AWS Cloud Enterprise Strategy Blog
Organising for Data
Best practices usually aren’t.
—Peter Thiel, Entrepreneur
I generally dislike the phrase best practice. It implies that a single truth has been discovered for something which cannot be improved upon. It engenders a complacency that, once the practice is implemented, thinking and innovation can stop. This goes for organisational models too. Business cases that propose a “world-class best practice <fill in name> team” along with an associated small budget raise a wry smile. It’s easy to slip into trying to copy someone else’s organisational model for a particular problem or domain, and yet this ignores the reality that just like culture, organisations need to fit your company’s context.
A common example of this is how companies organise for data. The intent is simple: how can data be turned into actionable insights and action quickly and cost-effectively? What I’d like to share here are three models to consider for your own organisation, expanding on my Centre of Excellence blog. These models apply not just to the technology, but also other aspects, such as where data scientists would sit. The models are grounded in how decisions are actually made in your organisation, and with a bias towards ensuing action based on data insights is prioritised over consistency and standardisation.
The models are all variants of the hub-and-spoke model. The hub refers to a centralised team and set of capabilities, and the spokes are capabilities within the line of business (LOB) or functional teams. In the case of a multinational company, the hub is likely to be the corporate headquarters and the spokes the individual business units or countries. The hubs typically strive for economies of scale and skill by standardising data technology and even utilisation. The spokes decentralise capabilities to be closer to the customers and to where the data can be turned into meaningful action. For organisations where decision-making is mainly driven from the LOBs, or where the LOBs have diverse needs and businesses, more activity would be driven into the spokes. In other words, the model should reflect the reality of decision-making power and commonality (or lack of) within an organisation.
Rather than make sweeping generalisations though, each aspect of the end-to-end data process should be mapped into this model. Aspects such as insight generation, governance, data wrangling, data stewardship, and value realisation each need to be considered individually versus trying to decide whether “data” needs to be centralised or decentralised.
Centralised Model
In this model, the hub contains the bulk of the data and analytic resources along with the centrally managed data technology. Priorities and strategies for the use of data are driven centrally. This model is most appropriate when there is a large degree of commonality across LOBs or functions with shared services that they use, and little differentiation across the LOBs. Capabilities in the LOBs are normally centred on basic reporting. Companies starting their data journey often opt to use this model to optimise the use of specialised, scarce resources such as data scientists until such a point that LOBs have sufficient data literacy to drive their own deep analytics.
Distributed Model
At the other end of the continuum is the distributed model, where the balance of power lies within the LOBs. The LOBs’ organisations will typically have their own analytic capabilities to support their high degree of autonomy and implementation of action close to the origin. Ideally, even in these decentralised organisations, companies should strive to establish and maintain a common data technology platform, freeing up LOB investments and resources to focus on differentiated work. The hub plays a limited role with its efficacy often in facilitating knowledge sharing across the hubs and providing self-service capabilities.
True Hub-and-Spoke Models
The majority of organisations I meet fall somewhere in the middle of the two extremes I’ve described. They combine the economies of skill and scale that centralisation allows, but the agility that decentralised capabilities brings. Over-centralisation leads to bottlenecks and dissatisfaction as the central team fails to keep up with demand. Over-decentralisation leads to duplicative expenditure and inefficient use of the available data and learnings, also resulting in slow results. This is also a balance that changes over time and needs to be assessed regularly.
This is where we get into shades of grey. The hub typically becomes a centre of excellence (COE), enabling the LOB spokes. The models can range from a larger decision-making COE with smaller LOB teams to a smaller technology COE and larger LOB teams. Such models can pragmatically address organisational constraints and complexity, such as limited analytics headcount and LOB differences.
The key here is to find the right balance of capabilities in the hub and the spokes, and to adjust this as the environment and data maturity changes. Initially like in the centralised model, limited specialist resources might sit in the hub. The spokes might have limited reporting and analytics capabilities, with more sophisticated requests being managed through a demand management process in the hub. As capabilities and understading grow, some specialist resources will be moved to the spokes and closer to the action, either permanently or for a period of time. Either way this helps ensure empathy and knowledge sharing between the hubs and spokes. The goal should remain the same throughout: to increase the number of data driven insights and their conversion into action.
What Goes Wrong?
When I reflect on these models, they seem blindingly obvious, and yet it is not unusual to see them poorly implemented. The three common anti-patterns I observe here are organisational blinders, a technology-over-results orientation, and a fixed mindset.
Organisational blinders refers to where companies ignore the reality of decision-making culture, believing (or wishing) that decision-making is centralised or that all LOBs have unique needs when the converse is true is a classic recipe for failure, as I discussed in my balance of power in multinational companies blog post. Bringing a virtual team together with representatives from the business units and central organisation to manage the correct balance in the organisational model is a powerful concept and something I will further expand on in a future blog post.
The technology-over-results orientation sees a myopic focus on the technology which misses the point of data: action. One-size-fits-all tools with an insufficient focus on training and change management inevitably lead to a lack of action, dissatisfaction, and a myriad of competing, fragmented initiatives. Technology plays a critical role, but it’s not the end game.
Finally, organisations change over time due to changes in leadership, markets, customers, and external pressures. You can’t view your company as a fixed entity. The organisational model which works today might not be the appropriate one tomorrow. Having a grounded view of what is working and what needs to be tweaked based on these changes and feedback from stakeholders can address this. For instance, the initial foray into data that requires discovery of what types of resource profiles are required might be best done centrally, but capabilities decentralised for speed as data literacy becomes more prevalent.
Regardless of approach, having a single-threaded leader accountable organisationally for making sure value is derived from data and that the use of data is championed regardless of technology or structure is a powerful step. This leader does not need to own the data; rather, he or she should focus on driving a culture of using data and turning it into action. After all, the majority of “data issues” in organisations are actually cultural and organisational problems.
—Phil
Twitter | LinkedIn | Blogs | Email
Big Data and AI Executive Survey 2019, NewVantage Partners
How to Build Data Capabilities, Ishit Vachhrajani
In Search of Silver Bullets: Moving Beyond the Dream of Data, Phil Le-Brun