There are so many users today, as so many customers. The customers are the backbone of this industry. Failing to please the mass means failing business. It is essential to make the customers feel important. We do that by connecting to them, by letting them have the best experience while they are using our sites and our services.
We join them to experience. How do we do that? We have what we call decision rules, picking rules, and predicting model. These three can be the frameworks that you can use to be able to bridge the gap between customers and experience.
This is trivial. It means if this [this] then [that]. It’s a basic construct. It is a logic that links observation to action. It is focusing on the actions as opposed to the data. Because really, fantasizing on data is not useful because data in itself has no value until it is being deployed.
There are two types of data, the observational data, and the CRO.
The observational data are the ones you passively collect, these are the easy stuff. We don’t actually do something about the data that we have been collecting, we just collect them and then store them somehow. The other type of data, the CRO which is experimenting and collecting different kinds of evidence.
These are causal data, when we manipulate the system then we are in a way learning and seeing different kinds of experiences. It is important that you understand this distinction. This causal effect cannot be estimated unless you do something to the system or the data collected.
Choosing rules is the optimization, and optimization is picking rules quickly. How do we do that? There are infinite numbers of rules in optimization. Any data sciences are almost the same as any amount of data. It would mean that there’s definitely a space for domain expert, there’s definitely a space for someone who knows something.
You might be worried about over optimization and losing your taste and quality. That isn’t a problem. Surely, there is no workable system that you can over optimize without some prior information.
We have AB Testing. This is when we collect data and compare one from the other to figure one which group is actually higher in the grid or information pool. Some other would often go looking for p values of whatever but in recent studies, the p values are of not much importance and are somehow invalid in evaluating marginal impact.
In picking data, you may need to offer a variety of options for the customers. Remember that the customers will not be answering and providing information that is the same. This is where segmentation takes place. Separated segments or groups are two mutually exclusive rivals. It’s always the answer to everything.
We use a model to combine the segments. The predictive models will be figuring out what works for whom. What it does is to mix and blend all features together of an individual system to give an estimate preference and intensity of each of the features present.
As a sort or review, to be able to know which one would work better among the three, you’d need to create a measurable goal. You must also sense the environment, apply the act and execute. There have to be experiments so that they are outcomes to be observed.
Lastly, it is important to learn the logic.