Launching your first assistant? We’re here to help you get going, but before we dive in, it’s important to take a step back, think about your overall goals, and plan out how you’re going to start.

Most of our customers have a vision of a truly personalized customer facing assistant that can work on any channel and handle any request at any stage of the journey. They want an assistant that can deflect questions from reaching support agents, speed up task completion, handle sales inquiries, capture leads, and reduce the cognitive overload of navigating other customer service channels.

If this sounds like your vision, our goal is to get you there, but this will likely require system integrations and team collaboration across organizational boundaries. In other words, time. So instead of focusing on all the details at once to go big from Day 1, we want you to start small, prove value, and THEN scale up your assistant.

Start small, launch fast

Yes, we think you should start small and launch your assistant quickly. Why? Because we’ve seen this time and again: many of our customers will spend months building and testing an assistant that tackles tons of topics. And once they launch, they find that most of the assumptions they made about how their users ask questions or want to interact were wrong, so they have to rebuild their assistant from the ground up all over again.

The reality is that humans use language to ask for things in unpredictable ways, especially when they talk to virtual assistants (which, by the way, is very different than the way they talk to humans). Although we can definitely help you make sense of your existing human-to-human customer service logs (more on this later), you really need to see some real-world assistant usage before you’re able to grow and perfect on it.

Organizations that started with a single domain of topics, launched quickly, and then iterated afterwards have always ended up in a better spot than those who spent months in the pre-launch phase.

You’re probably thinking, “Sure, but I don’t want to launch a rushed bot that will make my business look bad!” You’re right: a poorly built bot will exacerbate your problems instead of helping to solve them.

That’s why we have a number of safeguards built into watsonx Assistant to prevent dead ends and flat interactions. And that’s also why we always recommend a 3-stage, “walk, run, fly” approach to building, launching, and scaling your assistant.

Start by walking (not running, not flying)

We recommend that the first version of your assistant operate on one channel (e.g., your website or over the phone) and within one initial domain or department of your business (e.g., payment support). The focus of your launch should be on unifying the self-service and human-based customer service experience within this domain. More specifically, you’re focusing on one or both of the following value areas:

  • Make it easier for people to get the help info they need on the first attempt
  • Deflect common informational requests from reaching human agents

Once you’ve launched your first assistant and have started walking, you can continue to improve and grow the assistant into the run and fly stages. But for the purpose of this guide, we’re just going to focus on the steps you take to learn how to walk. Are you ready?

Planning your assistant

Once your initial assistant is launched, the model will look something like this:

Before you start building, it’s important to plan out the key parts of your assistant so that you can hit the ground running. The good news here is that you only need to make five decisions before you start building.

Let’s go into more detail on each step of the planning process. Once you’re finished planning, we’ll move on to actually building in the following post.

1. Starting channel

Before deciding which specific topics to build into your assistant, the first thing to decide is where you want to launch it. The major channels of communication with your customers are via phone or via your website. Since you’re just getting started here, let’s not go overboard — you’re not going to put this on your company’s home page, nor should it be immediately answering all the calls coming into your call center. We’ll get there, but again, start small and iterate.

Let’s break down where to start, depending on the channel you’ve chosen.

Your website or app

watsonx Assistant comes with an easy-to-embed web chat widget that most of our customers use as their initial channel. If this is your channel of choice, decide on which pages it should appear.

To that end, identify the pages where your customers most frequently ask questions from your customer service team. Another place to start is your self-help content like FAQs. It would be especially good if the questions asked on those pages are similar — i.e., lots of people are asking the same thing.

Tip: For page traffic, we typically see 1 to 5% of visitors use a new assistant in a given calendar month. So choose a set of pages that have enough traffic to give you enough data to improve the assistant. A few hundred assistant users per month is a good starting point, meaning you should launch on a set of pages that have at least 5,000 to 10,000 monthly unique visitors.

Over the phone

For the phone, it’s basically the same story as the web. You likely have an existing interactive voice response system with a tree-like branching structure (“press 1 for billing, 2 for payments”). You can integrate watsonx Assistant with your existing system to automate only one area of your overall IVR experience (e.g., press 2 for payments, then Watson handles the rest). Again, you want to start small to prove value incrementally.

2. Starting domain

At this point, let’s not get bogged down analyzing historical chats and data. Your goal should be to quickly find an opportunity to both showcase your assistant’s value and get comfortable building and deploying. So, how do you quickly and easily decide where to start? We can offer a couple of choices to help speed you along:

  1. Mine your customer service team for all the questions/scenarios they handle all the time, and look at them through the lens of potential for automation. Their answers will give you a clear idea of where to focus your efforts first.
  2. Choose from our pre-built actions templates to begin your build process. These already start you off with a fixed domain/topic framework that you can use as is, or customize to address the unique characteristics of your use case and organization.

Tip: Once you’ve chosen a domain, be sure that it aligns to the part of the channel (website or phone) that you can actually control. For example, don’t choose to automate billing support questions if the billing team won’t let you add the web chat client to the billing web pages.

Remember that you’re not going to build an assistant that can act on a user’s behalf to, say, look up recent transactions. We’ll get to that in later stages. You’re looking for questions that can be answered concisely with text and some links.

3. Frequent topics

As you can see in the “walk, run, fly” approach above, an assistant really shines when it can get jobs done for people faster and more personally than alternative methods. But before you get to that “walk/run” world, your assistant must have access to concise answers your customers are seeking today.

Now that you’ve selected a domain, it’s time to figure out which specific questions you want to teach the assistant to handle. We recommend using your assistant to automate answers to the top three to five informational topics that your customer service team encounters on a daily basis. To do this, you’ll build explicit action flows for each of the problems you want to solve for your customers.

Tip: Our pre-built templates can help you speed through this process and get up and running fast without having to put on the thinking cap.

For all other topics within your domain, you’ll either provide responses by using the search skill to find answers in your existing help content or escalate the conversation to a human agent.

4. Content sources

In addition to the top three to five topics, you should take a quick inventory of all the help content (product information, knowledge articles, FAQs, etc.) available to your customers today within the domain you’ve chosen. You’ll use this inventory when you set up the search skill to retrieve results.

You might be wondering, “When do I search existing content as part of a conversational flow?” To answer this, start by imagining you are the end user of this assistant. Your goal is to get what you need as fast as possible. Some things that slow you down might be:

  • Reading a long help article for 10 minutes and finding what you need at the bottom
  • Having to hunt around and read multiple help articles before finding what you need
  • Waiting on hold for human agents
  • Hitting dead ends and having to start over on another channel

As you build your assistant, for a given topic, you need to decide which resolution method (action, content, or human) would be the most expedient and complete.

5. Handoff strategy

Your customers hate hitting dead ends when they’re trying to get something done with your business. This means that 100% of the questions need some method of resolution. It’s likely that, in this first launch, you’ll be able to handle 40–70% of questions with the assistant itself between your top topics and other content sources. So what about the rest?

For any other on-domain questions that come through your assistant, you’ll want to escalate that conversation to the correct human agent who can handle the problem. For now, you just need to figure out what your escalation strategy is going to be. This will vary depending on the channel you’ve chosen.

Your website/app

If you’ve decided to launch on your site first, you have three options for escalating to a human agent, in order of seamlessness:

  1. In-line (recommended)
  2. Email
  3. Phone number

With the in-line option (shown above), you can directly escalate to a human agent in your existing contact center tool without making the user leave the web chat widget. This option provides the agent with full context of the conversation that just happened. We provide integrations with prominent service desk tools, or you can bring your own and connect it up:

With the email or phone number options, this is as simple as providing the user with the agent’s email address or phone number. While this is easy to set up, we don’t recommend this approach, as it forces the user to hop to another channel and then restate their problem to an agent. No one loves a disconnected experience like this.

But sometimes it’s hard to get internal agreement to connect the assistant directly to the appropriate contact center tool. If this happens to you, just start with the email/phone number route to prove value and then work to convince people with results.

Over the phone

If you launch over the phone, you only have one option for reaching a human agent: transferring the phone call to someone who can help. There are a lot of patterns for integrating into your existing telephony and contact center infrastructure, but the good news is that we have most of them covered — especially if your contact center/existing IVR can transfer calls over the SIP protocol (we like SIP). We won’t get into the technical stuff now though. Save that for the build phase.

Wrapping up

Hopefully by now you have a lot more confidence about how you’re going to start small and launch fast! Remember, before you continue to the next step, building your assistant, you should have these five key planning decisions made:

If so, let’s keep moving. Next up, Part I: The Build Guide!

If you’ve begun building your assistant, the Learning Center, located in the upper right navigation menu of the product homepage, contains interactive tours, feature articles, and product documentation. It’s everything you need to get up and running, consolidated in a single drop-down menu.

Was this article helpful?

More from Artificial intelligence

A new era in BI: Overcoming low adoption to make smart decisions accessible for all

5 min read - Organizations today are both empowered and overwhelmed by data. This paradox lies at the heart of modern business strategy: while there's an unprecedented amount of data available, unlocking actionable insights requires more than access to numbers. The push to enhance productivity, use resources wisely, and boost sustainability through data-driven decision-making is stronger than ever. Yet, the low adoption rates of business intelligence (BI) tools present a significant hurdle. According to Gartner, although the number of employees that use analytics and…

The power of remote engine execution for ETL/ELT data pipelines

5 min read - Business leaders risk compromising their competitive edge if they do not proactively implement generative AI (gen AI). However, businesses scaling AI face entry barriers. Organizations require reliable data for robust AI models and accurate insights, yet the current technology landscape presents unparalleled data quality challenges. According to International Data Corporation (IDC), stored data is set to increase by 250% by 2025, with data rapidly propagating on-premises and across clouds, applications and locations with compromised quality. This situation will exacerbate data silos, increase costs…

Where to begin: 3 IBM leaders offer guidance to newly appointed chief AI officers

4 min read - The number of chief artificial intelligence officers (CAIOs) has almost tripled in the last 5 years, according to LinkedIn. Companies across industries are realizing the need to integrate artificial intelligence (AI) into their core strategies from the top to avoid falling behind. These AI leaders are responsible for developing a blueprint for AI adoption and oversight both in companies and the federal government. Following a recent executive order by the Biden administration and a meteoric rise in AI adoption across…

IBM Newsletters

Get our newsletters and topic updates that deliver the latest thought leadership and insights on emerging trends.
Subscribe now More newsletters