We interviewed Kajsa, lead conversational designer at the Swedish firm Talking to Me about a voice-first chat-bot project they did for major Swedish delivery company.
A translated demo of the project is available here
I have a degree in media and communication studies and have previously worked as a manager within international customer service, before I co-founded Talking to me in 2018.
Yes, we mainly built our bots in Dialogflow, but have also worked with chatbot providers like Boost.ai and Telia Ace.
Since we generally want to be platform agnostic, I have tried many different tools.
The client was Budcompaniett, a large, innovative, Stockholm based delivery firm.
Budcompaniett wanted a delivery booking service, enabling customers with high volumes to book deliveries hands-free, for example while packing.
The initial channel was Google Assistant, but the scope later expanded to include Facebook Messenger and a web widget, because why not?
Taking inspiration from design thinking and Google’s conversation design guidelines, Talking to me have developed a design process, which amongst other things include mapping of the use case and writing sample dialogues to create a couple of voice only happy paths:
Bot: Welcome to Budcompaniett. Do you want to send a delivery from your preset address?User: Yes.Bot: Great. Where do you want to send to?User: To Talking to meBot: Okay, to Talking to me at Riddargatan 19 in Stockholm. Is that correct?User: YesBot: How many packages is it?User: 2etc...
Based on these happy paths, we built the first version of the bot in Narratory which was then extended and iterated to cover all possible user input, with visual elements added for the chat based channels.
Below: a snapshot from the bot builder - this particular turn querying the user for an address and then doing an address lookup using an API.
One of the main challenges was definitely how to collect the sender and recipient addresses from the user. We wanted them to be able to say either a company name or an address and for the bot to find the rest of the address automatically.
What we ended up doing was to collect a free-text input from the user, run it through a webhook which queried Google Maps API and do the form-filling automatically. This saved us several follow up questions. If we didn’t get an address hit from Google we would gather the address manually, step by step.
We were happy with the finished product, it went well beyond what we were hoping for. The client was also impressed by the fact that it could be so easily be adopted and deployed to other channels with just a few minor adjustments.
Below: the demo bot in English using a web client.
First of all, we needed integration with Google Assistant, including specific functionalities like account linking and transactions.
We also wanted a platform that was easy to work with for a non programmer without the need of too much infrastructure to be set up.
As this was kind of an innovation project, price was also an important consideration.
Given our criterias, it seemed like a great choice and we were eager to try it out.
The platform met all of our needs, and the Narratory team really went out of their way to fulfill every feature request that came up during the course of the project.
I had a lot of fun building it! It’s very rewarding to create something that can be tested straight away without help from developers.
I specifically like that it handles contexts for you, a Dialogflow/conversational AI concept that can easily get very complicated.
What we missed was having full visibility of logs in order to check the value of variables, etc. I am happy to see that, since then, Narratory has added support for this.
You can save time by sketching the whole flow out before starting to build. There are usually several different ways to achieve the same thing, and you want to ensure you do it in the simplest way possible.