Flows are the most important concept within AI Decisioning. You can think of a flow as a specific marketing initiative such as product cross-sells, onboarding, or win-back. A flow allows marketers to target an audience of eligible users with a variety of content, optimizing toward a set of outcomes.
Target audience: A Hightouch audience that defines which users are targeted by this flow
Action: A template in your engagement platform and, optionally, additional variations such as alternate subject lines, CTA copy, etc.
Outcomes: An important conversion that you are trying to drive towards, like a purchase, and any other intermediary steps such as email clicks or web sessions
A flow should be built in the context of the outcome that it is trying to drive towards, for example:
I want new users to purchase their first product. Which emails will be most likely to lead to a conversion?
I want existing users to refer a friend. Which SMS or push notification will be most likely to lead to a referral?
To create a new flow, navigate to the Flows page and click Add flow in the upper right. You’ll be prompted to select a target audience and a holdout percentage (to measure the incremental effect of this flow). Once created, you’ll need to add actions and outcomes to complete the flow setup.
Once you’ve defined a target audience, actions, and outcomes, you’re ready to initialize your flow. Once you press initialize, it will kick off a series of processes to create and train a custom machine learning model specific to your flow. This process can take up to a few days; the Hightouch team will keep you updated throughout this process to let you know when the model is ready.
Once initialization is complete, you’ll see an option to Start your flow. Once started, the flow will begin sending within 24 hours, or later if you’ve only allowed sends on specific days of the week.
To pause a flow, navigate to the Flows page and press the Pause button on the upper right of the flow you want to pause. This will stop sending all actions associated with this flow.
There still may be lingering messages that have already been scheduled when
you pause a flow that will still get sent over the next 24 hours.
An action represents some specific creative that can be sent to your users. By adding many actions, you’re giving AI Decisioning different options to find the best message for each user. Each action is linked to a certain message, campaign, journey, or template in your engagement platform (depending on which platform you use). AI Decisioning works by triggering individual messages to individual users via your engagement platform.
To add an action, navigate inside of a flow, and press + Add action. Depending on your engagement platform, you’ll have different ways of connecting a Hightouch action to a specific piece of creative in your engagement platform.
For more info on each integration pattern, see
integrations
You can use variables inside an action to test various creative/copy inside of a given action. For example, you may want to vary the URL of the hero image, pre-headers, subject lines, etc. These variables will be passed to your engagement platform in the API call and can be used in your messages via templating languages, most commonly using handlebar tags (e.g. {{firstName}}).
Hightouch has first-class support for recommending specific items to a user from a large catalog of options. Collections can be used for recommending products, events, streaming content, categories, and more. Each action can be configured to recommend a specific number of items. In your engagement tool, you can use these Hightouch-generated recommendations to populate product grids or other personalized lists.
Before collections can be used inside actions, they must first be configured.
Tags are a way for AI Decisioning to understand and learn trends across your messages. By default, the AI does not have any concept of what is inside of each piece of creative. If you have two very similar emails that both mention “back to school sale”, the AI doesn’t know this similarity automatically. In this case, you could add a tag promotion = back to school in order to tell the AI that these messages both mention back to school. This way, if the AI finds that user A responds very well to one of the back to school emails, it may want to try the other one as well.
Tags are especially useful when you have a large catalog of actions that may have overlap or common themes. Some common examples of how to use tags are:
Categorize actions into different types (category = promotional or educational)
Categorize actions based on the creative content (creative = GIF, image heavy, or text heavy)
Categorize actions based on the products being promoted (products = formal wear, shoes, etc.)
On the guardrails tab of an action, you can configure additional guardrails specific to this action. Specifically, you can configure:
Send limits: the maximum number of times this action can be sent to the same user
Start and end dates: if specified, this action will only be sent between these dates, otherwise it is eligible to be sent whenever the action is enabled
User eligibility: which users are eligible to receive this action
Once an action is configured, you must enable the action for it to begin being sent. Once enabled, you can disable an action by clicking on the … at the top right and clicking Disable.
Outcomes inform Hightouch on the results of each action that it takes for an individual user. By understanding which actions lead to positive outcomes, Hightouch can start to understand patterns where certain types of users respond best to certain actions.
To add an outcome, you can select any event model that has already been defined in your Customer Studio schema. You can optionally apply additional filters, for example, if you only want to count purchases in a specific category that you’re promoting with this flow.
Outcomes are assigned a classification based on how you want the AI to interpret this outcome. For example, a purchase is usually classified as Best whereas add to cart might be Very good and an email click might just be Good. On the flip-side, you can also define negative outcomes such as email unsubscribes.
The Weight field can be used to weight outcomes based on a certain attribute of that event like a purchase amount. This way the AI understands that a $100 purchase was a much better outcome than a $5 purchase.
Finally, for engagement events like an email click or a push click, you can specify which field contains the campaign ID from your engagement platform. This is to ensure that the AI is accurately associating an email click with a specific action.
The campaign ID field should contain just the raw id without any other text.
Ready to get started?
Jump right in or a book a demo. Your first destination is always free.