Onboarding a new organisation to “Starting up Smoothly” — A Migri perspective

If you follow our Inland blog or twitter even just every now and then you’ve probably heard about Starting up Smoothly: It’s the project to create a network of independent chatbots. If you haven’t heard anything about it, I encourage you to scroll to the bottom of this article and read through some of the links I provided there.

In this article I want to focus on how we onboarded a third organisation, the Patent and Registration Office, to the project. The goal of this process was to get the PRH team up to speed as quickly as possible. I hope that by sharing the phases of onboarding, other organisations can learn. This blogpost also shares some insight on how to create a chatbot in general and is rather documentary. If you want to read about the experience of onboarding to the project from the perspective of the new organisation, have a look at Olli Ilmarinen’s blog post here.

Phases of onboarding — from interest to start of being part of the cross-organisational team

Onboarding PRH took place in four phases, which I describe in more detail in the next subchapters. The phases might overlap, so their order is based on when each phase starts:

Phase 1: Initiate the collaboration
Phase 2: Starting the collaboration
Phase 3: Defining & sharing team working practices
Phase 4: Defining the new chatbot

When the Finnish Patent and Registration Office (PRH) joined the collaboration to create a chatbot network, the project had been ongoing for about 6 months. Previously it had been a collaboration between the Finnish Tax Administration (Vero in Finnish) and the Finnish Immigration Service (Migri). Working for only 6 months together in an experimental project meant that working practices were very flexible at this point. So, let’s go through the four phases of onboarding PRH to the project one by one:

Phase 01: Initiate the collaboration

This phase spanned from the first contact PRH made to the chatbot network team to express their interest, to the organisational kick-off event, where management-level project partners were present. In the organisational kick-off we discussed

a) the common goals for the project,
b) the high-level responsibilities that were needed for the next phase
c) needs for team members, their responsibilities and engagement in the project.

When onboarding PRH this phase lasted from May-June 2018 to mid-August 2018. Vesa Hagström, Migri’s Chief Digital Officer, was mostly leading the initiation and negotiating with the upper management level on PRH’s side. One feedback we got early on was, that the roles and responsibilities of the team members that would do the actual chatbot work were not clear in summer 2018. For this reason, when the four new team members joined the project, they were rather confused what their roles in the team would be. The main learning from this is that project member roles & related expectations are never to be taken for granted: We, from Migri side, knew the roles needed within the PRH team, but we did not manage to communicate it clearly to the new organisation.

Phase 02: Starting the collaboration

In the next phase we started the collaboration with the team members that were in charge of producing the chatbot solution for PRH. We started with a Team kick-off event in June and August 2018. Here it was important that all team members were present as we
a) discussed expectations of all parties from the new organisation,
b) demoed the current solution and explained how it works and why, and
c) established a common timeline for the upcoming collaboration.

Actual working together started only in August 2018 and one new team member joined at that time. After the second team kick-off the actual work started for PRH. They trained and got certified to use the boost.ai software. Boost.ai is our current software provider.

In a next step we shared our best practices and guidelines for designing chatbots. This included discussions on how to design a chatbot from the end users’ perspective.

Phase 03: Defining & sharing team working practices

In the third phase of the project we agreed on project roles and responsibilities within the project team. In the first two phases of “Starting up smoothly” we have used RACI charts for this. We have also defined the following roles within each organisation:

Project manager responsible of initiating, planning and managing the team work
Content lead Responsible for the content of the organisation (action replies, personality, test data, initial training data)
Content designer Create content within the boost.ai software (action replies, test & initial training data)
AI Trainer Responsible for the AI model of the organisation (model performance & maintenance, training data, synonyms, monitoring conversations)
Substance expert Assist in content creation, do not work in boost.ai. This role is not always defined and present within the Starting up Smoothly-project team. However, it is important to have access to all required substance expertise within the organisation.

In a next step we discussed and made initial plans how the PRH team works together internally and how they collaborate with the existing Starting up Smoothly-teams: As an example, we worked in the same physical space every Tuesday. This gave us time when teams could concentrate on their work, either for their individual organisation or among different organisations.

Phase 04: Defining the new chatbot

After training the new organisation’s team and defining the common working practices is when the actual content work started. If you want to be very picky on “onboarding” then you could say that anything from this phase onwards is not any more onboarding a new organisation but creating a chatbot. However, since we had 2 existing chatbots at this point and PRH was missing theirs, I include defining and developing the new chatbot as phases of onboarding PRH to the project.

As part of the Starting up Smoothly-project we had already decided our target audience, foreign entrepreneurs who want to start their business in Finland, and Vero and Migri teams had an overview of what kind of content needs to be built. For PRH we gave an overview of the user research outcomes from the previous phase and discussed together what they meant for content creation.

Based on this knowledge we defined new user personas, fictional representations of the service’s real end users based on real data and user interviews. This was a common activity among all participating organisations. The user personas needed to have touch points with all organisations in the network:

Next PRH needed to establish the personality of their chatbot: What kind of language does it use? What’s the tone of voice? Is it like talking to the CEO of the organisation or to a low-level expert? What does the chatbot look like? Is it humane or robot-like? How does it use humour or emojis? These and more questions needed to be answered before starting to design actual content. User research, such as interviews, surveys or user tests, were used to support the personality definition. We used insight from the previous phase of Starting up Smoothly, but PRH also conducted their own customer survey. The survey was based on previous user survey that we conducted at Migri, and the improved version done in the first phase of Starting up Smoothly. Having material to base their survey on, made it a lot easier for PRH to design their own survey.

Outcomes from PRH customer survey, which was based on Migri’s and Vero’s previous customer survey

In the next step we scoped the content of the new organisation. We created a high-level overview of all possible content and used the user personas to define a small area to start with: This was the most important content for PRH to build. The content was later expanded to other areas step by step. PRH decided to focus the content only on questions regarding starting a company, and even more specifically, private traders and starting limited liability companies. This helped the team to focus on a small area that they could actually finish in a rather short time. Based on the first company type, we then extended the scope to the second.

After we have defined the scope of content, we designed the high-level content structure for the organisation. This includes the high-level intent structure as well as the more detailed structure for one of the scoped content areas. Building this high-level and reusable structure from the beginning and thinking of future content areas helps in building a future-proof model from the beginning and saves lots of work later on.

Initial intend tree for Limited-liablity companies (short LLC)

When we had done all these four phases together with PRH, they started the actual content creation process. If you want to know more about creating content for chatbots I encourage you to read these posts in our inland blog:

1 year of chatbot development — 10 best practices

Collaborating to develop a chatbot network

Feedback and learnings

Reflecting back at onboarding a new organisation to the chatbot network project there are a few things I would do differently next time:

Make the roles clear earlier

Most of the confusion among the PRH team was caused by not knowing what is expected and who does what. During that time, I thought that the team members will learn by doing (which they did), but today I believe that by being more explicit PRH might have gained speed earlier and with less confusion for the new team members.

Speak the same language

This learning has to do with the first: The unclearness of roles, responsibilities and tasks in the beginning had to do with not speaking the same language. We all spoke English in the team, but being deep in the area and technology, we might have used lots of words that were clear to us, but much less clear to the new team members. I personally feel this is one of the areas that I will never stop to learn during my lifetime, so I can only try to get better.

Be clear about the effort

One of the struggles we had during onboarding PRH was time commitment. Again, this is related to the clear roles. It was not understood in the beginning that creating chatbot content takes a lot of time. This is of course related, to what type of chatbot you create: In our case we want to give answers within the chat, and this means that we need to create the content specifically for this purpose (we cannot copy-paste from a website).

The feedback we have received from the PRH team about their onboarding experience was overwhelmingly positive: They felt they have learned a lot during this collaboration, they have understood how much effort it is to create a chatbot and gained understanding how a chatbot can help their customers on a hands-on practical level.

I am the coordinator of this collaboration and was in charge of planning and conducting the onboarding of PRH to the project. I had previously been involved in setting up the chatbot team at Migri as well as given advice to Vero how a chatbot team can be run, when we started the first phase of Starting up Smoothly. I invested about 1 day a week to prepare the next onboarding session during August to October 2018. Luckily, we had documented the process of creating Migri’s chatbot personality and starting to create content for it. This made the onboarding of PRH more efficient for all sides.

All in all, I think onboarding PRH to Starting up Smoothly was really quick and effective. We had a few hick ups but generally we’re all happy with how it went.

author: Suse Miessner
edited: Mariana Salgado

More about Starting up Smoothly:
From Demo to Public Pilot — Documenting Starting up Smoothly working practices
Onboarding a new organisation for a joint chatbot service
Starting up Smoothly — Connecting government agency information through chatbots