Best Practices

Training Data

As Adapt is in Beta we are still in trial and error mode with getting the best results from the training data provided. We have found that the more time and effort you put into adding your training data, the better the results are when chatting with the bot. This involves cleaning the data if copying and pasting, removing any redundant training, spreading the training data over multiple entries and having meaningful intents from each training entry.

For instance if you were to train a bot on a menu, we recommend dividing up the menu into separate training sections with detailed intents. This allows our backend to process the training more efficiently and only utilize the training data that is relevant to the question. This also uses less tokens.

Privacy

By default all bots are set to Private when first created and must manually be switched to public to be accessible by anyone else. At this time there are no access controls around publicly sharing a bot, so if the bot is made public anyone with the link will be able to chat with it.

We recommend only making a bot public if the training data is NOT sensitive. If the bot is intended to be used by the API we recommend leaving it in Private mode.

Use Cases

As Adapt is so new we are still understanding all the ways the platform can be utilized. So far during our testing we have found that it is excellent in digesting and understanding instructional guides, troubleshooting information and product knowledge.

We have imagined through the use of a shareable link or QR code that it can be used to assist customers in many different ways. For instance, if a home coffee machine suddenly had a flashing light appear or a beeping noise, the customer could simply scan a QR code on the product and ask the AI bot what this might mean. Adapt will then be able to draw only on it's training data to find the answer. It can also provide the customer with relevant images to support the answer.

Another example using the API could be a workflow to respond to emails automatically. Adapt could be integrated with other no code solutions like Zapier to create rich automated email replies based off of it's training data.

We are very excited to see the kind bots our customers come up with and would love to hear any amazing ideas. We especially think the API will have some exciting potential when it comes to automating tasks with AI.