Based on our own experience implementing Yed.ai at Advanced Programming Solutions, some insights from our customers and research papers (Corea et al., 2020; Lewandowski et al., 2021), we dare to share here some tips to make your chatbot implementation journey smoother.
1. Decide early on what your main goal is:
Is it a task oriented bot? Or is it a bot to promote conversation and engagement? Clarify your vision. If task oriented: what are the KPIs? How will you measure success? Think about the conversational agent as a collaborator in the company. In most cases, organisations expect a single person to fulfill different roles. Make a list: what exactly will be the responsibilities of your bot? What are the most important to prioritize? Let’s follow an example: we want our bot to help fully remote employees to find information quicker within our company knowledge base. We have lots of information all over the place, in different repositories (Confluence, Drive, the ERP system…), and whenever new people join another project or a different team, we wish to provide them access, explain ways of working and adopt learnings. At the same time we do not want them to feel they are judged. We wish to ensure people are able to find this information and reduce the time they need to be ready for the role. We can measure it through the time new to roles take on average to get ready and time invested by senior members in the onboarding of new team members.
Once you have the goal clear, name a Content Coordinator for your bot, the person in the organisation who should be committed to serve this goal. If you agree for this role to be external, be sure to assign the right contact person internally, someone who knows who is who and is senior enough to look for the right answers to be provided and right people to be involved.
2. Identify your “sources of truth” and cooperate with them
This is probably the most relevant tip in order not to waste time and avoid rework… And here we are not referring to formal documentation only. Each organisation has knowledge keepers in different areas. These are the people that everyone recognizes as reference whenever they have a doubt, “ask Fran” or “let’s ask Dani” will be the answer… so you need to identify and involve them in the bot implementation process. They may give you all the key answers but they can also be part of the answer the bot will provide (hand-over to a human!): “I don’t know this yet, but I recommend you to contact Dani”. This might take a while, and might lead to discovering that there was old information being given, or something not working correctly. The Content Coordinator shall take time to agree on the right answers with these people. They may or may not hold a formal position of responsibility that is not relevant. They are prescribers and you need them to help you get to the truth. To continue with our example: We identify the people within the team that are recognized as the best by peers in order to solve a specific area of doubt before referring to documentation and manuals. We ask them to be our advisory team in case something comes up around their area of expertise and handle the question directly. We create a subject matter experts group email account, so all of them receive the queries at the same time and they can decide what to answer as SMEs. Call it “Domain Board”, “SMEs Committee” or “Wise Council”, just make sure they are recognized by the existing governance structures in the organization to empower their decisions.
3. Agree the process to follow for the conversational agent to learn
This is not about checking the quality of the neural network accuracy. Your tech solution should have that covered. You should worry about establishing who will ensure the chatbot learns the right stuff (training data availability and quality). We already know of big failures with chatbots learning just from user interaction… your organisation might not want to let trolls teach the AI what is correct. So, to whom should we go to when a doubt the organisation has no answer for comes up? Who decides which questions are out of scope of our bot? Be aware that learning what we don’t know is as important as providing answers.
You might think this issue should be solved by technology: partly right. Knowledge Management experts are in the tech field that can help orchestrate the effort and define the right information structure to start with, nevertheless, if you wish the bot to learn, say the truth and be accurate, you better agree how you will get answers and where the data will be stored, how it will be managed and who will feed it into the bot.
In our example: Once we have identified the advisors, this includes setting up a procedure that works for them and the bot development team. Keep it realistic. We assign the advisory team of SMEs an hour per week to resolve new doubts coming from the bot, facilitated by the bot Content Coordinator. This time guarantees a high level of quality in the answers. Plus, we have the Content Coordinator conducting the review of the bot answers once per month, with a minimum of 30% of conversations being supervised by humans during the first six months. You need to be able to manage big amounts of data of course, have them correctly stored, protected, cleaned and available. This is where your tech partner comes into play: to make the Knowledge Management job easier.
Trending Bot Articles:
2. Automated vs Live Chats: What will the Future of Customer Service Look Like?
4. Chatbot Vs. Intelligent Virtual Assistant — What’s the difference & Why Care?
4. Be ready to learn bottom-up and commit to continuous improvement
As per research papers, a group oriented perspective and voluntary participation are the best way to kick off your efforts of implementation.
A top-down enforcement to use the bot does not work. Plus, you need to be aware of your own organisation learning curves. The bot will learn as quickly as your organisation does: the pace to teach the bot a new answer is the pace in which you create new knowledge that is agreed by the right people and therefore can be taken seriously.
This is why one of our conclusions is that the chatbot implementation is the best way to conduct digital transformation in organizations: it makes the information gaps evident and gets the missing elements to the surface. Chatbots also make you reflect on stuff you did not expect to come up in the first place.
5. Make sure your implementation team is diverse and interdisciplinary
As you are designing conversations, you need to recognize that people make questions in different ways, and perceive information differently. Therefore, a chatbot development team composed of great scientists alone will not be useful. Neither a team of senior engineers on their own. In order to move the needle implementing chatbots you require not just great technical skills (programming, data scientists and NLP experts) but also marketing people and communication related skills, preferably with knowledge of your organisation’s main domain area. These people have to interact with the chatbot and each other. Going beyond disciplines, you will benefit from having generation-Z and baby-boomers working together on the organisational knowledge corpus development… as well as minorities representatives and people with disabilities. They will not only help to avoid biases, but will also be able to discuss ethical issues that may arise. For example: you might face questions such as “how much does the CEO earn? or “How do I report bullying in my team?”. In this case you need to make a decision on whether to get them on your scope, or you wish the bot to say it does not know. In any case, it is up to the organisation to make a choice. Some conversations will be required to be held that were not expected. Getting different perspectives heard is a must. Research papers confirm that AI projects cannot be simply outsourced or developed in isolation by any internal team. Get the right people to facilitate your journey, it is not about tech only.
If you wish to exchange best practices or collaborate in any way please contact firstname.lastname@example.org
Corea, C., Delfmann, P., & Nagel, S. (2020). Towards intelligent chatbots for customer care: Practice-based requirements for a research agenda. Hawaii International Conference on System Sciences.
Lewandowski, T., Delling, J., Grotherr, C., & Böhmann, T. (2021). State-of-the-Art Analysis of Adopting AI-based Conversational Agents in Organizations: A Systematic Literature Review. PACIS 2021 Proceedings, 167.
Don’t forget to give us your 👏 !
5 keys to boost chatbots adoption in the organisation was originally published in Chatbots Life on Medium, where people are continuing the conversation by highlighting and responding to this story.