Artificial Intelligence seems to be on everyone’s lips. Despite its success, many companies still feel quite intimidated by it. Some struggle to figure out where to start or whether they have sufficient data. So how do you make the jump into the AI pool? Not only do quick wins provide immediate value for your company or department, but they are also often low risked, narrow & focused in scope as well as inexpensive. In this blog, you will find steps on how to identify AI quick wins for your business, along with some real-life examples that might be relevant to you.

How to start?

Step 1: Ask around

If you are not familiar with the specifics of the processes, it would be quite hard to identify quick wins. Most of the time, quick wins are determined by the people who work with these processes daily. Often it is a manual, repetitive, and prone to error task that consumes much time. However, it has been done like this forever, so nobody thinks of making a change.

Step 2: Start small

Now that you have identified the process that might lead to potential quick wins, what is the next step? What if nobody believes in it? The easiest way to open people’s eyes is to use AI to make a (small) improvement. Once they experience how it works, notice the difference, other similar cases will follow. That is when the ball starts rolling. For example, at one of our customers, I have implemented an automated e-mail routing; automatically forwarding e-mails to the right department. It began as a simple model, not perfect, no user interface, just some running code. Within a month, all departments would receive forwarded e-mails (most often correctly, sometimes not) and therefore feel involved with this AI-driven process. Soon other departments independently came to me with other tasks such as categorizing IT tickets, requests, mails, or different types of communication they had identified as potential quick wins. That is what we want!

Step 3: Look at your customer contact data

The customer service department is often the place where you can use AI to get extra information about your customers. For example, you would like to know why your customers contact you and which topics are handled the most frequently. You could start reading through thousands of messages, or you could ask some of your agents if they have an idea. However, this is too much information for one person to give you a complete and objective overview. Using AI techniques, you can easily extract the most prevalent topics and how often customers talk about them. That will not only give you insights but will allow you to optimize your website, for instance, by providing the specific information that your customers are looking for. By making some controlled, easy changes, you might be able to reduce the need for lots of customer contact. Customer contact is expensive, so this is a common quick-win that can be easily achieved using AI.

Real-life examples

Project 1: E-mail routing

Scenario

An insurance company often receives hundreds if not thousands of e-mails. Customer service employees spend hours of their time sorting these e-mails and making sure they are sent to the right department. Not is it an inefficient process, but it also increases labour costs. The question is: how can we make this process more efficient? By using historical data and machine learning techniques, an AI tool can quickly learn to predict which department should pick up the e-mails.

Highlighted issues

  • Customer service employees spend a large amount of time sorting out e-mails and sending them to the right department. It is challenging to provide customers with a fast response, especially during peak seasons.
  • The most experienced employees have to do this task because they know about each possible subject and to which department it belongs. These are precisely the type of employees that should be handling the contact with the customers, not sorting e-mails.

Data evaluation

AI needs data to function, but it doesn’t necessarily need much data. What it needs is relevant data. In this case, we made use of old e-mails that were forwarded manually to find patterns and behaviour we wanted to replicate.

ROI

This project manifests both cost savings and improved customer service. The automated e-mail sorting allows improved communication and more accurate inquiry handling. The customer was able to cut back up to 2 FTE in labour costs. Due to faster responses, customer service is improved. That can be measured by the improvement of customer retention and positive customer feedback.

Project 2: Automatization of the task registration process

Scenario

Every time an employee handles some tasks, e.g. responding to a question from the customer and closing the corresponding ticket, the employee should register the details in the system. That way, the manager can keep track of whether all tasks are handled on time and can monitor how much time employees spend handling specific tasks (e.g. for planning reasons). If later, another question about an old task is asked, the employee can look up the specifics of that task and how it was handled.

Highlighted issues

  • Employees spend a lot of time registering tasks after completing them. This reduces the time available to handle requests.
  • Employees often make mistakes while registering the details, because they are under time pressure to finish other tasks in time and find this less important.

Data evaluation

For this solution, no data was needed. The only information required was which details need to be registered. Then we had to figure out how to retrieve them automatically. To do this, we used an out of the box automation tool, Microsoft Power Automate, to automatically track and register certain tasks. E.g. when an email is sent to a customer, we automatically register who sent it, what the message was, and how much time it took to reply.

ROI

Implementing this AI solution took only 2 (!) days and we are saving about 30 minutes a day for every employee handling these requests. This means a 15% increase in employee’s capacity in handling requests per day!

Project 3: Identifying frequent customer questions

Scenario

Every day, a company receives about 2000 messages, contact forms, or e-mails. Sometimes the customer has a choice to select a category for that message, but often it is inaccurate, or there are no categories to choose from. Once a month, the manager asks the customer service employees to send her the top 10 questions they come across during customer contact. This way, the manager can identify improvements in communication to reduce the number of questions.

Highlighted issues

  • It is very subjective, but it is doubtful that all employees can remember all questions; therefore, mistakes are unavoidable.
  • There is no accurate overview of which questions are often asked by the customers and to which categories do they belong to.

Data evaluation

Raw text data of all communication was required for this project, no labels needed. We used topic modelling and other NLU techniques to identify frequent topics and get an overview of the frequency of each topic. Comparing questions from one time period to another as well as identifying differences often leads to interesting findings.

ROI

The most significant discovery we found during this project was that 20% of all questions in some period were all about requesting a new card. However, customers could make this request independently on the website. By analyzing the website, we found out that the page for customers to make this request was too laborious to find. The problem was quickly fixed by putting a link on the front page, reducing calls and e-mails by 15%!

Conclusions

What is the verdict? Are quick-win AI projects worth it? Whether you’re a small or big company, quick wins will help you demonstrate the value of AI, win people over, and sow the seeds for your more significant AI projects. Just start small, ideas have to come from within the company. That is a great way to find opportunities that add the most value.

This blog is written by Maartje Keulen – Data Scientist at Mediaan.