Let Agents and Bots Play to Their Strengths: A Product Manager’s Perspective

As a Product Manager at Helpshift, it’s always a priority to carefully identify narrow goals and objectives for new product features and capabilities from the get go — and to stick to them. The tactics and approach will change as the team iterates, but the objectives should not waver. So, when we began to strategize about integrating bots into our messaging-based customer service, we first needed to solidify our philosophy of human agents and where bots could provide maximum benefit.

What do your agents do on a daily basis?

Whenever any of your customers have a problem or a question, they may start a conversation with you. Depending on the type of problem, your well-defined workflows and processes allow the conversation, or ticket, to be routed to a capable agent. Depending on the complexities involved, the agent gathers more information from the customer and provides a resolution. This is a very simplistic way of describing the work. Of course, there are other mechanisms like deflection through self-service that allow your customers to sometimes resolve their problems themselves without the need of a human agent.

Now, let’s break down the agent resolution process into simpler steps.

• Step 1: Once a customer has started a conversation, an agent may suggest canned responses or knowledge articles for the customer to be able to self-serve.
• Step 2: If the customer is unable to self-serve, the agent will gather more information about the issue if necessary.
• Step 3: The agent may need to check the internal customer data portal to retrieve any information that needs to be shared with the customer. Alternatively, the agent might need to update data in the internal customer data portal.
• Step 4: After the above steps, the agent finds a way to resolve the issue and communicates that resolution to the customer.

Can bots do what agents do?

Sometimes. Bots can be used to resolve those issues that are similar to when an agent sticks to a script and is able to resolve an issue through a narrowly pre-defined workflow. Here are some ways that they can work with agents:

• A Helpshift custom bot is built up of multiple steps and can relay messages to the customer and request additional information. They can collect text responses or ask customers to fill out email addresses, phone numbers, or dates. They can also offer a drop down list of options.
• A custom bot can seamlessly work with agents, routing queues and other custom bots. So, after the information has been collected, the custom bot can assign the issue back to an agent, a queue or another bot for the next step.
• Custom bots can be connected to one another through a decision-tree structure to create a workflow that does not require human agent intervention for certain issues.
• In addition to the above actions, they can also end conversations. Similar to agents, custom bots can resolve and reject issues.

How can we achieve a harmonious bot and agent relationship?

Our objective is to create bots that are able to recreate the rote and routine tasks that agents do on a daily basis. Agents often provide solutions that are unambiguous and aimed at resolution. Our bots need to be able to do the same. So, how do we achieve this?

The question we had to answer became: would this bot’s purpose be achieved better through a non-deterministic natural language processing (NLP) algorithm or deterministic rule-based bots?

In order to answer that, let’s discuss how NLP Bots work. First, a user sends a message. The intent is then derived via NLP, information is then extracted and processed via the bot. Sounds simple, right? The reality is that NLP bots can only work after they have been trained using a vast amount of data. Even at that point, there is no guarantee they will work, let alone with the high levels of accuracy needed for a great UX.

Rule-based bots, on the other hand are built to guarantee outcomes. Administrators will be able to make sure that when bots work on cases, they will deliver the same outcome as agents. There is never any ambiguity. Additionally, we believe bots can work on simple cases that are repetitive and follow standardized processes. They are not meant to replace agents; you need agents to work on complex issues that are not straightforward or standardized.

Out bots today work seamlessly with agents. And they have surpassed expectations. In one month, one of our largest gaming customers saw the following results:

• 51 percent reduction in time to resolve
• 131 years saved in agent interaction time (yes, years!)
• 21 percent reduction in time to respond

Each case can be assigned to bots and then agents, or agents can manually invoke bots. The bots are highly customizable, hence their name, and can be tweaked according to different use cases. These bots are 100 percent designed to help the customer in the most user friendly way, saving massive time for agents and money for brands in the process.

For more information on use cases and getting started with bots, check out this ebook.