PHASE 1: BUILDING THE CHATBOT
It's easy to treat a chatbot as a project, first gathering people from different departments in the company and matching calendars. Once schedules are matched, a meeting invitation is sent to the chatbot supplier and it is assumed that after the meeting, there will be a chatbot offer in the email. The chatbot may also be treated as an IT project, in which case it is considered for purchase in the same way as any other IT system.
These two approaches are unfortunately common, as in our experience homework has been left undone before considering a chatbot system. Few technology vendors are business experts, and while chatbots are generally seen as a technological choice, business insight is needed far more than is often initially realised. If a chatbot is treated as an IT solution, it is easy to think of security as a top priority, for example with the theme that "the chatbot must run from its own machine room because sensitive data is processed through it". Security is important, but a chatbot does not exist just to be secure. First you need to think about how the chatbot will benefit the organisation. There may also be feature requirements for the chatbot: it would be nice for it to be conversational. Then you just have to decide what it will talk about. To help you avoid pitfalls such as technology orientation, we've put together a list of things we've learned from working with chatbots.
The starting point should always be the business and its needs.
The starting point for a chatbot should be: what business challenge do you want the bot to solve? In other words, think about solutions that genuinely support the business. A chatbot cannot be bought off the shelf in a one-size-fits-all manner, but must be tailored both in terms of usability and technological solutions. Chatbots may also be considered when trying to correct mistakes made elsewhere. If a website interface has become a failure in terms of usability and information retrieval, the introduction of a bot will cure the bug, but not the disease.
Is the chatbot a way to increase sales or save resources? For example, if you want to automate the customer service process, you can theoretically expect that the introduction of a chatbot will reduce the number of customer service calls. However, if you're talking about improving customer service efficiency, you can't draw the equation that a chatbot is automatically more efficient. A good example of this is the slowness or sluggishness of internal IT systems. Even if a company tries to improve customer service through a chatbot, the benefits achieved will be marginal if the company's own internal IT systems slow down the service process. In addition, there are many hidden costs associated with a chatbot project - the price of a chatbot procurement is not just the price paid to the chatbot supplier. Chatbot readiness requires a lot of internal efforts from the company, which requires both working time and the use of other tools. Therefore, enough man-hours should be allocated to the project! On the other hand, a chatbot also enables you to serve your customers in a whole new way. Chatbots can be used to extend service hours to 24 hours a day, to collect leads or to help with e-commerce problems even at night.
Project resourcing
In addition to the business side, it pays to have both technology experts and customer service staff involved in the project. Customer service staff may often have fears that a chatbot will take their jobs. However, chatbots will never fully replace human labour and they will free up customer service staff's time for more challenging and usually more interesting customer service tasks. Customer service staff also have a lot of tacit knowledge about the company's customers, products and services. This knowledge should be used to build a chatbot. When embarking on a project, it is worth making sure that the people involved have the peace of mind to do their job. If the developers are constantly getting requests from different departments in the company to implement this, that and the other features, the bot will never be finished.
The project must have a management mandate but also room to move
Are you a manager? Your role is to encourage. Even if you are not directly involved in the project, you need to create an inspiring atmosphere. You need to emphasise the importance of the project, but give room to work for the executor, who is involved in the day-to-day business. Don't set boundary conditions at the beginning of the project that will prevent innovation as the project progresses. This brings us to the most challenging aspect of a chatbot: the problem should be business-oriented, but you should not lock in ready-made solutions, because in our experience there are always surprises along the way, both good and bad.
Positive surprises, through ideas generated in the process, offer the opportunity to serve the customer in ways that were not thought of at the beginning of the project. Even if from the beginning there are concrete ideas about how a chatbot can improve efficiency, there may also be only inklings about other ways in which the bot could help. It is therefore important that the organisation is involved in the development process. In our experience, management support can get people to come up with really exciting and feasible ideas, but it requires an open atmosphere and trust in the process and the suppliers. This is often difficult for large organisations that expect predictability.
A chatbot investment is not like buying a CRM, it's more like building a restaurant kitchen where you don't quite know what kind of food you'll be cooking. The physical kitchen is the integrations with all the other systems and the bot tools (e.g. chatbot editor) the spices of the soup.
The more prepared an organisation is for a flexible approach, the better the chatbot project will progress. The unpleasant surprises are often related to the type of data needed and the time it takes to collect it. As a general rule, chatbot projects take much longer to finish than companies expect. But by being prepared to collect data when needed, you're one step closer to a finished bot.
Is your company considering building a bot? Book an appointment for a free 15-minute coaching call! You can also join our webinar on the topic.
5 small things to do in the early stages that can become big problems:
- A chatbot project has no management mandate.
- Chatbots are treated with a "would be nice to do" attitude, but when the work needs to be done, budgets disappear.
- The end result is a foregone conclusion.
- Bot is approached from a security or technology perspective, not a business one.
- Chatbots are expected to be like any other information systems project. A complex bot is wanted to be deployed in a short timeframe.
PHASE 2: Technology enters the picture
Once the business objectives have been defined, you are one step closer to deciding on the technological solution. What kind of bot would best serve you? Would a so-called ´dumb clicky bot´ help? Such a bot allows you to test if the approach is OK, but it has no AI.
If you want more intelligence, for example, to process more natural language, it requires a deep understanding of what the role of the chatbot is in the end. In this case, there is some big data in the background that needs to be harnessed so that the bot can be taught to identify the visitor's needs and, based on that, produce or retrieve a response.
Data before and after
To build a chatbot, data is needed both before the bot is deployed and after the chatbot is in action. Before the bot is deployed, data is needed on the situations in which the bot would be most useful. A good place to start is to analyse the top 10 reasons why customers contact customer service. Or, for example, how the time spent dealing with a customer contact is distributed (e.g. whether 90% of the time is spent discussing to understand the customer's problem and 10% to fix it, or vice versa).
Such data is provided, for example, by recorded customer service calls. In order to analyse them, they must first be converted into text format using a separate tool. Only then can they be text-analysed to see which situations cause the most problems. In an online shop, a bot can be used, for example, to find a spare part for a car or for additional sales. The chatbot can also be used to guide you to the checkout and remind you about unfinished shopping baskets. The chatbot needs data to retrieve information from.
So what is this data? The first thing to consider is whether the data already exists, or whether it needs to be collected. Data may also be collected from different background systems, in which case it needs to be harmonised and indexed. The 'dumb clicky bot' can only answer a limited number of questions. At what point do you need a natural language bot?
Segmentation
Will the bot be shown to everyone? When building a chatbot, you also need to consider which customer segment(s) the bot will be shown to. These segments can be selected on the basis of, for example, behaviour or the content of the pages browsed.
System integrations
When building a chatbot, it is important to consider which systems it will integrate with. Does the integration work in both directions, i.e. does the bot fetch and send data, or only either or? Examples of such systems include:
- CRM (Customer Relationship Management)
- Product Information
- Website
- Call centre -system
- Labeling system
Do you have questions about the technology choices for your chatbot? Book an appointment for a free 15-minute coaching call! We will also cover this topic in our webinar, sign up now!
5 small issues at the technology stage that can become big problems later on:
- Starting a bot project with technology choices, not with figuring out vital use cases.
- No manpower is dedicated to identifying and analysing data sources.
- System integrations are not taken into account when buying a bot.
- The Chatbot will be built on an "everything for everyone, now" model.
- Some important participant group is excluded from the initial discussions.
PHASE 3: Continuous improvement
Chatbots should not be thought of as a project that starts at point A and ends at point B, but as a process of learning and development.
Since a chatbot cannot be built with an 'everything for everyone' mentality, the initial use of the bot should be limited to a specific problem or target group. Once the piloting is done, the main lessons learned from the project are compiled and either the bot is extended to new areas or the previous processes are refined to make them even better.
Bot-whisperers to the rescue
A chatbot will be of little use if it is not continuously "trained" to ensure, for example, that the underlying data does not expire. A good example of this is the addition of new products to a company's range. The bot will not know about this addition unless it is explicitly told about it. So a chatbot needs "training" just as much as humans do. Many companies use bot-whisperers. The job of the bot-whisperer is to go through the conversations that have taken place via the chatbot and determine whether the bot has been able to answer the question or not.
Roadmap for development
It is typical to start a chatbot project with a "clicky" bot and, once it is built and working, move forward on the development roadmap on to the next step of using natural language. In this case, a free-text search field is added to the bot, allowing the bot to retrieve more complex information or "skip" some steps of a clicky bot.
The key to the success of a chatbot project is that the company takes the initial leap and learns along the way. You can do endless research on theory, but only experience can steer you in the right direction. In a chatbot project, there are endless decisions to be made on the fly, so choosing good partners and skilled people for the project is a great way to ensure great work flow.
Wondering how to develop a bot? Book an appointment for a free 15-minute coaching call! Tai osallistu webinaariimme.
5 small things in the development phase that can lead to big problems later:
- The chatbot is treated as a project with a clear start and end.
- The bot is not trained.
- The underlying data is allowed to expire.
- Information is rationed and the chatbot provider is not allowed to work in genuine collaboration with the customer's employees.
- Management has no confidence in the process.
THIS IS HOW YOU BUILD AN INFURIATING BOT:
- Choose the technology first, then think about the use cases.
- Build a bot to answer all questions immediately.
- Prioritize the development project according to the company's internal hierarchy, not the customer's needs.
- Decide on the end result in advance.
- Rush the project.
TIPS FOR BUILDING A SUCCESSFUL BOT:
- Define the problem that the chatbot should be solving.
- Involve the necessary people from different departments.
- Prioritize the areas for development - it's not a good idea to try to make a bot that answers all questions right away.
- Align data sources.
- Clarify interfaces.
- Choose the technology and provider that best supports your needs.
- Build a chatbot and be willing change plans along the way if necessary.
- Publish and test.
- Analyse.
- Develop and train, both bot and people.
Wondering about building a chatbot? Book a free 15-minute coaching call!