Skip to main content


Customer service 3.0 — what are chatbots really for?



Mailiis Ploomann, Head of Telecom Services in Elisa Estonia

We are living at the age of googleing and youtubing. Should you wonder what’s the actual difference between an Indian and African elephant — you’ll find the answer in thousands of pictures within seconds. Or if you are left puzzled after yet another Bond movie how is a Martini made shaken and how it should be made stirred — youtube will provide more than 15 000 videos in 0,24 seconds. This is the actual customer experience global giants are spoiling us (and our clients!) with. And day-by-day — an expectation of a similar customer service experience is sneaking towards all the other service providers as well.

Couple of years back, it was totally acceptable to answer your clients email within a few days, today the expectation is immediately. It really doesn’t even matter if a customer has a problem or wants to share something positive with you — customer service must be able to answer within seconds and a solution must be provided within minutes.

Historically, most of the customer service agents tend to be human beings, so it is important to keep in mind another trend that is rising and constantly picking up speed.
Work is gradually growing into so much more than just proving an income. It must have a meaning and serve a purpose. People are not attracted by jobs that contain routine and mundane tasks. They want to be part of a story that produces something more than just EPS for shareholders or EBITDA. So if a company still holds a number of jobs with high share of routine and dull tasks, these positions will soon be considered temporary and they will be constantly facing the problem of employees coming and going.

So on one side we have customers (human beings) who demand quicker and more punctual customer service (in a way that is as simple as googlingto them) and on the other side customer service agents (again human beings) who don’t want their days to be filled with routine and simple repetitive tasks. On the contrary — they are looking for challenges and puzzles, where they feel that they can really use both sides of their brains and put in use all those cognitive and emotional skills.
There is constantly growing demand but rapidly declining supply, which is a situation that is not tolerated not in wild nature, nor business economics. So in an era where human expectations are exceeding the capacity of human supply — we turn to robots, which is the main reason why customer service chatbots are often the first AI products a company implements.

Elisa-s chatbot Annika will celebrate its first birthday in the spring of 2019 so it feels like a perfect moment to take a look back to our learning curve and share some insights with all of you as well.

Whenever I hear someone talking about chatbots I tend to recognize two types of people.

First there are those who have watched maybe one too many sci-fi movies and therefore imagine a sophisticated artificial mind, that thinks and talks like a human being, while self-consciously finding the answers to the customer’s questions and deliberating on the correct solution completely on its own. Those people tend to be quite skeptical about whether they should trust a robot with their own problems, because well, we all experience how often regular computer programs fail to perform — so is it really a good idea to let some computer sort my problems out on its own?

And then there are those of us who have had an unbelievably horrible customer service experience with some low quality (and rather stupid) early versions of chatbots and therefore passionately hate the idea of talking to a bot. To be completely honest — I cannot blame the latter, because it is just idiotic to be stuck in a menu bar or navigation tree that for some reason has been placed in a chat window. There are simply too many terrible IT solutions out there that get called robots by their PR departments for no justifiable reason at all. And chatbots tend to be a victim of such a crime most often.

Well, as always, the actual truth lies somewhere in the middle.

Yes, in theory there is a technology for creating a conversational AI that would be able to hold a Turing-test-proof conversation by its own, but for today, this is still mostly in the labs of scientists and not in the corporate use.

The chatbots we have working for us today, use ML algorithms (NLP) to “understand” what a human being means if he/ she puts several words in a row in a sentence and based on that “understating” it calculates the best / most probable answer that the customer should get. 
But these answers are pre-defined and programmed by human beings beforehand — so we could say that a chatbot 
understands what you are asking but does not know what the suitable answer would be — it only calculates the best match from predefined answer flows.

Now, from this point on there are two principally different approaches of how to build a chatbot for your company. It could either be part of customer service process (to simplify the work of your customer service agents) or a part of your product management process (to automate customer contacts completely).

Which one of those two suits your company the best is an important decision you have to make really early on, because large part of IT development will depend on that decision (plus of course your KPIs and ROI calculations).

In a company that sets its goal to simplify the work of its customer service agents, the business (problem) owner is usually the customer service process owner or head of Customer Service department. The project will usually begin with going through the (previously written down) customer service process and identifying the parts that could be automated. Since we are talking about chatbots, it is usually the beginning of the contact / a chat where bots can be most helpful.

AI (ML model) will be able to define the customer contacts intent and can route the client to the agent that feels or performs best with that specific topic. For that you could use predefined skills for your agents or over time, after a sufficient amount of data has been gathered — build a recommendation engine that would take into consideration the customers feedback. If after every interaction both an employee and the customer evaluate the process and end result — you will be able to create clear routes to bring together the parties that would generate the best result (higher NPS for instance).

That data could afterwards be immensely valuable insight for a team leader in daily management routines. You would get a constant flow of objective data how your team is handling specific topics. Plus, you could see which topics tend to be more preferred by whom and who might need extra training on rehearsal on some customer cases.

In such companies, the customer service agents are usually the ones who “train” the bot, meaning that they are the ones who generate the answers a bot should give to a customer once a specific question is recognized. The business owner of the project will be someone from the customer service department and a fair amount of front end development hours should be considered — as the main goal will be making employees work easier and more enjoyable.

This would be a perfect solution for a company that offers customer service as an (outsourced) service and therefore is not the actual owner of the products / services they are supporting. Or for those, whose products are mostly tailor-made to a specific client or so complicated in nature that there simply are not enough standard answers (or even questions) to predefine.

Now in the case where the company owns, develops and supports its products/services all by themselves — a food for thought might be to figure out how to automate the customer service as much as possible all together. Especially when there are a lot of customers, with reasonably similar questions.

In such a case the business problem owner will be someone from the product management department (most probably a product manager) and human customer service agents will be considered as a middle step towards completely automated customer care.

Of course it is never realistic to think that 100% of all customer contacts could be automated — but (based on 10 years of telecom product management experienced gut feeling) 2/ 3 definitely could– as in those cases the human agent identifies the problem (question), then checks 2–3 parameters from CRM and then gives an answer to the customer based on that information. There are very few reasons why a robot (bot) couldn’t perform those exact tasks.

For instance, a customer bought something from your website and now wants to know when the package will be delivered to him/her. A human agent cannot simply answer something that feels right, they have to identify the customer, check the purchase timestamp, courier pick-up time and registered delivery time. That is a totally obvious and normal question — but an answer from a human being does not create any added value to the customer. It is the actual information (from CRM), the speed and accuracy that count — and a robot has the upper hand in those for sure.

Finding and automating such customer contacts is the main task of product management driven chatbots with a bonus feature that shouldn’t be underestimated as well — the business intelligence gathered through the analysis of all those customer contacts. All the product managers will get an objective and near-live feedback of the questions that customers have with their products which will be extremely valuable input to the product management, internal processes and business decisions in general.

Another aspect that creates value in this scenario is the quality and accuracy of the answers a customer gets. Human customer service agents usually follow the instructions or manuals written by the product managers, but you will always have a certain percentage of employees who have just joined the team (and are just learning themselves) and that other percentage that considers themselves way too experienced for instructions and simply do not read the manuals anymore. Or even worse — they have printed out the manual a few years back and still follow the hardcopy… 
In companies where the product / service lifetime cycle is rapidly and constantly changing — it is not uncommon for a customer to get three totally different answers from three different customer service agents for the exact same question. And to be fare — there is no one specific to blame. That’s just the way it is when humans are involved.
Then again — if we have a technology that can help us improve the situation 2/3 of the times — then why not use it?

Elisa’s chatbot Annika has been the second type (product management driven) bot from day 1 and it has been our main goal to rid our human customer service agents from routine and simple inquiries as quickly as possible. We have put a lot of effort into building a system where product managers are responsible for the answers that our bot gives out to customers — as this is the only way to guarantee that it’s knowledge is always in line with the latest business decisions and so far it has proven to be the most fruitful path.

It is important to keep in mind that no-one should start an AI project just because its sounds interesting (or because everyone else seems to be doing something :) but if you do run a business that deals with (a lot of) customers and digital customer service is your priority, then it is the perfect time to start building your own bot — as it will be impossible to keep up with the demand otherwise.

Figure out which will be the right track for you (what sort of business problem will you be solving), find a partner that can help you with the data science part and start experimenting!
Keeping in mind that text based chatbots are also good base for building digital customer service solutions through voice as the main interface in the future as well — but that is a whole new story for the next time :)


Seotud märksõnad

Vabandame! Teie veebilehitseja on liiga väike meie kodulehe külastamiseks.

We're sorry! Your browser is too small for this website.

Приносим извинения! Размеры вашего браузера слишком малы для посещения нашей страницы.