This blog post was inspired by a conversation that I had recently with some people who were interested in hiring a data scientist. This is a company of very switched on people, but they’ve never hired a data scientist before. So there needed to be a lot of communication. A lot. Of communication.
Here’s the a list of the ‘soft skill’ lessons that I learned from these conversations:
- Always trust that you will come to a mutual understanding, it will just take time
- Assume everyone is equally intelligent, but also assume everyone has been given a different definition of every technical term that is mentioned
- Assume everyone is using the wrong word for everything all the time
- Don’t waste time being frustrated; take every agreement as a victory and keep moving forward
- People absorb ideas and explanations better in writing - listening and thinking at the same time is hard
- You and the client are probably using the same words to describe different things and confusing each other; ask questions to clarify
- If people aren’t understanding you, you need to find a better way to explain yourself
- Everything will always be much clearer tomorrow
But on top of the regular… ‘challenges’ of communication, a much more specific misunderstanding happened that I thought was worth writing about. This whole story started with a single, innocent request:
“We want someone to do some marketing analytics for us.” - Real Person, 2018
After a massive length of time we separated out four distinct ideas, which at one point or another someone in the conversation was using as their definition of ‘marketing analytics’.
1. Analysing Customer Behaviour
GOALS: Optimise marketing choices to increase sales, profit, or revenue (or any metric that your company deems important).
In this scenario, an analyst would being considering data about a customers behaviour as they encounter the various touch points of the marketing pipeline. Here’s a general example of data that could be collected to use for these types of analysis:
- Webpage clickthroughs
- Email or phone correspondence
- Actions before and after customers making a purchase
- Previous purchasing history (for existing customers)
- Actions before and after a customer stops using your product
With this data you can then do all manner of analyses, employing any appropriate statistical methods. For example calculation of retention rates, identifying features that correlate with purchases, or predicting future behaviours to motivate future marketing strategy.
Knowledge of a customer’s behaviour is imperitive in marketing to them effectively.
This was the definition I had in my head when we began discussions. Probably due to the fact that I have a whole series of other blog posts on a specific class of models.
2. Analysing Effects of Marketing Efforts: Experiments and Studies
GOALS: Provide evidence to support hypotheses designed to discover if specific marketing activities have had their desired effect. Or conclude that you have no evidence to support this, and try to redesign a marketing campaign that will have an effect.
Experimental design deserves it’s own series of blog posts, and there are approximately 1 billion textbooks written about related topics. So I am going to skate over a lot of details, and just try to highlight the differences between this aspect of analysis and the other points mentioned here.
Point 1 above covers the behaviour of customers before you roll out a marketing campaign to develop your relationship with them. You want to know how they’re interacting with your business and what they’re likely to do next.
Point 2 here is about analysing what happens after you roll out a marketing campaign.
If your companies revenue has increased as a direct result of a marketing campaign, then you will want to be able to support this with data. For example a t-test may be appropriate: this requires you to design a controlled experiment in which one set of customers is exposed to a marketing campaign and one is not. A t-test then allows you to test whether there is a difference between the mean purchasing behaviour of these groups, and whether that difference is statistically interesting.
Revenue may increase after a marketing campaign, but it is important to use statistics to decide what this wonderful outcome can be attributed to.
So this differs from Point 1 above in that it helps a company to understand the effectiveness of the marketing campaigns it chooses to implement. This would be analysing marketing efforts.
3. Developing an Analytics-Driven Product
GOALS: Develop a product that uses a customers data (which you store with permission) to provide additional services to them. And then market that new product to them and to others.
This one is extremely dependent on the nature of your business and the products that it offers. The example that everyone’s heard before would be Netflix’s recommender engine. Initially they just offered a streaming service that may have suggested other movies, but they would all be terrible. So they went and used all of their data science power and built a monster of a recommendation system, based on data that they had collected about their customers’ use of their existing product.
This then serves two purposes. The first is that the performance of this product differentiates them from others in the market. The second is that they can then brag about this endlessly. And more importantly, they can brag about it in ads. Ads of all shapes and sizes.
Developing a product driven by analytics gives you a fantastic asset to market to people.
So this is marketing an analytics product. Which is a different concept from marketing analytics (Point 1 above).
Both are fantastic opportunities to employ a data science team to increase the value of your company. But in the discussions I had, this was a source of confusion that none of us realised for a while.
4. Analysing Effects of Products
GOALS: Attempt to infer drivers of a clients behaviour based on the adoption of existing and new products.
This is basically experimental design for products (as opposed to point 2. above which is experimental design for marketing campaigns). There are a couple of use cases for this situation.
In one case you may like to know how much of an impact your product has had on a client. Suppose that your company offers softare to optimise the amount spent on marketing for a client. You have a client that is about to go bankrupt; they buy and use your product for a few months, and because you’re such great developers they claw their way back to making a profit.
Inferring causality in this situation would be innapropriate, but with enough data you would be able to quantify correlations. And with enough data about the market in which a client operates, there would be opportunity to conduct hypothesis tests to form the basis of a claim that your product does make a significant impact to a client’s business.
What can you do with this information? Tell everyone. Ads. Ads ads ads ads ads.
Clients will always be skeptical. Back up your claims with data.
So this would be marketing the effects of products. Which is different to marketing a specific analytics product. And both are different from marketing analytics.
So in summary: Data scientists are more than capable of doing all of these things and they are all a worthwile investment for a business. Also communication is difficult and confusing, but it is important to remain optimistic during the process.