Ready to learn how to get predictions into your stack with Faraday? If you haven't already, check out our basic concepts guide for a high-level overview of what to expect in this tutorial.
In Faraday, you can approach your predictions in two ways:
- A guided experience through our most common use cases via the recipe wizard
- Manual configuration for needs outside traditional use cases
We'll dive into both of these methods using a use case of likely buyers, with a CSV file as both our method of connecting data and deploying predictions since a CSV can easily be uploaded to your preferred activation platform.
- Navigate to the recipe wizard.
- In the recipe search bar, search for likely buyers in CSV, and click to open the recipe wizard for this recipe.
In the likely buyer recipe wizard, we can see that all we need for this recipe is to select a cohort–or a group of individuals that have all taken the same action, such as a purchase–as your customers cohort.
In the cohort selection area, we'll click create a new cohort to create our customers cohort.
Next, we're prompted to either use data we've already uploaded, or upload data now. If you already have customer data uploaded, you can choose the event that defines your customers–orders, for example. For the sake of this tutorial, we'll select I still need to connect data so that we can plug in our CSV.
Once making the selection, we'll choose CSV, then click next.
Here, we can use the file picker, or drag and drop, to select our customer data. With our data selected, we'll click next to begin the upload, and move on to data mapping.
When the file is done processing, the new cohort's data mapping window appears. Here, via the dropdowns, we'll make sure each column in our customer data matches what Faraday expects in the field in dataset column (email matches to email, etc). For any properties that we don't have in our data, we can leave those rows blank. We'll do this for both our identity mapping–fields that define the people in our data–and event mapping–fields that define specific events in our data, like order data and value.
- After we've ensured the columns are mapped correctly, we'll click create cohort, which will not only create the cohort we need to complete this likely buyers recipe, but will allow us to use it in other recipes as well by defining it in our dataset.
With our new customers cohort selected, a green checkmark at the top indicates our recipe checklist is complete. At the bottom of the recipe, we can see that when we confirm the recipe, Faraday will create a predictive outcome based on our business goal of finding more likely buyers, a scope–or a pipeline–to connect that outcome with the customers cohort we created, and lastly, a target deployment for where our predictions are going.
Finally, we'll click proceed to complete the recipe, which will bring us to the processing screen for our new pipeline.
When the pipeline is finished building (indicated by the loading bar at the top), we'll click the download button to download our list of hashed, likely buyer data for upload to an ad platform, and we'll toggle on the pipeline in the upper right to ensure our predictions are always up-to-date.
In order to get predictions into our stack, we'll need to perform the following steps when configuring manually as opposed to using the recipe wizard.
- Create a dataset based on customer data
- Create a customers cohort from the dataset
- Create a predictive outcome for likely buyers based on the customers cohort
- Create a pipeline to combine everything into a deployment
- Navigate to Datasets, where we can upload our customer data CSV.
- Click new dataset in the upper right to create a new dataset.
- In the connection selector, we'll select CSV, as we're using a CSV as our data source for this tutorial, then we'll click next.
- Next, we'll drop our CSV into the file picker, name our dataset something relevant like "customers," and click create dataset. We'll be taken to our new dataset view and presented with a loading bar. After a short time, the loading bar will complete and enable us to define our identities and events.
- With our dataset created, our next step is to define the people in it through identity sets by clicking + add identity set.
- Here, we'll give our identity set a name. Via the dropdowns, we'll make sure each column in our customer data matches what Faraday expects in the field in dataset column (email matches to email, etc). For any properties that we don't have in our data, we can leave those rows blank.
Only lowercase letters, numbers, and underscores can be used as names for technical reasons.
- Next, we'll do the same for events. Since each row in our example data includes a single transaction event where a person became a customer, we'll include that as the timestamp event property, name our event transaction, and click finish.
- Lastly, we'll click save dataset to complete the dataset.
With our dataset created, we can move on to create our customers cohort.
- Navigate to Cohorts, where we'll create our customers cohort.
- Click new cohort in the upper right to create a new cohort.
In the new cohort screen, we'll click add event to select the event–or the action each person in this cohort has experienced–that we created in step 7 in the previous dataset section. Once we've selected the transactions event we created, we'll click next, then finish.
Advanced options, such as recency, frequency, and value, can be configured prior to clicking finish on the new cohort.
- Next, we'll name our cohort customers, and click save cohort.
Now that our customers cohort is created from our new dataset, we can create our predictive outcome.
- Navigate to Outcomes, where we'll describe the business goal we want to achieve.
- Click new outcome in the upper right to create a new outcome.
- Since the business goal we want to achieve is finding more likely buyers, we'll make the following selections:
- Eligiblity: everyone, as we are simply looking to find more people to market to.
- Attainment: customers, as our customers cohort is who we want the people we target to look like.
- Attrition: empty, as defining people who fail to become customers isn't relevant to a likely buyers outcome.
- Finally, we'll name our outcome likelihood to buy and click save outcome to finish.
With both pieces of our predictive recipe ready–our cohort and outcome–we can put them together in Pipelines.
- Navigate to Pipelines, where we'll create our likely buyers pipeline.
- Click new pipeline in the upper right to create a new pipeline.
- To put the pieces we built together, we'll make the following selections:
- Payload: the likelihood to buy outcome we created.
- Population to include: everyone, as we want to score the entire population on how likely they are to buy.
- Population to exclude: empty, as we don't want to exclude anyone from these predictions.
- Next, we'll click save pipeline to confirm the pipeline, after which it will begin building. Once the pipeline is complete, the system sends an email notification, but in the meantime we can create the deployment so that we're ready to enable the pipeline as soon as it's available.
- In the pipeline, under deployment, click add under Faraday hosted.
- Since we're planning to use these predictions on an ad platform, we'll leave the data hashed for privacy and for ease of upload into ad platforms. Optionally, we can select human friendly so that our prediction columns are easily recognizable. For this example, we'll leave it on the default of machine friendly.
Click next, skip the advanced settings, and click finish to wrap up the deployment.
When the pipeline is ready, we'll finish up by clicking the enable toggle in the upper right to prepare our predictions for download.
And that's it! Our pipeline is finished, and we're free to click the download button to take our CSV of likely buyer predictions to our favorite ad platform.