I've been in the enterprise resource planning (ERP) and customer relationship management (CRM) business for over 15 years and one thing has always remained the same, that system-based decisions rely on quality data. The challenge in the past was that the complex and fixed code written was only good for a point in time.
We didn’t have machine learning (ML) capabilities like we do today in the old ERP/CRM worlds. We relied on outside algorithms as ‘black boxes’ to handle forecasting and other complex processes. But in reality, we relied more on providing a lot of information to people and relying on their tribal knowledge on how to deal with that information.
The challenge in today’s market is that customer preference changes easily - forecasting still is not accurate and identifying anomalies in data is still a very manual process. This is where data driven organisations are now moving to the front of the pack in their respective industries. Let’s look at a few examples of how companies are transforming with data-driven decision making.
For example, a leading travel company is leveraging an integrated CRM, marketing automation and machine learning ecosystem to:
Profile customers more effectively based on historical similarities, augment their captured data with historical data and add in outside data such as credit scores to determine the propensity and capability to book various travel packages and stack rank the leads into highest probability to close with the highest value.
Once the customer is profiled and top of the funnel, then direct to the appropriate call centre agent who specialises in that type of travel booking and customer classification. This all happens within seconds and has netted the customer an expected 32% increase in revenue and a higher net promotor score with customers due to the focused knowledge set of the call agent versus generic knowledge.
So how do you prioritise and determine the data that will impact your business?
The first step is to determine the personas in your business you want to target - call agent. Then determine the questions you want to answer such as who should I call back? How can I ensure I'm calling the best quality lead with the highest probability to close with the largest value of deal?
Next define the data model required to support these answers - identify both the data you own and the external data sources that you can leverage to augment the data to reach the best qualified decision. Then determine the outputs or actions from this data - who gets what lead? What do we do with the low end, low quality leads? How can we validate this process?
In today’s cloud-based world - this type of proof of concept (PoC) can be realised in a matter of weeks versus months with large expenditure. I recommend to any of our clients to start small and build on success.
Some examples of recent discussions and implementations:
Forecasting algorithm for a high volume, high mix distributor with a demand that revolves heavily around key events that they hold. Integrating marketing data, product order history and promotions into the algorithm provided a highly accurate forecast.
Dynamics pricing for steel manufacturing - allowing dynamics pricing based on prospects location to their site versus competitors, current stock levels and quantities ordered.
Churn analysis for SaaS software provider - augmented with telemetry data from their software which had number of logins, number of reports run, hours on the service etc. When a drop off occurred - flag them to the account rep and get them to retrain or communicate new features and capabilities before they cancel their subscription.
Customer profiling for retailers and manufacturers to build up fan bases for top customers and reward them across social media, email marketing, in store promotions and recognition.
The biggest mistake I see is that companies think their data is all that they need. Augmenting your data from external or multiple data sources is the key to a well-rounded profile to make your decision on. Data.com, Experian, Discovery.org, Weather or other services provide valuable information that you can augment your internal data with to provide a better decision.
The second mistake I see is trying to bite off too much up front and wanting the perfect algorithm. ML/AI is a journey not a destination, the more information you collect and analyse - the more tweaks you can make to get to that quality data-driven algorithm. Start small, build on that success and you will be far more successful.
Tim Harris is the vice president of strategy and solutions at Arbela Technologies