Big data and data science are on everyoneʼs lips. The business models of Google, Amazon, Facebook and the like are based not least on their ability to collect data en masse and to use them. With deadpan statistics of the past, however, this has nothing to do anymore. Data management and analytics are sophisticated, vibrant disciplines – a must-have for large enterprises.

In addition, data become increasingly important for startups – though usually in a complex way, given the know-how requirements in general and the startup’s resources in particular. This article by INWT Statistics points out what startups should consider when collecting and evaluating data, ideally in a profitable way.

Central data questions

Most startup teams are skilled in business (especially marketing) and IT. But only few teams have an on-board statistician – or data scientist, as in the modern and more comprehensive job description meaning of the word. Yet someone in the startup has to take care of central data questions like these ones:

• Do we need a CRM and or a BI system (Customer Relation Management / Business Intelligence) and how should it be specified?
• What software is best suited for us?
• What metrics of what systems should we collect?
• How do we structure our data?
• What do we need to calculate key figures and how do we turn them into reports?
• Are there groups of customers whom we should target with separate marketing (customer segmentation)?
• What value do our customers have (CLV, Customer Lifetime Value)?
• How much do individual online marketing channels contribute to our success (customer journey analysis)?
• How should we balance the budgets for different channels optimally (attribution)?
• What contribution does TV generate to online sales (TV attribution)?
• How do we implement the most reasonable highest bids in display advertising (RTB) and SEA?
• How can we tap on the up- or cross-selling potential of our customers by making appropriate recommendations (recommender systems)?
• How can we optimize our website for A/B testing?
• How do we recognize and prevent fraud in credit card payments, or click fraud in display advertising (fraud detection)?

Creative solutions vs. comprehensive data science

Founders and creators are makers. And many of them are business and IT people. While they recognize statistical issues as being essential to the companyʼs success, they often draw on their knowledge of statistics from their university studies, which conveys at least knowledge of descriptive analysis of data, of the statistical significance tests and simple regression models. IT graduates complement this spectrum mostly with the knowledge of common machine-learning algorithms. With this knowledge in mind one can find creative solutions for many of the above issues. But oftentimes the following is overlooked.

Statistics, although perceived by many as an auxiliary science, is a full-fledged discipline as diverse as computer science or business administration, both in terms of breadth and depth of techniques and applications. In consequence, for performing a sufficient CLV analysis it is not sufficient to have merely statistics knowledge from two or three college lectures. This is also why an IT system electronics technician is not given the task to adapt SAP modules.

Theory plus practical experience

Graduates with a degree in statistics are equipped with theoretical knowledge in areas such as robust and nonparametric methods, survival time analysis, design of experiments, Bayesian models, extreme value distributions, etc. Those who only minored in statistics are little to nothing familiar with these. In addition, graduades from Dortmund, Munich or Berlin for example (where academic statistics is offered in Germany) gather a great deal of experience in practice.

In professional everyday life, data scientists gather hands-on knowledge on particularly important metrics, popular pitfalls concerning the data quality and best practices in different fields of application. These skills allow experts to solve problems top-down much faster and with more meaningful results than part-time statisticians could do.

Advice is worth it

Optimized forecasting models (e.g. for staff scheduling, for bid management in real-time bidding or for detection of fraud) amortize their development costs multiple times within one year in direct comparison with homegrown solutions. The latter are cheaper in the making, but more expensive in the long run in terms of not saving all the costs that could be saved with professional forecasting models.

Competence in handling data and algorithms is a competitive advantage, just like efficient IT or an innovative marketing strategy. Startups without the on-board senior statistician, but with team members who are familiar with statistical sub-disciplines, they can recoupe consultancy costs fairly quickly in most cases.

Professional use of data

An experienced external data scientist structures the process of finding solutions, thus saving time for research in too many directions. Moreover, experience in similar projects accelerates the solution-finding process thanks to precise selection of the right metrics and by avoiding typical errors, and thereby provides high-quality implementation. Both factors save money and resources. The ROI of professional quantitative advice is above average. This holds even more true when the examined problem is critical for market success and when the startup has staff who in future will maintain and wait the algorithm, after being schooled by external consultants.

Finding best solutions: external and internal

In the world of quantitative advice it is common to offer a free, no obligation initial consultation. On this occasion, interested startups can feel the tooth of potential consultants and get an idea of ​​what are the advantages of a consultation compared to an in-house solution. Own resources permitting, it is worthwhile to develop a strategy for the orderly transfer of hands-on knowledge from consultants to internal personnel. This allows to combine the wish for best practice with the one for an in-house solution quite well.

[July 2015]