Skip to Content

The Causal AI conference is returning to London on 24 Sept-24, and it's bigger and better than ever.

Register Interest

Next-Best-Action with Causal AI

Enterprises don’t want to just predict the future, they want to shape it. At causaLens, we build solutions that enable organizations to make data-driven decisions that optimally shape their future, going beyond mere predictive analytics. These solutions are powered by Causal AI, a new category of machine intelligence that can reason about the world the way humans do, by analyzing cause-effect relationships.

Action optimization is one among a set of powerful Causal AI tools that enable organizations to make more effective plans and better decisions. Action optimization identifies the most resource-efficient combinations of actions that you need to take to reach your goals. We illustrate how it works with a specific example from sales and marketing.

Action Optimization — an example

Suppose that your marketing team has identified that one of your business’s key demographics (18-25 year-olds) are at risk of “churning” (abandoning the service). If you can increase retention by just 5% that is likely to translate into an increase in profits of 50%.

So what can you do about it? You have several options: implement targeted sales outreach, run ads (on smart TV, paid search, out-of-home, social media, and so on), discount prices, or offer promotions. Which of these actions should you take, and how much should you invest in each?

This is where action optimization comes in. There are several constraints that the technique factors in. 

Firstly, Causal AI is capable of ruling out courses of action that aren’t feasible or possible. For instance, whilst macroeconomic factors like GDP do significantly influence customer retention (think of Netflix’s recent announcement of massive subscriber losses, partly caused by inflation and the cost-of-living crisis) — but a business can’t meaningfully change GDP. 

The machine then evaluates the feasible actions based on how reasonable and cost-efficient they are. For instance, giving away the service for free may induce the customer to renew but is not economically viable for the business — reaching out to the customer with a targeted sales call may achieve the same result for a fraction of the cost. Generally speaking there are lots of ways to persuade a customer to renew (in the technical AI literature this is sometimes called “crossing a decision boundary”). Causal AI can search through these actions to find the most efficient pathway across the decision boundary. 

Finally, Causal AI can uniquely factor bespoke business constraints into its recommendations. Perhaps for strategic reasons you can’t discount prices by more than 10%, because doing so would undermine customer confidence in your pricing and compromise margins. No problem — action optimization will take account of these hard constraints. 

This image has an empty alt attribute; its file name is algorecourse%20(1).gif
The causaLens platform allows all users in the enterprise to optimize actions at the click of a button.

Correlation-based ML cannot optimize actions

This type of optimized action planning is only possible with a Causal AI model. Current correlation-based machine learning does not capture relationships between different actions, simply assuming that they do not interact with and impact each other at all. More formally, this “independence” modeling assumption requires that a change in one feature has no effect on the value of another.

However, we know this is rarely the case in practice. An ML system may be able to spot that an increase in the overall number of customer interactions is likely to influence a customer to renew. But it misses the obvious fact that an increase in the volume of sales calls or emails result in an increase in interactions, and so it is liable to naively recommend needless additional actions (call and email the customer).

This image has an empty alt attribute; its file name is 20220519dfclCausalGraphV01-1-1024x640.png
Causal AI understands the dependencies between features, and hence it can anticipate all the complex downstream effects of a given course of action. Correlation-based ML has to assume that all actions are independent of each other, a clearly unrealistic modeling decision.

It’s generally incorrect to assume that actions are independent of each other — correlational machine learning models that assume independence recommend disproportionate (and therefore expensive!) courses of action. 

Traditional correlation-based methods can only show the business or user what variables would need to change to achieve their goals, but not how to act. Causal AI provides the best actions for reaching business targets and helps the user to better understand their problems. 

We have illustrated how action optimization works by focussing on a marketing example. But the technique applies generally to a very wide range of use cases. To name a few examples: we have applied action optimization to help consumers understand their options when they’re refused a mortgage; the technique facilitates complaint resolution for victims of fincrime and fraud; and in a very different context altogether, action optimization has been applied to reduce vaccine hesitancy during the COVID-19 pandemic. 

Our experts will be happy to discuss your business use cases and explore how Causal AI can address your business challenges. 
Action optimization builds on a new technique known as “algorithmic recourse” in the technical AI literature.

causaLens provides next-generation AI for the enterprise. Our technology harnesses Causal AI to build models that are not just accurate but are truly explainable too, putting the “cause” in “because”. Find out how we can help you to better understand and explain your business environment.