Interviews, insight & analysis on digital media & marketing

Frustrating chatbots. And how to avoid them

By Laurent Brickell, Technical Director at Organic

Chatbots aren’t all they’re cracked up to be. But is the problem with the tech or the way brands perceive and implement them?

The world had high hopes for chatbots. Near-human levels of interaction were just around the corner, and brands were set to automate all kinds of customer service tasks. The efficiencies were very appealing. And with Google, Amazon, and IBM throwing their weight behind it, surely it would live up to the hype?

The pandemic accelerated the roll out of chatbot automation in an effort to lighten the load for teams, with the deployment of the tech stretching further than ever before. All well and good, but are they working? Have chatbots delivered on their initial promise? Or have brands scrambled to implement too soon, ultimately delivering disappointment for customers?

So, let’s explore how brands can make chatbots work for businesses and their customers.

Are chatbots effective?

IBM research reports that significant improvements to customer satisfaction and decreases in time spent by employees dealing with requests when chatbots are deployed.

On the face of it, this looks great. But IBM drives home one point over and over in its report: “Continuous improvement is a critical component of Virtual Agent Technology performance”. Herein lies one of the biggest stumbling blocks. This is not a one-off project. It’s simply not enough to stick up your FAQs and hope the bot deals with customer queries. You know where that ends up: irritated customers and an influx of calls and emails requiring attention from real people.

These are crucial insights from IBM that unfortunately do not ring true with the current market implementations of chatbots. Time and time again we see chatbots rushed through their development process with huge scope implemented but lacking any research or understanding of what customers are actually after.

This issue can be broken down into a number of factors, such as:

  • Needing to upskill teams
  • A lack of data around customer behaviour
  • A lack of scope or goals around the precise problem brands are actually trying to solve for customers

Ask upfront: What problem is the chatbot solving?

It’s important to do nothing until this is clear. There are typically two areas where chatbots excel:

  1. Surfacing information for customers quickly so they don’t have to search online help centres, order systems, or get help from a support agent.
  2. Collecting data for top of funnel marketing. This can include things like requesting demos.

Laying out customer goals will help brands develop a script to test against the dialogue the bot is using. It also helps create the intent phrases needed by natural language bots that understand how a customer may ask a specific question.

Take the scenario, “I want to book a software demo”. From this we can create a few intent phrases to help train the bot. This removes a big chunk of the guesswork element from the setup phase, and helps the engineers produce better training for the bot.

Secondly, who is the brand solving the problem for?

This is important for a number of reasons. Chatbot users will have typical journeys dependent on the specific goals that they want to achieve, needs they want fulfilled, and pain points they want alleviated.

So, by understanding who the brand is helping, and what they need, it helps them to design the bot in a more human-centred way.

Understand how users talk

It’s vital to consider that different demographics may respond in varying ways to how the chatbot responses are written. Likewise, different languages and cultures will often use different terminology or numerical systems.

Understanding how users search for the information they want is also important. Brands can use site search logs from their own website as well as queries from Google Search Console to piece together a good idea of how customers are currently trying to find information on a website. If you combine this with human agent requests, brands should be able to build a comprehensive picture of the language their customers use. By continually monitoring queries through site search and working with SEO teams to understand what users are looking for, it allows for consistent improvements to the bot.

Create content that leads to satisfaction and engagement

If the information that the bot serves up is unhelpful, then users will inevitably be frustrated. Content needs to directly answer a problem or an FAQ. By always reviewing the data received from chatbot interactions, it will highlight areas for improvement to the existing content or the need for new content.

Watson Discovery is a particularly powerful tool as you can scan content hubs and have AI index copy and images of interest that can be delivered via chatbot. This allows the bot to answer specific questions using existing content with incredible accuracy and speed.

Choose a platform that reduces frustration and enables success

Not all chatbot platforms are equal. There is a huge gap in features and capabilities even among the market leaders, and the wrong choice can easily lead to wasted money and angry customers.

Here are the important features to look out for:

  1. Having full control over natural language capabilities
  2. Integrating with data sources and API capabilities
  3. Understanding the contextual insight provided by users
  4. The ability for agents to handover to humans

The future is now, and it’s all about data

There is still a big gap when it comes to creating good chatbot experiences, and that is down to personalisation.

We repeatedly see brands using multiple systems and logins, often separating different chatbot platforms for different departments, with little information shared or accessible. Customers are left having to repeat themselves, which can make for a disjointed and jarring experience.

With the right data, it’s possible to have a chatbot that takes care of a customer throughout the entire customer lifecycle. The key to this is sharing data between systems. The bot needs to be able to identify a sales lead, store it in the CRM, then access the order system and support documents. The easy part is then just taking what was learnt from making super bots and applying user stories to the process.

Chatbots are the final, and potentially most powerful, piece in the marketing automation stack. With the right implementation they could reduce workloads and increase positive customer experiences.

Do them wrong however and watch your customer satisfaction fall.