Measure twice, cut once: Building a smarter, more adaptable chatbot

Building automation into contact centers and messaging channels is essential for brands looking to deliver high quality care without breaking the bank. The benefits and appeals of automation are obvious, but when you turn a manual process into an automated one, you run the risk of creating a system that’s difficult to modify or update.

Here, we’ll cover why it’s important to maintain an agile chatbot to handle changes that come your brand’s way. We’ll also go over some chatbot do’s and don’ts so that you can have the tools you need to design a world-class bot. For more detailed information and tips on this topic, check out the do’s and don’ts of chatbot design.

Think of your chatbot as a modern factory, full of conveyor belts and robotic arms assembling parts. If at any point you need to change the size of a part or the order of the assembly, it would take a lot of effort to move around and retool all that heavy machinery. If changes like this are common enough, the level of effort to update your automations start to outweigh the level of effort to do those same tasks manually.

For brands trying to deliver relevant and personalized customer care, those changes should come often. Chatbots and routing workflows should always reflect the most recent changes in service, like outages, shipping delays, or contactless pickup. They should also reflect changes to sales and marketing experiences, like new product availability and promotions. Bots should be frequently iterated on to remove unsuccessful dialogue flows and add or modify intent detection as customer behaviors and preferences change over time.

The best chatbots are iterative and agile. They have intuitive, modular structures that make it easy to add or remove functionality, and make it easy to identify where time-sensitive updates should go.

Now, let’s cover some of the basics. When it comes to building your chatbot, take care to follow these best practices before you start. Measure twice, cut once.

1. Build short, linear dialog flows

The top consumer inquiries in most industries can be handled in short conversations. But once bot dialogues get too long or complex, they become difficult to build and more likely to frustrate consumers. The most successful bot flows tend to be short, one-way decision trees — powered by quick reply buttons, drop-down menus, and forms — that are small enough to be easily measured and optimized. If things get more complicated than that, it’s almost always worth bringing in a human agent to avoid frustrating customers.

2. Reprioritize and swap out elements

Centralize your updates and early optimizations in a few specific points, like welcome messages and early branches in the decision tree. As you learn more about what questions consumers ask your bot, and how successful different flows are at resolving them, resist the urge to add more steps to your decision trees or options to each step. If you have new intents or self-service content that you’d like to test adoption for, establish a dedicated space to insert them into your routing flow before adding them permanently to the list of options. Establish short interstitial messages after a user selects “sales” or “support” that can be a permanent real-estate to publish timely information and expectation-managing updates.

3. Create naming and training convention for intents

When classifying customer intent, use an easy-to-read, hierarchical, descriptive naming convention. Here’s a basic example:

category | primary intent | secondary intent |

Sales | Price | Shipping

As you master the highest volume intents and your bot matures to be suited for more niche inquiries, you can easily add a fourth tier, such as shipping promotions or international shipping questions.

Similarly, create a standard process for the number and variety of training phrases you use to train each intent. It can be as simple as a checklist for different tenses, permutations, or speech patterns. This standardizes the “depth” of natural language understanding (NLU) across your many intents, making it easier to compare success rate across intents and to add new types of training data like regional slangs or new industry jargon.

4. Set a threshold for when to make timely updates and optimizations

Establish a list of types of current events, operational changes, or product and service updates that would be useful and relevant to the majority of your consumers. Documenting this threshold for timely updates helps hold teams accountable for actually updating the bot, and makes it easy to identify, reject, and avoid spurious update requests. And of course, as you would with any other consumer engagement tool, set a schedule for periodic review and optimization.

1. Use rich features in frequently updated flows

For the early routing menus or popular FAQ responses that get the most enhancements and updates, using rich cards, carousels, and interactive media adds a layer of complexity that can slow down updates. Channels like Apple Messages for Business and Google’s Business Messages all use different specs for images and character counts in these features, and almost no bot platforms can easily re-skin a single bot flow into every rich channel — they have to be rebuilt manually.

2. Go crazy with APIs

Integrating your bot with reputable, stable platforms like Google Calendar for scheduling or Shopify for product catalogue automations is a relatively low-maintenance, low-risk way to level up bot efficiency and customer experience. But be careful not to rely on a large, diverse stack of integrations. Doing so increases the likelihood of a disruptive update or a feature being phased out, breaking the entire experience. Outages and service interruptions are common even with the largest providers, so the more you layer in a single flow, the larger the house of cards becomes.

3. Create too many timely messages and flows

Establishing some clear real-estate in your bot dialogs where updates or seasonal messaging can live is a great tool for educating consumers and staying relevant. But adding too many creates a diminishing return where the time necessary to measure and maintain all of them isn’t worth the marginal personalization, convenience, or efficiency it offers. A restaurant menu’s “fresh catch special,” “seasonal vegetable,” or “soup of the day” are important placeholders that can offer peak-season quality and variety, but a menu full of them would be exhausting to live up to and difficult to read.

Perhaps the most important thing to remember is that your customer’s experience matters, and if you do things right, you’re spending a lot more time thinking about it than they are. As much as AI should make things better for you, it’s just as important that it makes things easier for them. That’s why, when you’re designing your chatbot, you should opt for choices that make their experience faster and simpler for the people who matter most: your customers.

To learn more about artificial intelligence, chatbots, and how you can improve your customer experience, check out The Bot Balancing Act.