Should your analytical team be centralized or decentralized?
When this question is raised during an organizational design, you will find enthusiastic supporters on either side.
Proponent of the centralized analytical function will quote the benefit of streamlined data, consistent metrics, models, algorithms, insights, knowledge base…
Their opponents voice their concerns of bottlenecks, lack of innovation, low efficiency, difficulty to scale...
If you are an analyst or data scientist, you probably have worked in various organization structures; even survived a few reorgs that swung between these two.
Why does this matter?
It matters because analytical capacity is a precious resource to be allocated across the company. How to make sure the allocation aligns with the areas that will generate the biggest lift to the business; and address the most critical risks?
How are human resources usually allocated? The largest scale of such example is a country’s economy. An extreme example of a centralized system is Chinese planned economy. In the twenty years prior to 1978, Chinese agricultural and industrial yearly output was determined by a small group of “experts” for the whole nation; and assigned to each factory, village, household. The goods created was then allocated to individuals. Families were given ration for everything from rice, cabbages, bicycles, TV sets, to apartments. In this economic structure, China suffered three years of famine during which over 20 million people perished; and ten years of cultural revolution, when another million people vanished. You can’t blame the economy for all the social turmoils. But it made you wonder how things would be different with a more open economical form.
If China planned economy is a bad example of a centralized system, what is a great example of a decentralized system?
How about blockchain?
By having a “shared public ledger” (blockchain), and “a distributed consensus system that is used to confirm waiting transactions” (mining), Bitcoin is clearly designed to be an open source, decentralized ecosystem.
Yet, a hacker and blockchain researcher at Harvard (Primavera De Filippi) and a professor (Benjamin Loveluck, Associate professor at Télécom ParisTech) argued otherwise. While everyone can mine and submit changes to the software, only a small number of core developers made decisions on what changes shall be incorporated to the main branch of the software. They called such a system of governance “autocratic-mechanistic”. It has no formal governance structure, only implicit, with the project often relying on a “benevolent dictator.”
Can you imagine an analytical function structured as the blockchain system?
In such an organization, everyone has access to data, can mine it, run analyses, develop machine learning models, generate insights, and contribute to the knowledge base. Analytic is blended into everyone’s job - Product Managers, Engineers, Marketing Operators, Accountants, Sales Executives, Customer Service Reps. They are provided training and tools to ensure they are efficient in doing the analytical portion of their work.
In the center of this disturbed analytical network, sit a few “benevolent dictators”.
First is Data Engineering. They build a coherent, scalable and high-performance data infrastructure for the whole company. They also design data schema to be followed by everyone who creates data assets. They may not develop all the data, but they have a complete map of where each data asset is located.
Second is Finance. They define a set of standard metrics with clear business logic; and continue evolving the metrics as the business grows. These metrics are well communicated and documented. They are the common language for analyses and measurements for machine learning models. In a public company, they are also the medium to explain the business to the external investors.
Third is a “CIA” function. In this decentralized organization, knowledge spreads out across multiple functions, business entities or even geo locations. The distributed knowledge calls for a function dedicated to collecting the information, consolidating the intelligence, and drawing insight from the connected data. At the same time, the “CIA” team serves as the diplomat between functions, coordinate the development of new data, selection and deployment of new tools, and alignment of different interpretations of data. They are the virtual headquarter where everyone can check in to get the most updated information on data and analytics. They can also help the executive team to digest the vast information generated across the company.
Another amazing example of a decentralized system with a centralized goal is the swarm of 1,000 simple robots built by Harvard researchers. Likewise, the most efficient and innovate company makes tons of decisions “locally”, but coherently toward the overall business vision. Such design is especially beneficial during the early stage of a company, when fund is needed in many business critical areas. A dedicated analytical function has to wait after engineering, product, sales, marketing... It is certainly beneficial to larger companies. Encouraging, or even “forcing” everyone to be analytical provides a mental framework for them to evaluate the impact of their work.
[Reference:
The invisible politics of Bitcoin: governance crisis of a decentralized infrastructure