We use cookies. You have options. Cookies help us keep the site running smoothly and inform some of our advertising, but if you’d like to make adjustments, you can visit our Cookie Notice page for more information.
We’d like to use cookies on your device. Cookies help us keep the site running smoothly and inform some of our advertising, but how we use them is entirely up to you. Accept our recommended settings or customise them to your wishes.
×

How to Scale Your Orchestration Engine to Handle Omni-Channel Personalization

We all get why omni-channel orchestration is important. It breaks down communication silos, enables cross-sell opportunities, supports customer journeys, and allows marketers to deliver relevant, frictionless and efficient customer service. As a marketer, having a person’s undivided attention is exceedingly rare, so if you have it, make the most of your interaction by presenting something relevant. (Learn more about Decision Orchestration here). 

As your orchestration engine evolves to take on more channels and product lines, it is important to consider the challenges that come with scaling your operations and IT teams to handle the additional scope of the tool. By nature, multi-product and omni-channel means your orchestration engine has a lot of ground to cover. If you do not consider efficient management of your marketing strategies (via business rules, models, etc.), there is a strong risk of ending up with a tool that is too unwieldy and expensive to maintain. This could hamper your business agility and even slow your IT program’s momentum.  

Here are several ways to reduce the overhead and to avoid these risks:  

Use a Self-Service Tool and Keep the Strategy Close to the Business 

Having the ability to update marketing strategies quickly so that your business can respond to changing market conditions should be a critical objective in your omni-channel program. When you consider the overhead that may be associated with changing your strategies, the typical process that exists in most organizations can be too slow. Consider the following: there might be requirements gathering, development, testing, defect resolution, regression testing, approvals, and deployment. 

One way to reduce the overhead is to develop a self-service tool so that marketing strategies are managed and deployed by those who know it best the business. The idea is to have IT develop the logical framework, guardrails, interface and approval processes, which the business can leverage to make changes in production. Because the framework and guardrails are already laid out, strategy updates are “configuration” rather than “code” changes. This eliminates the need for IT involvement, reduces cycle time and improves business agility. The business would still need to embed a quality assurance process to ensure that their strategy is producing the desired outcome. Even so, the business would know the success criteria better than any other team. 

Use Models and Machine Learning 

A rules engine is great for a team starting off with orchestration and laying the basic framework. However, as the orchestration engine takes on additional scope, the sheer volume of rules can become a nightmare to maintain. If your business operates in an environment that requires it to respond rapidly to competitive threats, a rules engine is not the most agile nor scalable. 

To simplify the tool, a model-driven approach should replace business rules where possible. For example, a single model that assigns offer propensities can replace a large set of business rules written to cover the full array of possible scenarios.

 Online real-time vs. offline batch models: Because of the lengthy development period and the large datasets involved to create offline models, customer preferences may shift by the time they are deployed. In contrast, online models use a small but continuous stream of data as a training set and can put more weight on recent trends than on older data. Since online models are integrated with the framework, it could calculate outcomes at run-time, and the response would be fed right back to the training set. This approach ensures that models are always refreshed and can therefore quickly react to market signals. 

Machine Learning: Consider for a moment the characteristics of potential Tesla customers from 2010. A model might correlate these car shoppers with being high-income earners and tech-fanatics. But in 2016, with the introduction of the more affordable Model 3, the assumptions around high-income earners may no longer hold true. Likewise, if a state such as California introduces rebates for electric vehicles, the customer’s state of residence might become a more significant predictor than it was previously. A model framework that continuously challenges the underlying assumptions of its models can be summed up as machine learning.  

Having a framework that is online, self-maintaining, and self-learning can help to remove overhead from human actors. However, a certain degree of comfort has to be established before your organization hands over the reins to an autonomous algorithm. To establish this trust, you can run tests comparing a model-driven approach against your current approach. This can help prove that models are just as adept in driving your marketing objectives. 

Although the numerous advantages tied to machine learning does to a large degree justify all the hype surrounding it, business rules are still important. Businesses often have constraints that would require a hybrid rule-model approach. For example, there may be contractual obligations to present certain offers above others, or to generate a minimum number of clicks for content. Models might also not be appropriate if the question being solved is trivial. We don’t need a model to tell us that the propensity for a customer to accept a $50 retention offer would be higher than a $10 offer.  

Overall, each of these strategies can help to lighten the overhead associated with maintaining an orchestration engine and should be considered to achieve omni-channel personalization at scale in a large organization.