Shadow Data Teams: A pain in the CDO’s ass!
How to Harness Them and Turn a Liability into an Asset
The CFO opened the quarterly business review (QBR) meeting with an overview of key metrics – "Revenue growth is up at 12.3% year-over-year."
Not two seconds later, the Head of Sales points to her dashboard and says – "Actually, we're at 15.7% growth based on our CRM data."
Uh-oh, I thought to myself.
Then Marketing Guy chimes in, "Well, according to our analytics dashboard, we're actually at 10.2%."
The CEO slowly turns to look at me. Not to the CFO. Not to the Sales Head. Not to Marketing Guy. To Me, the CDO. The supposed "single source of truth" guy.
"So," says the CEO with that dangerously calm voice, "which one is it?"
This total mess wasn't happening because we didn't have data talent. It was happening because we had TOO MUCH data ‘talent’, just all doing their own thing. Every department had secretly built their own little shadow data team.
Finance had hired these four ex-banking quants who built a parallel data warehouse that only Finance could access. Sales had a few “reporting specialist" (read: shadow data engineer) who pulled directly from Salesforce and applied his own custom calculations. And Marketing? They had built a large team who built them a "totally revolutionary" analytics platform that literally no one else knew existed, and the head of marketing paid the platform subscription cost from his corporate credit card!
Each team had their own definition of revenue and no one was sharing any data. The Finance reporting specialist was literally downloading exchange rate data from some random website and marketing was copy-pasting inputs from an email report.
No wonder three different executives have three wildly different numbers!
The reality is - any positive outcomes delivered as a result of data it's because "the business made smart decisions”. But when data goes wrong, as CDO, you're the face of the problem – even if you didn't create it. So you need to lean in and fix it!
Yes it’s not fair, but that is life. This is an occupational hazard of being a data professional, especially for the top dog!
Signal from Noise
Your company too has shadow data teams
Your spend on data resources is likely costing more than it should
Yes - the problem is yours and only yours to fix
As CDO, you probably have limited ability to fix this problem, and you will likely create a political s**t storm trying to solve it.
Being Brutally Honest (Ouch!)
If you are a CDO of a large or mid-sized company, your current structure would be the organisational equivalent of using Excel as a database – it works (kind of), nobody loves it, but everyone's too scared to change it.
Your "Federated" Model (not really a model) has grown into quite a mess! The Product, Marketing & Customer Service departments each have built their own ‘Business Intelligence Group”, because they wanted dedicated resources and their data needs were ‘unique’. Each department is running their own siloed ‘data warehouse’ (homegrown database) which hadn’t conformed to any architecture principles, and many of your processes are manual and off-system (Excel, local databases etc.).
Your data architects are all busy on ‘strategic’ projects, and your data engineers seemed very content being pipeline builders and transporting data from source to data warehouse, with no regard for fundamentals like data modelling, conformance and data governance.
Your central data team (stood up only 2 years ago) has created a modern data platform and a single source of truth data model, but has struggled to build insights solutions because you are just a data team, and creating reports is the role of the ‘business’. Plus platform adoption has lagged due to resistance from the business intelligence teams.
As a consequence:
The majority of your reports are manual and untimely, and there are 100’s of Excel reports/models being used to run critical operations and pose a significant risk.
Data in reports is often wrong, the same question often has different answers depending on which team created the report and which system the data was extracted from. Of course these reports are prepared by business intelligence teams, but the CDO gets blamed for this.
Data is being hoarded by a small number of people, which means your end users did not have easy access to data. Also, users do not know who to go to for data and analytics requests as there are multiple teams with similar data.
You don’t have the ability to leverage data as a strategic asset, and your data culture is suffering.
Your labour cost of data is blowing up - it is likely 3X of what you think the cost is.
Playbook - to transform the business of data
Appoint a single accountable person - your Chief Data Officer (CDO)
An organisation with multiple data teams and multiple data leaders will inevitably face challenges due to its fragmented structure and execution, leading to unaligned teams, conflicting direction, and an inability to harness data strategically. An obvious solution to this problem is to appoint a senior leader as the single accountable leader for data analytics - a Chief Data Officer (CDO). The CDO needs to have centralised oversight over all aspects of the data landscape, including capital allocation and prioritisation decisions, and should be tasked with steering the data organisation in the strategic direction.
Because the role of a CDO is ill-defined and not understood in the industry (e.g. compared to a CFO), you need to take extra efforts to clarify roles & responsibilities, especially in relation to the various Directors or Heads of data, performance monitoring, business intelligence, etc., you have in your company. Chances are many of these Directors or Heads consider themselves to be the de facto CDO (at least for their division) and assert that perceived authority. Make it clear the CDO is the executive leader responsible for data. Like each ship can only have one captain, you should only have one CDO at the helm.
Note that your flavour of data operating model (federated vs. centralised) shouldn’t have a bearing on the appointment of a CDO - the CDO can effectively oversee the data organisation in both a federated and a centralised model, as long as they have the right authority delegated to them. This model can be scaled based on the size of the organisation, e.g. divisional CDO model in larger organisations vs. a central CDO model for mid-sized companies.
Empower your CDO with Analytics & AI responsibilities - don’t confine them to just data.
The CDO role is a business (not technology) leader, and the composition of the CDO organisation needs to reflect this. The worst thing you can do is to restrict your CDO team to technical backend functions like data governance, engineering, etc., without any business-facing functions like analytics, AI, data science, and Business Intelligence. You want your CDO to drive business outcomes - this requires the CDO to own analytics & BI responsibilities.
Some companies also make the CDO responsible for delivering business outcomes by using analytics and AI, but leave analytics and AI teams scattered within business units with no intent to coordinate and oversee these teams. Generative AI has compounded the issue for CDOs with many CIOs having staked their claim on AI in their portfolio - but the risk here is over-focussing on trendy and simple-to-implement AI tools (e.g. Microsoft co-pilot) which certainly has some productivity and other benefits, but ignoring deeper and more meaningful business transformation. The other challenge is Technology departments standing up new AI teams, which adds another siloed team in the company.
Speaking from personal experience, this is the worst possible scenario for a CDO to be put in - all the responsibility but little to no authority. This is one of the key reasons for very high CDO turnover - most CDOs quit in 24-30 months, which is too short to drive real transformation, especially in large, legacy businesses.
Ask your CDO to develop a unified data, analytics, and AI strategy.
Given you have multiple data and analytics teams, you probably have multiple "strategies" floating around. Or worse, no one has a strategy at all. Business units have a smattering of ‘data-driven’ & ‘AI first’ in their business plans, but there is no substance behind the buzzwords.
You need a unified data, analytics, & AI strategy to:
Provide clear direction on the role of data, analytics, and AI in your company, and how exactly it should be used to deliver to your business strategy.
Align all data, analytics, and AI initiatives to a single strategic vision (ideally to the company’s strategy).
Establish priorities that everyone agrees on (yes, this will be painful).
A strategy without a funded roadmap is just shelfware and pretty useless - so create & fund a single roadmap that balances quick wins with long-term value.
Define clear success metrics (and no, "being data-driven" isn't specific enough).
Rationalise (but harness) your shadow data & analytics operating model - remove duplication and set them in a single direction.
Your federated data teams are a result of years of departments solving their own problems, either because nobody else would, or they wouldn’t let anyone help. Now you've got to untangle this mess without breaking what actually works.
Find all the bodies (including hidden ones) - You need to know what you're really dealing with, and it’s probably worse than you think. Conduct an inventory of resources across the entire company - don’t just look for people with ‘data’ or ‘analytics’ in their title, look at unofficial data roles like ‘business analysts’ or ‘strategy lead’ who mostly wrangle data or do data science. Assess skills and willingness to adapt to new ways of working. Define your target state organisation structure, and define a transition plan, including mapping resources to roles in your target state.
Design a fit-for purpose operating model that’s suitable for your company. Don’t just copy-paste a generic model from a consultancy. Spend time assessing current state maturity, identify pain points in the operating model, and design a structure that not just solves for these problems, but also allows you to achieve your strategic objectives.
Provide employees with clarity on their roles in the central team, what’s changing and what’s not, and transition timelines. Secure buy-in from employees transitioning from federated teams on the data strategy and the future state.
The best data analysts have a combination of data and specialist business/domain expertise, and you typically find them working in federated teams. Identify these folks and treat them especially well as they will play a crucial role in making your data strategy a success.
Map out shadow reporting processes - you won’t be able to scoop up these reporting analysts as part of your centralisation mission as reporting is usually only one aspect of their roles, but you will be able to unlock productivity savings once you roll out your automation agenda.
Stop or slow down work on legacy platforms and focus your efforts on strategic priorities. If a legacy process is mission-critical, consider hiring an offshore or low-cost consultant to perform this task (you don’t need a senior data analyst who is paid $200k manually refreshing data in an Excel model). Even better, build a bot to automate the process where possible.
Note: Many people have spent years in their current roles. Be aware that you are dismantling (and rebuilding) entire careers and working relationships, and approach this with empathy. The employees you would like to retain may be exploring external opportunities - implement retention plans, budget for retention bonuses. Be prepared to deal empathetically (but swiftly) with people who may not make the cut, either because they lack the required skills or the willingness to be part of your team.
Deploy reusable data and analytics products and promote collaboration between different teams
3 different models to calculate customer lifetime value, 2 potential sources of sales data, 3 customer satisfaction dashboards - you need to put a stop to solving the same problem ten different ways. You need to create standardised products.
Catalog all existing data products, even the unofficial ones (especially the unofficial ones) - this is basically your baseline requirement for your target state data products. Your legacy systems contain a whole heap of crap that isn’t relevant anymore. Identify the metrics/reports that actually matter and serve them as standard pre-canned analytics and insights products. Discard unwanted metrics and reports.
Design standard data products for each business domain (customer, sales, marketing, etc.) and make sure they can talk to each other. Don’t make the mistake of attempting to create a full and comprehensive data model - be agile, prioritise, and build incrementally. Bake in data quality and privacy controls into the data products.
Target high-visibility use cases, e.g. delivering Executive reporting as a quick win is a great way to get some visibility and CEO sponsorship. Also, make sure you enlist power users in your company to be early adopters of these data products, enable self-service, and put in sensible governance processes to govern data usage.
Use data & reporting catalogs to help users find, understand, and use data. Provide training to users on accessing and using this data appropriately.
As you start to migrate reports, actively decommission Excel models & legacy reports currently used. Change manage and transition users to the new platform. Transition employees from the old to new platforms. (I have oversimplified this complex topic; I will look to detail this at a future date).
Prioritise Change Management
I cannot overemphasise the importance of change management in data transformations, especially in organisation redesigns. Change management is the most important factor in the success or failure of your data transformation, not your data platform, not your delivery partner, not the choice of your BI tool.
To be successful in your role as CDO, you will need to transform the way your company fundamentally operates by changing behaviours. You will need to navigate politics at all levels, and will face severe resistance from employees who have spent many years working with existing systems and processes. Without good change management, even the most perfect technical implementation will fail due to low adoption or misuse of your new platform/model. And really great change managers play a dual role and act as the unofficial cheerleader of your team, helping keep morale up and provide creative solutions to resolving problems and conflicts.
A change manager is usually the first key hire I make when I start a new gig. You should do the same. Unlike good data engineers or analysts who are relatively easy to find, good change managers are a rarity, especially ones who have some data transformation experience.
Good change management (and a good change manager) can dramatically improve your chances of success - prioritise this over all else.
Key Takeaways
Shadow data teams are disruptive. Harness and channel their strength towards a single strategic direction.
Your ship needs only one Captain - your CDO.
Deliver visible and high profile use cases early on to get your Executive team on board.
Invest time and top dollar to hire a solid change manager.