On average, companies spend over a quarter of their marketing budget on content, while the most successful spend closer to 40%.
However, creating the content isn't enough, and finding the right outlet for distribution is key.
The first step to distribution is making sure the content is relevant to your target audience, but it should also be tailored to the outlet you are using for distribution. One popular outlet for blog content is Medium, which has countless "publications," and it's crucial to find the right one for your piece of content.
Traditionally, marketers would do this manually, but this process can be automated with AI.
Content creators can send a title through a classification model that will then recommend the best publication for it. The database is created by analyzing historical published articles, and then matching the style of a new title with the right publication.
This way, marketers don't have to guess which outlet to use for distribution, and can instead completely automate the process. While we'll dive into how to build an "outlet prediction" model in this article, there are many more AI marketing use-cases, such as churn prediction, augmented lead scoring, and sentiment analysis of consumer text.
To get started, sign up to Akkio for a free trial. Just like any supervised machine learning task, our first step is getting historical data and selecting a column to predict.
For demonstration purposes, we’ll be using this Medium articles dataset from Kaggle. This dataset consists of over 6,500 Medium articles scraped from various Medium publications, across topics like data science, entrepreneurship, self-improvement, writing, and marketing.
We’ll get started by creating a new flow in Akkio, and hitting “Table.” We can then upload the CSV from Kaggle.
After the dataset is connected, Akkio will automatically figure out the data types for each column, and also show some basic details, including the number of rows and columns. You’ll also see a scrollable preview of the dataset.
Next, we click on the second step in the AI flow, which is “Predict.” Under “predict fields,” you can select the column to predict, named “publication.” We’ll ignore the rest of the columns in the dataset, since we want to keep the model very easy to use in production. Then just hit “Create Predictive Model,” and you’re done.
Once the model is created, you can get a quick overview of your new predictive model.
When you click “See Model Report,” it will open up a new tab in your browser with an interactive report which allows you to quickly understand what was predicted for each field.
In our example below, we achieve nearly 75% accuracy from just one column of text input.
Let’s dive deeper into the model to better understand this metric, via the model report. The model report highlights the following:
You’ll be able to see specific details like the accuracy, model type selected (in this case, Sparse Neural Network was used), and so on. Below, we can see that some of the mispredictions were in cases with very similar classes, such as “Towards Data Science” instead of “Data Driven Investor.” In other words, our model performed very well with such a challenging classification task.
The best way to improve model accuracy is to get additional, high-quality training data. With Akkio, you can easily merge on a new dataset, such as with another source of feedback data. You can also try increasing training time, which may help increase accuracy, particularly if you have a large dataset to begin with. In our example above, we use the “higher quality” training time setting.
Now that we have our classification model trained on some historical data, let’s deploy it. The final step in the flow is to select where your prediction should be made.
In this example above, we’ve built a model to predict the best outlet to publish on, but we could just as easily build a model and deploy it to classify incoming leads, predict customer churn, optimize emails, or a variety of other possibilities.
This is the final step in the process, and it's where you'll plug your classification model into your system to make predictions. With Akkio, we can easily deploy in a variety of settings, including Salesforce, Snowflake, HubSpot, Google Sheets, and more. We can also make fully custom integrations with the API, or use the no-code automation tool Zapier to integrate with thousands of other applications.
For example, we can hit “Add Step” and then select “Web App” as a highly simple deployment method. Since we trained our model on just input field, or title, that’s all we need to make a prediction. We give our web app a simple name and description, and then hit “Deploy.”
We can then see and share our web app in action, where a marketer can enter a title, and figure out where to publish it, all in an instant. You could also upload a list of titles as an Excel sheet or simple CSV, and get a series of predictions as a result. In the example below, we entered the “The Comprehensive Guide to AI For Entrepreneurs” and got “The Startup” as a result.
With Akkio, teams can scale globally without needing to build or maintain any code or infrastructure themselves. As a result, deployments take moments instead of weeks or months like other machine learning platforms.
Marketing and PR teams can use this approach to find the right outlet for any piece of content they are creating. This is just one simple example of how AI can help automate tasks that were previously done manually. The possibilities for this type of technology are endless, and companies who invest in it will be well-positioned to stay ahead of the curve.
This project can be a stepping stone to more advanced AI marketing automation, such as lead scoring and consumer sentiment analysis.