Consumers are ready for chatbots. A recent study by DigitasLBi found that almost 40 percent of consumers would make a purchase from a chatbot. Even more impressive, nearly 60 percent of consumers would engage with a chatbot if it meant receiving coupons or special offers.

So bring on the bots, right?

Not quite. The problem is, chatbots might not be ready for consumers. On Facebook’s Messenger platform, chatbots have successfully fulfilled a paltry 30 percent of user requests. Yikes. Online fashion retailer Everlane, one of the first two companies to use Messenger for business purposes, pulled the plug on its use of the platform to return to good old-fashioned email. Facebook decided to pump the brakes on its chatbot program, and chatbots now are largely more famous for their flops than their victories.

In many ways, the anticipated chatbot boom has been more of a blip. While countless developers rush to build chatbots to ensure they’re not behind the curve, it’s important to consider whether your chatbot should exist in the first place.

Can you hear me now?

Many people assume the failure to launch that characterizes the first wave of chatbots stems from a communication failure.

A recent survey found 71 percent of millennials were interested in chatbots, though 55 percent of respondents said chatbots needed better accuracy; 28 percent of respondents thought chatbots could improve if they had more natural-sounding conversations. Natural language processing (NLP) would seem to be the missing ingredient.

Adding NLP to any technology or service sounds like a smart choice. In theory, bolting a snazzy new feature onto an existing service feels like a valuable addition. In reality, NLP multiplies value (or magnifies existing value). If the existing value of something is zero, multiplying that value by NLP still gives you zero value.

That’s not to say chatbots and NLP go together like peanut butter and mustard. In fact, this dynamic duo could be the ideal way for some developers and business leaders to engage their customers. Before you dive into NLP — or chatbots, for that matter — nail down the ultimate value you want to provide. People will use your chatbot to accomplish a certain task, whether it’s checking the weather, ordering a new hat, or listening to a favorite band. NLP could spark an amazing user experience, but it’s essentially useless unless it helps people accomplish your task.

Chatbots are essentially a user-friendly interface for back-end systems. EBay’s ShopBot, for instance, is a shopping assistant that delivers personalized recommendations while users chat with their friends. EBay has always had a powerful search engine on its website and apps, but ShopBot makes it possible to search directly from a chat interface. The chatbot lowers friction for consumers who aren’t sure what to buy and makes it easy to complete a sale, because it accesses the back-end services.

Building a better bot

While chatbots haven’t been quite as revolutionary as some predicted — not yet, at least — plenty of useful chatbots are out in the wild. And yes, some of them deliver incredible results thanks to NLP. For instance, H&R Block has used IBM Watson to teach its online system the intricacies of tax code, allowing it to suggest deductions and credits to users as they fill out their taxes.

But before you set off to become a millionaire through your brilliant combination of NLP and chatbots, ask yourself four simple questions:

1. What task will customers accomplish using your chatbot?

Before you do anything else, set a concrete goal for how your chatbot will serve users. Think about common pain points for your company or in your industry, and brainstorm ways your chatbot can offer solutions.

By thinking several steps ahead to your ultimate goal, you’ll be better able to judge whether your chatbot can accomplish that mission. Note that your customers might need some convincing that your chatbot is actually useful: While 79 percent of marketers think customers are excited or very excited about bots, about 50 percent of consumers said they’re either somewhat unexcited or very unexcited by the technology.

2. How do they currently accomplish that task?

Once you have a goal, look at existing solutions. By extension, consider whether a chatbot or NLP would improve that process. According to research by Gartner, a third of all customer service interactions still require another human to perform the desired task. NLP might help users feel as if they’re interacting with another human, but it won’t change the functional capabilities of your chatbot. You don’t need to reinvent the wheel, but try to make sure you’re offering a better — or at least different — solution.

3. Do you have the necessary back-end services?

If you don’t already have an API that powers your existing interfaces, then adding chat will only add more technical work to your plate. APIs can drive revenue for your company (just ask Salesforce, eBay, or Expedia), so focus on implementing them first. Look at existing APIs to see what you can fold into your system, or work to build your own from the ground up. You wouldn’t try to build a house without a solid foundation, and you shouldn’t develop a chatbot without a strong back-end system.

4. Why is your chatbot better than other interfaces?

This is a big one. You can’t expect someone to use your chatbot unless it offers a much better or more unique way of accessing information. Your bank might let you pay bills and transfer money from a chat client, but it probably shouldn’t let you open a new account from that interface: Chat doesn’t feel secure enough for that sort of interaction. Identify the context in which your customers might prefer a chatbot to another solution, and focus on optimizing your chatbot for that situation.

Chatbots and NLP aren’t magic. They’re new tools we can use to build incredible customer experiences in a new environment beyond the web and apps. But simply stuffing a chat interface into your existing website search won’t provide much value. Always consider what you want the customer to accomplish first — the rest will follow.

Josh Marinacci is the head of developer relations at PubNub, a data stream network for mobile applications.