It’s become something of a cliché to say that chatbots are the new apps, but the notion is no longer fanciful. Last March, Microsoft released its Bot Framework, and the following month Facebook announced its support for bots in Facebook Messenger. We know there are more than 34,000 chatbots running on Facebook Messenger, with companies as diverse as FedEx, Capital One, and Taco Bell lining up to use voice-activated interfaces and conversational virtual assistants.

While the use of chatbots has exploded in the past 12 months, the technology in many cases has not kept pace — chatbots running on Facebook Messenger have failed to fulfill as many as seven out of 10 user requests.

When it comes to banking and finance bots in particular, industry experts claim that many bots are not ready for prime time, but are being launched anyway. Banks are facing a Catch-22: They need to utilize innovative approaches to better serve their customers, but if they introduce a lackluster technology, they risk annoying their customers by delivering an uneven experience.

We all recognize that bots are still a new thing and will take time to mature. At the same time, there are several key capabilities they must accomplish in order to be seen as useful and trusted by customers when it comes to their finances.

1. Are you with me?

Many of today’s bots utilize simple natural language processing (NLP) that starts falling apart pretty quickly as the conversation progresses. This can be a frustrating experience, and certainly not one that would make the bot easy to trust. The bot needs to be able to carry on an intelligent, natural dialogue while maintaining the context of the conversation. This includes being flexible enough to process out-of-context replies and adapt on the fly to changes in the conversation.

2. Get to know me!

If all the information a bot provides is generic, it will be seen as shallow, not engaging, and most likely, not very helpful. A smart bot can understand individual customer situations and needs and provide content, advice, and offers that are tailored to each customer at any given moment.

3. Go real-time or bust

If the bot presents recommendations based on data that is outdated and inaccurate, customers will lose trust in the information and stop using it. For example, if you make a payment that resulted in an overdraft of your account due to outdated information provided by the bot, you’re obviously not going to trust the bot again with your finances. Bots need to be powered by real-time data to avoid these sorts of scenarios.

4. Reuse, reduce, recycle

As the number of channels used by customers continues to increase, creating new content repositories for each channel becomes a challenge that most banks can’t meet. At the same time, customers expect the bot channel to be at least as knowledgeable as existing channels. It’s crucial for the bot to be able to access content created and maintained in digital repositories used by existing channels, such as websites and FAQs.

5. Go beyond ‘what’s my balance’

If all a bot can do is show account balances, it’s not going to increase customer engagement. The bot must have deep enough knowledge to be helpful, at least on some specific topics (e.g., save more money, manage rewards). That’s why bots have to be purpose built — you can’t expect a bot to know everything, but you want it to have deep knowledge on the issue you’re trying to get help with.

6. Follow me wherever I go

Customers expect a consistent experience across the digital landscape, through online apps, Facebook Messenger, Amazon’s Alexa, and other channels. A conversation might start in Facebook Messenger, move to Alexa, and continue in the bank’s mobile app. Producing bots that sit in their own operational silos is a real danger for banks — they must integrate their bots into their overall omni-channel strategies.

7. Never stop learning

To deliver a personalized experience, the bot must get to know customers better through ongoing interaction. It should be able to take cues from the way individual customers react to information and advice provided by the bot as well as feedback given by the customer.

8. Lean in

Usage data shows that almost half of all bots are only used once, never to be contacted again. This is not a situation of “build it and they will come” — you can’t sit back and wait for the customer to approach your bot. To get customers in the habit of conversing with your bot, it needs to proactively reach out to them with information, insights, and advice presented at the right time and place, based on predictive analytics of individual customer needs.

In today’s highly competitive environment, banks and other financial institutions cannot afford to antagonize or alienate their customers. By following these 8 best practices, banks can create useful bots that customers trust and actually enjoy using.

Ron Shalit is director of product innovation at Personetics, a leading provider of personalized digital guidance solutions for the financial industry.