How To Scope a Conversational AI Project
I am sure that you can think of examples of Conversational Bots (either Chatbots or Voicebots) that have caused you frustration because they did not fulfill their purpose. There can be many reasons Conversational Bots fail to do their job, as there are several ingredients that need to be in place to implement a good Conversational AI project: having the proper platform for your journey; assemble the right team; planning the project growth and expansion in terms of languages, channels or geographies; thinking on scalability of both the solution and the infrastructure, etc. One of the fundamental ingredients is defining the content that needs to be included in your Conversational Bot, or in other words determining the scope of your conversational AI project.
Even if this seems like an obvious statement, I have seen projects failing because a lot of effort was put into implementing features and content that, eventually, no one used. This article gives some insights and tips on how to scope your project to make your Conversational Bot truly helpful.
From having an objective to the Business Case
The first question that you need to answer is: why do you want to implement a Conversational AI solution in your company? If you don´t have a clear answer, you don´t pass the first gateway. By experience, successful projects are those that have a clear, identified use case and clear drivers aligned to company/department objectives, as this allows focusing energies in one direction.
The objectives may vary from company to company, but they are typically three non-exclusive but complementary ones: increasing customer satisfaction by lowering waiting times and quickly resolving customer issues; costs efficiencies in the Call Center by freeing up valuable agent time; or increasing sales conversions by helping the user and reducing abandonment rate.
Let´s look at an example. Imagine a retail company in expansion who wants to avoid a growing Customer Service cost by implementing a Conversational Bot that automates the handling of certain user questions and hereby frees up the time of customer service agents. This will secure customer satisfaction by giving immediate answer to those using the Bot and to better serve their customers in complex queries. At first instance, you may think, great, they have a clear objective on why they are doing it, so all good. Still, I am afraid that this is not enough to start your project as you want to evaluate if there is a business case that justifies the investment. This is the second gateway to go through.
To build a business case for this retail company, you will first have to list out all the different questions (user stories) asked in the Call Center. You don´t need complex tools, a simple Excel will do the job. Once you have the list, you need to map them out based on three factors:
- How frequent is a certain question asked in the Call Center? (frequency)
- Are you able to automatize the answer or its resolution (the process)?
- How much effort do you need to spend to automatize it? (complexity)
The first question is easy to resolve as all Call Centers have statistics on the type of questions that are asked. And if you don´t, no need to worry, a workshop with a few agents and team leaders will give you all the info you need as they know the top questions inside out. When analyzing Call Center statistics, you will realize that there are a few questions that are clearly outstanding over the rest because the Pareto principle, also known as the 80-20 rule, always applies i.e. 20% of the calls drives 80% of the volumes. You would be surprised to see how this rule is consistently met irrespectively of the industry.
The second question is very fundamental one, which in essence says: will the Conversational Bot be able to provide the same answer as your Call Center agents would? If the bot provides a general, non-resolutive answer, the user will finally call the Call Center and you are then back to those frustrating Conversational Bots that I was referring to in the beginning of the article. You don´t want to be there. To answer this question, you will soon realize that some answers will be straightforward to provide, but others will require that you add backend or 3rd party integrations to fetch the right information. Imagine a client on the website of the mentioned retail company, looking at a red shirt and then asking the bot: “Are there any in size M?”. The Conversational Bot will have to, apart from understanding context, check stock to provide a final answer, which is the only way of serving the client and saving the company a call.
This leads us to the third question, which complements the information from a different angle. You need to understand how much effort (“complexity” in the graph) is required to automatize an answer that is connected to back-end systems:
- What integrations (if any) are needed to answer such question?
- How much effort is needed to put them in place?
Bear in mind that there may be questions that you don´t want to automatize but still are needing integrations in place. Think of a user complaining about your service, you really want to hand over this conversation to a real agent i.e. you need to connect with the Live Agent provider; or think of a difficult query that you can´t answer but you still want the Conversational Bot to record it in the ticketing system.
What about the savings/value?
By now, you have listed out all the questions asked in the Call Center and assessed them based on frequency and efforts to automatize. To close your business case, you now need to map each of them with the (monetary) value that you think they are going to provide. This is now an easy task as you have done all the preparation work already: if you are fully resolving a client’s question, you should associate savings that are aligned with the cost of one Call Center phone call; if you are not resolving but directing the query to cheaper channels like a Live Chat, you should equally associate savings but, in this case, in a smaller portion (Live Chat channel is less expensive than phone); if your objective was to increase sales in your website, you should assign a value generation based on closing a purchase on your website.
In our retail company example, they would now have the full picture on how to expand their Customer Service footprint without needing organic growth, they would now know how much they need to invest and how much they will get in return (ROI).
Plan the project
If you have passed the second gateway, you have, in essence, created the backlog of your project. You now have a list of questions categorized with the value that they will bring should you automatize them, so everything is ready to plan your project. Note that, based on your analysis, there will always be content that you will not want to implement i.e. out of scope content, which will be addressed in a generic way.
At this point, you want to involve a broader team to prioritize the backlog and to make sure that you are splitting up all the backlog items into different phases. The recommendation is: don´t try to do too much in your initial phase, start your project with a small chunk of content and grow from there (Small, Medium and Large phases as shown in the picture above). Focus on the low-hanging fruit, which will allow your organization to learn by doing while providing measurable quick wins. Make sure that you have clear KPIs that prove the value, which will reinforce the benefits of developing an intelligent conversational AI solution and will get full internal support for further phases. The project will be ready to kick-off.
If you are implementing a Conversational AI project for the first time, you will realize that, once you go live, the behavior of some of your customers may change as well. This is a dynamic effect where you might see that some clients will stick to traditional channels, while others shift to the new ones, or a combination of both. On top of this, expect the overall number of contacts in all channels to slightly increase as you will be reaching out to more people.
Conclusions
This article has outlined a simple process to take informed decisions on where you should invest your resources to implement a Conversational AI solution: from identifying the use case to mapping each user story to its value provided. The result is a clear business case that will allow you to start a solid Conversational AI journey by aligning all the organization resources to row in the same direction, which is the only way to be successful.