The case study presented here - including root cause analysis and solution - was performed for a digital publisher. It offers a different perspective on what data scientists are capable of. The expert involved here is not a coder, certainly not a production guy, yet is able to leverage his business acumen and domain expertise to

• Imagine dozens of scenarios and rank them by chance of occurring
• Get silo-ed data from various departments (finance, sales, marketing, product, IT)
• Analyze the data in connection with the scenarios (including checking data validity)
• Get external data (competitive intelligence) as needed
• Find the causes (not just correlations)
• Find the remedies
• Detect issues well before anyone else can see them, by looking in summary data
• Complete the analysis with a 48 hours turnaround

Such a data scientist who can save billions to a company, is usually not hired, for the following reasons

• Companies are looking for coders, not business solvers, when they hire a data guru, despite claiming the contrary
• A data scientist without Python on his resume is unlikely to ever get hired
• Hard work gets rewarded, smart work does not.

This is why companies erroneously think that data scientists are unicorns, because they won't even interview such a professional when a position becomes vacant. Yet this guy does not consider himself a unicorn. And it's not the story of a smart data scientist, highly capable, misunderstood and poor, unable to find a job. It's the opposite: a data scientist (granted, not on any payroll, not considering himself a worker), who will compete aggressively against whoever (small or big) is in his path, and win time and over, including financially. 

This story also illustrates that data science is not necessarily about big data, big statistical models, or big Python code. It might indeed involves none of them. Anyway, this "unicorn" shares his secret with you in this article. Bookmark it, it might become very handy one day! And you won't need to find that elusive unicorn: just follow his recipe below.

Unicorn Data Scientist Finds Root Causes of Sales Drop, Fixes it

Helping a digital publisher, this data scientist (let's call him the unicorn), among other things, monitors the company finances, using a dashboard  - Freshbooks - that advertises itself as "bookkeeping in the cloud". And it's true that Freshbooks is fantastic, making many high level KPIs easily accessible. One set of metrics is in the "invoicing tab", where you can see all invoices submitted , sorted by date. Gaps between subsequent invoices range between 1 to a7 days. So by looking at the distribution of these gaps (no need to develop a stats model) it is obvious after a 30 second visual inspection, that something happened if the most recent invoice is two weeks old.

This was the starting point for this investigation; the discovery of an issue with invoicing. Note that the discovery and the choice of the data tracking platform (Freshbooks), was the job of the unicorn, in this company. In all other companies, it takes much more time to react, as data scientists look at sales, rarely at invoicing. In short, detection took place one month earlier than in most companies. Interestingly, the alarm was raised at a time when revenue was higher than ever before - but there's a 40 days lag between invoicing and revenue.

The unicorn then imaged scenarios, and the kind of data needed to rule out most of them, and detect the real culprits:

• Is one of the sales reps invoicing less, or is it generalized across sales reps? Is one of the sales reps in vacation?
• Are specific clients responsible for the drop, or is it generalized across clients? Impacting old or new clients? No pattern found. Also there was no client loss or cancellations.
• Some clients moved from monthly to quarterly invoicing? The answer was yes, in this case.
• Were the previous month numbers unusually high? Yes. It also depends when the bookkeeper issues the invoices, and she switched from twice a week to less than once a week - creating a bigger gap than usual.
• Has time-to-close (a deal) increased/ No indication of this, though targeting new clients always increase the metric in question.
• Could a price increase be the cause? Yes for few, little clients.
• Did competition decrease their prices? Offering more competitive products? Gaining market share. Or is the industry experiencing a sudden downturn? No evidence of this. But this happened just before the elections - and this could have played a role, clients waiting until after the elections.
• The products do not work as they used to? No, traffic is at an all time high, and traffic quality high. But some advertisers keep repeating the same announcement over and over. Need to educate advertisers. Data to counter this hypothesis comes from newsletter total clicks by client (vendor data), Google Analytics (after filtering out noise), and client data (leads generated).
• Has the web traffic decreased, or not as relevant as before? More from outside US? Different demographics? No. And anyway, it would have a progressive effect, not a sudden effect.
• Is it a general market trend? No, but in the process, the unicorn discovered new markets to target. He also looked at the kind of ads displayed on competitor websites, as well as traffic growth from competitors, making sure to keep a top position. 
• Is there a reputation issue? No. Check tweets and other postings about the client. And effect would be progressive, not sudden - unless the reputation issue is huge and sudden.
• Was advertising budget reduced, with negative consequences? No.
• Is there inventory left to sold? It turned out that this was the main factor. All ad inventory was sold in September/October for the Christmas season.

Solution

Since lack of inventory was the issue, growing traffic by adding a relevant channel, and offering new advertising products, was the proposed solution. Also, blocking some inventory in advance for potential large clients won't solve the invoicing issue, but can boost total revenue in the long term. Again, the unicorn came with the solutions.

Note that this investigation required collaborating with many teams: sales, inventory management, finance, marketing, competitive intelligence. As always, multiple factors were involved, with two dominating factors (inventory sold out, bookkeeper working with client less frequently because she has acquired more clients).

Note that this is a causal analysis, not just a search for correlations. Easier to perform quickly if your company has fewer than (say) 400 employees.

DSC Resources

• Career: Training | Books | Cheat Sheet | Apprenticeship | Certification | Salary Surveys | Jobs
• Knowledge: Pure Data Science | Competitions and Challenges | Webinars | Our Book
• Buzz: Business News | Announcements | Events | RSS Feeds
• Misc: Top Links | Code Snippets | External Resources | Best Blogs | Subscribe | For Bloggers

Additional Reading

• 13 New Trends in Big Data and Data Science
• 22 tips for better data science
• Data Science Compared to 16 Analytic Disciplines
• How to detect spurious correlations, and how to find the real ones
• 17 short tutorials all data scientists should read (and practice)
• 10 types of data scientists
• 66 job interview questions for data scientists

Follow us on Twitter: @DataScienceCtrl | @AnalyticBridge