“Come with me if you want to live” —The Terminator
You’ve probably watched The Terminator featuring Arnold Schwarzenegger and ever since then associated robots with his character.
But robots or “bots” are not just physical walking metal parts; they come in many different shapes and sizes, and we interact with them every day: whether we know it or not.
You may have asked your Google Home or Alexa what the weather was like today or asked Siri for directions. These are some of the most prominent artificial intelligence (AI)-powered bots in the world, but there are so many more.
Some of the most basic bots show up when you call your phone provider and you get this message, “Press 1 for Support” ... That’s an IVR: and it’s a bot. A simple and annoying one but still a bot.
So, what makes a bot?
Most bots we interact with today are user interfaces typically used for messaging via chat or voice. In this article, we’ll go deep with chatbots. Specifically, we’ll explore the strengths and weaknesses of chatbots in a business context and underscore the importance of their swift deployment.
Bots are not a set-it-and-forget-it product. Users...ask questions and take uncharted paths that the bot needs to be trained on.
Bots use conversational interfaces as another way to display software services. For the most part, bots are digital users that live in messaging apps, such as Facebook Messenger, WhatsApp, and Slack or on your website via web chat. Unlike human users, they are powered by software and use the company’s brand voice.
To be clear, the bot is only an interface for the product or service: in the same way a website, a phone call, or an app might serve as the interface to book an appointment.
Why are we talking about bots? Why should you be investing in yet another interface after investing so much in web and mobile?
Here are just a few reasons:
- About 27 billion text messages are sent daily around the world (source: Increditools).
- Over three billion people use WhatsApp and Messenger every month (sources: Meetanshi [blog] and Datareportal).
- In 2021, an estimate of 3.09 billion people used messaging apps. This number is expected to grow to 3.51 billion in 2025 (source: Statista).
Anything that bots can do to reduce call volume and deflect effectively is a massive win.
It’s about reaching your customers where they are. Messaging interfaces are where they spend most of their time and what they are most comfortable with.
When was the last time your son, daughter, or niece called you? I bet the last time they reached out was via text. Messaging gives users freedom through optionality:
- Reply now.
- Reply later.
All these options are possible because messaging is an asynchronous communication medium. How can you leverage it for your business? There are many ways. Let’s review the types of bots and then we’ll cover their use cases.
Defining the Different Types of Bots
Consumer versus Business
Consumer bots are built for 1-to-1 interactions with the purpose of entertaining, educating, or improving our lives. A good example would be DAWN AccuWeather’s Messenger bot.
Business bots are built to facilitate a business process. An example would be getting a customer’s insurance benefits verified to see if they qualify for the services before putting them in touch with an agent.
General versus Domain-Specific
General bots are the biggest ones which I named earlier: Alexa, Google (Assistant), and Siri. They facilitate a variety of different services and rely on large amounts of data to be practical.
Domain-specific bots focus on a single service or a handful of services associated with the brand. An example would be an appointment bot to book a massage.
Business Use Cases for Bots
Now, let’s narrow it down to our potential use cases.
1. Customer support and FAQs are the most common use cases. These are obvious ones, given that the support centers are costly to operate.
Anything that bots can do to reduce call volume and deflect effectively is a massive win. These are, however, some of the most challenging bots to build because they need deep integrations with other systems to answer very specific customer questions and resolve issues without escalating to agents.
2. Business automations allow businesses to streamline certain tasks and workflows, such as asking qualifying questions to a new job applicant. Entire business operations can become more productive with the help of a digital personal assistant.
Marketing automation bots are used in business-to-business (B2B) companies to automate the workflow from an unqualified inquiry to a marketing-qualified lead (MQL). Next, bots can deploy automated messages to nurture MQLs into SQLs (sales-qualified leads) and enroll contacts in a targeted account-based marketing (ABM) program via chat or text.
Marketing bots differ from support bots in that they need to be flexible. Marketing campaigns frequently change to align with the user journey, so marketing bots need to operate in context.
For example, a marketing bot should be able to analyze which ad or keyword attracted a user to your site. It should be capable of supporting chat re-engagement and nurturing (B2B sales cycles are typically long). It should also provide robust analytics that give insights into what the prospect is looking for.
Finally, a marketing bot should be able to integrate conversational marketing automation seamlessly with CRM systems of record, such as Salesforce, Marketo, Eloqua, or equivalent systems.
3. Conversational commerce is a relatively new use case that is gaining popularity in Asia on platforms like WeChat.
From booking a flight to booking an activity at a destination, this can all be done without ever having to download an app. Just open Messenger, search for your airline, and book. KLM is a great example: from managing your booked flights to issuing a refund, their bot is there to help.
4. Bots can also be used to unify systems and erase data silos via third-party integrations. In fact, integrations power every use case because they are truly a necessity.
If you want to send data collected in your web chat to a CRM application like Salesforce or HubSpot, you’ll need to connect your bot to it via an API.
A great example I’ve been a part of recently is enabling verification of insurance benefits within chat to automate the qualifying process of potential patients and then sharing that data with their electronic health records (EHRs). This has saved behavioral health centers, insurance companies, healthcare providers, hospitals, and admissions officers hundreds of hours spent on calls that didn’t convert.
5. Entertainment is another great use case for bots: whether it’s via a game, fun facts, or a quiz don’t underestimate the value these can bring to a wide or niche audience. These bots have been used as great hooks to bring in an audience to generate brand awareness and build rapport with each individual.
This use case is one I’ve worked on a lot from building my own Messenger fantasy sports bot to a lead generation movie quiz for LG Electronics. In both cases, we engaged with over 50,000 users, generated over five million messages, and captured 37,000 emails which were then nurtured via chat and email using custom attributes collected with the bot.
6. Brand bots is the last on my list. I’m adding this one because even though it doesn’t drive direct ROI as some of the other use cases, it helps build your brand and expose it wherever your audience is.
An example of a brand bot would be a grocery chain like Whole Foods sharing recipes via chat based on certain tastes and foods a user may pick.
Many brands have invested in custom mobile apps to deliver this experience. However, they underestimated the effort to acquire and retain users in their apps.
By using messaging as the medium, users no longer have to download an app. They can interact with the brand whenever they want within their most popular apps, like Messenger and WhatsApp, or on the brand’s website. It removes friction, increases engagement and it reduces the effort required to deploy such an experience.
Generative AI and Chatbots
Generative AI machine-learning algorithms, such as GPT-4, are a recent addition to the world of bots and are transforming the field of natural language processing (NLP).
Generative AI uses large language models (LLMs) that can generate human-like text and can be trained to perform a wide range of tasks, including language translation, content creation, and customer support. They have the potential to revolutionize the chatbot industry by providing more advanced and versatile bots that can understand and respond to natural language queries more accurately.
One of the most significant benefits of GPT-4-powered bots is their ability to learn from and adapt to new scenarios and contexts. This means that they can provide more personalized and human-like interactions with users, which can improve customer engagement and satisfaction.
However, these models are still in the early stages of development, and there are concerns about their accuracy and the ethical implications of using AI for customer interactions. Nonetheless, GPT-4-powered bots offer exciting possibilities for businesses to improve their customer service and engagement.
These models are key to enhancing the user experience, so make sure to consider implementing these in your bot strategy. There are many elements to think about:
- Data quality. The quality of the data used to train the LLM is crucial to its effectiveness. The training data should be large, diverse, and representative of the language that your bot will encounter.
- Training time. Training a LLM can be time-consuming and resource-intensive. You will need to allocate sufficient time and resources to train the model adequately.
- Model architecture. Choosing the right architecture for your LLM is essential to its performance. There are various architectures to choose from, including transformers, LSTMs, and CNNs.
- Hyperparameters. Hyperparameters, such as the learning rate, batch size, and number of epochs, can significantly impact the performance of your LLM. You will need to experiment with different values to find the optimal settings.
- Fine-tuning. Fine-tuning your LLM on specific tasks or domains can improve its accuracy and relevance. You may need to fine-tune the model on specific datasets or tasks to optimize its performance.
- Testing and evaluation. It’s essential to test and evaluate your LLM thoroughly to ensure its accuracy, relevance, and usability. You should use a diverse set of test cases and metrics to assess the performance of your model.
- Maintenance and updates. Language is constantly evolving, and your LLM will need to be updated and maintained regularly to remain effective. You will need to monitor its performance and update it as necessary to ensure that it continues to meet the needs of your users.
These are all elements we think about and work with daily at my company.
Bot Strengths and Weaknesses
Now that we’ve identified which use case will solve our challenge, how do we get started? In this section, we’ll go over the strengths and weaknesses of bots, common pitfalls to avoid, and what to look for in a chat provider.
Here are some key strengths:
Bots aren’t real people; they don’t work 9-5. They’re available to help customers every day of the week at any time. And they’re available when most needed and customers love this.
Customers don’t like waiting. It is one of the most frustrating parts of any customer service experience.
Instead, potential customers will rather find an alternative: which means speed could be the difference between closing a customer and losing one to your competitors. Bot responses are automated, so customers receive answers to their questions much faster than via email or phone.
Popularity of Text
As I said earlier, consumers use text communications more often than voice calls, with many people outright preferring to text with a bot than talk to a human.
We’ve seen this be the case when dealing with behavioral health treatment centers. Consumers prefer sharing personal information with a bot over a human.
Consumers like to reply whenever they want, so having a bot on Messenger, SMS, or WhatsApp gives them that freedom. They can answer or ask questions whenever is most convenient for them without having to restart the conversations every time.
A bot can take many shapes even within your website. You can have it set up as a chat widget in the corner of your page, be embedded into the layout of your website, work as a pop-up, or as a stand-alone landing page.
Personalization at Scale
This is by far the most interesting. As your bot exchanges messages with your customers, it collects data that it can use to personalize the customer experience. Different questions and answers can lead to different journeys for thousands of users simultaneously.
Bots aren’t perfect. Here are some weaknesses:
Bots are not a set-it-and-forget-it product. As users interact with it, they’ll ask questions and take uncharted paths that the bot needs to be trained on.
These optimizations are necessary to make sure it’s giving the right information to your customers. If business priorities change, or if you update content that your bot is leveraging as data to answer users, you’ll need to update the bot or your net promoter score (NPS) will take a hit.
Unfamiliarity of Data
Analyzing bot performance is a challenge for anyone new to conversational interfaces. There are many different data points that are different from web or mobile to take into consideration, such as sentiment, fallbacks, attributes, handovers, engagement, and much more.
It takes time to get familiar with these new metrics and their relationships. It’s not quite as cut and dry as a conversion funnel on your website.
Time to Market
Many companies have found it difficult to deploy a bot in four to six months. It takes the right expertise, which very often is not found in-house: which means your team is not comfortable with the build phase and delays the launch.
The important thing is that you launch the bot as soon as possible...and get feedback.
There are usually two reasons delays occur:
- Lack of expertise with the bot and natural language processing (NLP) technologies.
- Not satisfied with the results.
I’ll cover how to overcome the first problem in the next section. As for the second problem, it’s completely normal.
The important thing is that you launch the bot as soon as possible. You need to treat it as a minimum viable product and get feedback. Too many companies get caught in the build trap and end up wasting time building a bot that doesn’t serve their customers.
Now that you understand your use case, and the benefits and risks tied to building your bot, what’s next? You need to find a bot technology provider.
There are many providers on the market and they all bring something different to the table. There is not a one size fits all solution here, it really depends on what you’re looking to achieve.
What To Look For in a Bot Provider
Here are a few things to look for in your bot provider:
If you’re not a technology company with a dedicated team tasked to solve the problem with a bot, I strongly recommend working with a provider that will be very hands-on. This may seem expensive but spending months on a project that you end up killing is much more costly.
This is not a requirement, but I strongly recommend you work with a provider that has domain expertise for many reasons. They will have the niche integrations you need, have solved similar problems for another customer, understand your users better, and you may even benefit from using an AI model that was heavily trained with data from your industry.
Fast Time to Market
If you can find out how quickly they launch bots, this will be valuable because it means your time to value will be faster. If they say it’s a six-plus month process, to me that’s a red flag because you’re likely falling into a build trap and not focusing on delivering immediate value to your customers.
There’s so much more to cover, but this will be a great starting point for you and your team.