arranged the voicebot sprint

Day 0 - Voicebot virtual kick-off

We organized a two-hour video call to align on everything prior to our in-person sprint. This is when we met the client, discussed their pain points and received the official briefing. 

Since we had four parties involved, we also clarified the roles and responsibilities of each party as follows:

  • MINDD: (Client) Responsible for the conversational flows
  • Chatlayer: (Bot-building platform) Responsible for the platform and integration with V Care
  • V Care: (Telephony partner) Responsible for telephony integration and queue management
  • Campfire: (Service partner) Responsible for implementing the flows and API calls from MINDD’s backend to Chatlayer

We then shared all the technical documentation via email so that we could read through it before the Sprint.

Day 1 - Voicebot sprint

The agenda for Day 1 of the voicebot sprint

09:30-10:30 - Team alignment

After quickly shaking hands in person, we sat down and got straight to work. We began with our “What does success look like?“ exercise, where we asked five crucial questions that ensure the entire project team is aligned and working towards the same goal.

10:30-11:30 - Getting familiar with Chatlayer

MINDD shared their high-level conversational flows with the whole team, and then we split off into smaller groups to divide and conquer. Our implementation expert opened Chatlayer and explained to MINDD how to build the bot as they moved through the flows. In the meantime, our developer started working on the API setup to connect Chatlayer with MINDD’s backend. 

11:30-12:00 - Technical voice setup

Before taking a well-deserved lunch break, the Sprint Team called V Care (telephony partner) and Chatlayer (bot-building platform) to discuss the specifics of the phone line integration. 

13:00-16:45 - The scoped flows

In the afternoon, we focused on building the scoped conversational flows. These are the designed user-bot interactions that follow an ideal script. As we made the flows in Chatlayer, we simultaneously trained the NLP model. We immediately added any obstacles or ideas we uncovered during the building process to a backlog to tackle on Day 2.

Geertina Hamstra says that at the end of the first day, it was already possible to call our Digital Assistant!n

Day 2 - Voicebot sprint

The agenda for Day 2 of the voicebot sprint

09:30-12:00 - The fallback flows

Working with AI means that the user-bot interactions are infinite. To accommodate this, we work with fallbacks to handle deviations from the scoped flows. These fallbacks are in-between responses that attempt to move the conversation forward before the bot replies with a not understood message. Fallbacks can take many forms, such as providing a menu with options related to a recognized keyword, a database search, an offload proposal, etc.

13:00-14:00 - Tackling the backlog

We then tackled all the obstacles and ideas we noted down on Day 1, which included all types of tasks, such as choosing a voice for the bot, using SSML to optimize the speed, creating a summary, etc. 

14:00-16:00 - Testing our way to a working demo

We tested until we had a working demo! Because all key stakeholders were in the room, we could quickly identify blockers, resolve them and try again. We repeated this process until we could pass through the different flows multiple times while activating the fallbacks. 

16:00-17:00 - Retrospective & next steps

We close every sprint with a short retrospective and define the next steps. This allows us to learn from what we just experienced and ensure that the project doesn’t “die” once we close the door on the sprint. 

N+4 weeks - Weekly voicebot status calls

We scheduled a 1-hour weekly status call for the next four weeks following the sprint. This allowed us to align with MINDD, ensure their testing was still running smoothly, and answer any questions they had about their implementation. 

What made this voicebot possible?

“Thinking about our planned 2-day POC hackathon, a lot of challenges came to mind. But, after preparing with Chatlayer and The Campfire team, all thoughts and doubts melted away. We went to Antwerp and had an incredibly productive 2-day hackathon. At the end of the first day, it was already possible to call our Digital Assistant! The second day was all about fine-tuning and establishing POC. The Campfire team is very agile; they think ahead and have extensive knowledge of the bot-building platform.”

- Geertina Hamstra, Voice Project Manager at Moet Ik Naar De Dokter

The client was prepared and knew what they wanted 

Because MINDD was already working on the voicebot before contacting us, they already had a clear scope, their technical framework documented and their key flows mapped. They also knew who needed to be part of the sprint team to get the POC off the ground. 

Virtual kick-off to pre-align expectations

Meeting before the in-person sprint allowed us to get to know each other, understand their pain points and frustrations, pinpoint their priorities and learn how the team worked. All this information was crucial in prepping the two days of intense collaboration. 

In-person, with ZERO distractions

We love the flexibility of remote working and embrace the technology that connects us, but when we want to get things done fast, there’s no substitute for in-person collaboration. Booking a large conference room allowed us to bring the entire sprint team around one table and force everyone to give their undivided attention to this POC. We could pause for five minutes to take a key decision, divide and conquer on different parts of the implementation, reconvene to gather feedback, walk across the room to ask a question… You get the picture. It was a hands-on, intense 48 hours, where everything else was on hold. 

The right people were in the room

This might seem like a logical one, but it’s a crucial one. We got all the right people in the room to progress quickly.

Our voicebot Dream Team include a Project Lead and Member from MINDD, an Implementation Expert and Developer from Campfire and Technical Support from Chatlayerr

The Dream Team, what we refer to as our core sprint team, included:

  • Project Lead from MINDD
  • Project Member from MINDD
  • Implementation Expert from Campfire
  • Developer from Campfire
  • Technical Support from Chatlayer

V Care was “on-call” during the sprint and joined video meetings when needed. By having all parties focus & allocate resources during these two days, we could immediately tackle any (technical) roadblocks.

What do you need to run a POC bot-building sprint?

According to Sophie, our Head of Implementation, you need four things to run a successful POC bot-building sprint.

✅ A clear scope

This is the main reason most POCs fail. A clear scope for your POC ensures that your entire team knows what they’re working on, when the deadline is and what the end result should be. Your team’s purpose during a sprint is to validate a leap-of-faith hypothesis, learn fast and make informed long-term decisions. If you’ve ever taken an agile management course, then you know all about scope creep and how it can derail an entire project. 

If you’re unfamiliar with scope creep, here’s an example of how it applies to Thanksgiving dinner. Scope creep is when your crazy Aunt Jackie comes into the kitchen on Thanksgiving morning and asks you to make a regular version AND a dairy-free version of mashed potatoes because Uncle Tommy has digestive issues. An hour later, she returns and informs you that Cousin Chloe just texted her that she’s now gluten-free, so you need to find an alternative side dish for her to replace the stuffing. At lunchtime, Aunt Jackie walks in and shares the wonderful news that her son Jackson changed his mind and is bringing his new girlfriend Rachel to dinner… so you need to squeeze in two extra place settings in the already overpacked dining room. 

Sounds fun right? Not only is it annoying and inconsiderate, you went from having to make a traditional dairy and gluten-filled Thanksgiving feast for 12 to a Thanksgiving feast with dairy and gluten-free options for 14 in less than five hours. This drastic change in last-minute to-dos is what we call scope creep. It transforms a clear task into an impossible mission by adding extra requirements after you agree on the initial plan.

✅ Documentation

Gather, read and organize your information and documentation beforehand so you don’t waste valuable sprint time looking for things, asking so and so to mail something over, etc. Know which questions, conversations or processes you want to automate and have them listed or organized in one place. Understand how your tech stack is structured, which things connect where and what APIs you use.

✅ The right team

It’s hard to find the right balance when building a sprint team. You need the correct profiles to get the work done, but you don’t want too many people, which will slow the process down. One person can serve multiple roles! The ideal sprint team is 3-6 people. Here’s what Sophie always recommends to her clients when assembling a team for a bot-building sprint:

  • A facilitator who’ll run the sprint, keep everyone on track, take notes and deal with all the practical stuff
  • A decision maker who has the power and agency to make all the final decisions
  • A developer who understands your tech stack, including your bot-building platform 
  • A customer advocate who serves as the voice of the customer and can share insights on how they’ll use and interact with the POC
  • An employee/process advocate who currently does the job that you want to automate

✅ Dedicated time and space

Book a large space that can accommodate the entire sprint team. Arrange all the practical stuff ahead of time: refill your workshop box, foresee lots of coffee and water, and order lunch and snacks. Remove all the additional to-dos and choices, so your team can funnel all their brain power into the sprint. 

Also, make sure to manage expectations about focus and dedication before the sprint kicks off. Encourage your team to set an out-of-office, close their Slack, Teams, mail, etc., and keep their phones on silent in their bag. By eliminating distractions from the outside world, you’ll be shocked by how much you can achieve in such a short time.

Thanks for that! Here's the download link 👇
download now
Oops! Something went wrong while submitting the form.

Thank you! Your submission has been received!
You will receive the resources by email shortly.
Oops! Something went wrong while submitting the form.

Talk with a human

If you want to discuss AI in more detail, then reach out to Alexis.

He's ready to chat in French, English and Greek.

Talk with Alexis →
Ask me anything
Chat with Marshall