Sasha is the head of customer service at a growing software company with 200 employees. Her team of 15 customer service agents are well trained and have access to all the documentation for the company’s software offerings. Day after day, the agents answer a mountain of repetitive queries and struggle to have enough time for the more complex and serious problems. Target response times remain unattainable, customer satisfaction scores are dropping fast, and two agents quit just last week.
Sasha’s ready to quit herself. How can her team spend more time on those complex problems and less on the routine ones? Can’t her customers just search the website? All the user documentation is there. It’s just a little … disorganized.
Then one her agents asks about creating a GenAI chatbot. They could train it on all the information customers can’t seem to find on the site. The bot could answer those questions, freeing up the human agents for the bigger problems. And if they added the bot now, they’d beat out all their competitors and maybe keep more customers.
What Is a GenAI Chatbot?
I hear you asking, “Haven’t chatbots been around for years?” Yes, they have.
But GenAI chatbots are different.
You’ve likely interacted with a rules-based chatbot. This cousin of the GenAI chatbot is preprogrammed with questions and responses. The bot matches the keywords from a customer’s query to those in its database and then delivers the prewritten response.
That’s good as far as it goes. But when a customer asks a question that doesn’t exist in the database or when they ask it in a way that the bot can’t understand, customer service stalls. A good setup would transfer the customer to a human to help resolve the issue. But if this is a basic question, everyone’s time has been wasted.
The goal of a GenAI chatbot, also known as a conversational assistant, is to answer basic questions, no matter how they’re phrased, in everyday language. The bot isn’t quoting its database; it’s writing a fresh answer, word by word.
That’s a key difference with GenAI: It uses natural language processing (NLP) to compose answers. Customers don’t need to frame their queries in a specific way to get the desired results.
There are still plenty of questions a GenAI chatbot won’t be able to answer. Customer issues can be complicated, going beyond the information the bot has been trained on. To ensure that your customer’s issues can be resolved, your best choice is GenAI chatbot that can hand a customer over to a human, sometimes called a hybrid chatbot.
Why Should You Consider a GenAI Chatbot for Your Site?
It seems every company is adding GenAI to its services or its backend. In a 2024 McKinsey & Co. study, 65% of respondents said their companies were using GenAI in at least one business function. Most often, GenAI was found in marketing and sales, product development, and IT.
But you shouldn’t add GenAI just to join the crowd. You need to know how it will benefit your company.
Midsize companies struggle to balance costs and growth. When you have more customers, you also have more customer service issues. Yet you want to keep staffing lean. Targeted use of automation can stretch your team further. And if customers can resolve their issues quickly with a bot, you’ll increase customer retention, too.
With a GenAI chatbot not only can you provide 24/7 customer service but you have the opportunity to resolve common issues permanently and save money. For example, Delta Airlines saw a 20% decrease in calls to customer service after it implemented a GenAI chatbot. DoorDash has saved an estimated $2 million annually by implementing its GenAI chatbot and it was able to resolve 18 of its most common issues, saving a further $3 million.
Your customer service bot is best used for routine queries. You’ll be training it on your materials, so issues need to be resolvable with those materials (more on that below). If your customers could work out the answer themselves by digging around your website, that’s a good query for the bot. Think queries like:
· How do I reset my password?
· Where’s my order?
· Does this product come in another color/size/quantity/etc.?
· How do I cancel my order?
Take a look at your customer service data: What are the most common, easily resolved questions you get? Now imagine what it would be like if your human customer service team no longer had to answer those questions. How much time would you give back to your team—time they could invest in resolving more complicated issues and winning back frustrated customers?
You also will be able to gather more data about your customers. Knowing more about when and why your customers are unhappy with your support and your offerings allows you to make changes that will grow the business.
And as the company grows, you can keep staff additions to a minimum because AI can handle multiple conversations at once. Instead of hiring—and training—three new customer service agents, for example, you could increase your bot’s capacities and hire one new agent.
Finally, just like your human agents, a GenAI chatbot can learn to improve its responses. A rules-based chatbot needs humans to write better responses for it.
What Makes a Good GenAI Chatbot?
GenAI chatbots sound pretty good, don’t they? Except that, right now, a lot of people don’t trust GenAI—or any AI. There are a lot of ethical questions around its use, and not everyone wants to talk to a bot. Automated customer service lacks the human touch, something we crave when dealing with emotional or complex issues. As customers ourselves, we’ve all been trapped in that endless loop of non-help some automated systems are prone to.
You can do better by following a few best practices.
Be Transparent
When a customer opens the chat function on your site, the first thing they should learn is that they will be chatting with a bot. This will help them set expectations for responses.
Continue that transparency by programming the bot to list its sources for responses. If your customer asks about shipping costs, your bot could summarize the information and include a link to the page where shipping costs are listed on your site.
Teach Your Customers How to Use the Bot
Consider your customers’ familiarity with chatbots. If this is brand-new to them, at the start of an interaction, provide some basic directions. You could create some test prompts for users to try or create a how-to video users can click on.
Customers who are familiar with GenAI and chatbots might not need a how-to, but a quicker way to ask a question is always welcomed. You could program buttons that will send prompts of some frequently asked questions or give a list of sample queries.
Create an Easy User Interface
We tend not to think clearly when we’re frustrated or angry. Make the user interface simple and prioritize a mobile environment. What’s easy to do on a desktop may be much more difficult on a mobile device. Consider building in buttons for actions, questions, and links to more information.
Avoid Dead Ends and Circular References
Please, for the love of Lenovo, avoid dead-end directions and circular references. Nothing induces a customer to leave you faster, I promise.
Take this example: I recently upgraded my phone. When I opened my authentication app on the new phone, the app didn’t recognize the phone and wanted me to respond to a push notification … on the app I was trying to log in to. What? It took three emails with customer service to gain access to the app on the new phone and all the authentications I had in it were wiped out.
I’ll be switching apps, thankyouverymuch.
Ensure Accuracy and Safety
Make sure all responses are accurate. This is a particular peeve of mine because outdated instructions prevent customers from helping themselves. Check any steps in the documents you’ll train your bot with: Have button labels changed? Are the menu options the same? Do the steps still work?
Make sure, too, that any information given is safe. It should go without saying that you don’t want to endanger your customers. But you also don’t want to be liable. Increasingly ,companies are being held responsible for the responses their chatbots give customers. Include trying to get harmful information out of your bot part of your prelaunch testing (more on that later).
Make Reaching a Human Agent Easy
Train your bot to recognize when a customer should be handed over to a human agent. For example, if the bot has asked for clarification twice and is still not able to resolve the customer’s problem, it should connect the customer to a human customer service agent.
The bot should also be able to identify scenarios that immediately require a human agent. Go back to your customer service data to identify those sticky problems that need the human touch and write up directions for your bot.
Finally, tell customers how they can redirect themselves to a human being. Don’t make it a secret: customers will figure it out anyway and they’ll resent you for hiding the option. Recently the app my gym uses for class schedules and memberships went down over night. The first class of the day is at 5 am. By the time I walked through the door for the 6 am class, the owner of the gym was furious. She wasn’t angry because the app was down; tech issues happen. She was angry because she had been trying for over an hour to reach a human being.
She had gone to the website and opened the chat function to report a problem with the app. At that point, she didn’t know it was a company-wide problem. The bot gave her vague responses that didn’t help. It didn’t offer to transfer her to a human and there was no way for her to do so. In the past, her method had been to close out of an existing chat and open another until she reached a person. That morning, she reached no one.
Her entire business depends on that app. It needs to work. And when it doesn’t, the problem needs to be addressed quickly. How long will your customers stick with you if your customer service app doesn’t help them? If they feel tricked into talking with a bot that can’t help and if they can’t reach a human?
The short answer? Not long.
Follow All Relevant Laws and Regulations
Laws around AI are a work in progress, and keeping up with changes can be a lot of work. Work closely with your legal team to stay in compliance. Determine the laws and regulations you need to follow in your company’s jurisdiction and your customers’.
Utah, for example, signed the Artificial Intelligence Policy Act (UT AIPA) into law in March 2024. The new law states that companies must tell customers when they’re interacting with GenAI. Companies also are no longer allowed to use GenAI as a defense against any liability claims.
Create AI Policies
To help you follow those regulations, and your own ethical guidelines, create internal and external AI policies. Everyone in your company should be familiar with both policies and at the least know who to contact when they have questions about it. The external policy should be easily accessible on your site and be part of your bot’s training database.
At minimum, consider detailing how AI can be used, by whom, and for what purpose. Or to create a shorter policy, detail how AI can’t be used, who can’t use, and why.
Will you collect data? Will you train your AI on it? How will you protect customer data? Who will have access to it? How long will you keep it? Can customers opt out of having their data used to train your AI? These are all things to think about.
Again, consult with your lawyers to ensure that your policy reflects what you want it to and that it’s enforceable.
Ask for Feedback
Be open to hearing what your customers really think about your new bot. After all, if they hate it and you don’t know it, you could lose many customers and, ultimately, your business. Better to know and address any problems when they’re small and keep your customers.
Try a one-question survey at the end of the conversation. Ask something like, Did we solve your question today? Or, Are you happy with the response you received?
If you purchase a more feature-rich bot, you could allow customers to rate individual responses, the way ChatGPT and Claude do. You’ll be able to identify and resolve problematic responses quicker that way.
Preparing for Your GenAI Chatbot Journey
Making a GenAI chatbot part of your customer service strategy could help your company successfully serve more customers and jump past your competitors. The potential for affordable 24/7 customer service, cost savings, and happier customers are not to be shrugged off.
In part 2 of this series, I’ll look deeper into this decision, covering costs, timelines, potential risks, and the organizational steps needed to add a GenAI chatbot to your website. Get an alert when part 2 publishes by so you don’t miss it!
Erin Brenner is the owner of Right Touch Editing, a boutique editorial agency that specializes in helping small and midsize businesses to be more engaging with their audiences, more persuasive in their marketing, and clearer and more precise in their communications.
Erin is also the author of The Chicago Guide for Freelance Editors: How to Take Care of Your Business, Your Clients, and Yourself from Start-Up to Sustainability, Marketing Yourself Guide (with Sarah Hulse), Copyediting’s Grammar Tune-Up Workbook, and 1001 Words for Success: Synonyms, Antonyms & Homonyms. She is an Advanced Professional Member of the Chartered Institute for Editing and Proofreading and a Full Member of ACES. Follow her on LinkedIn and Bluesky.