It’s Too Late for Data Governance

When do you know you need to lock your car?
When do you know you need a home alarm system?
When do you know you need a strong password?

After you have been stranded, burgled or impersonated; when it is too late. The same goes for data governance.

When decision makers discount your opinion because they feel your data–the raw material of your insight–is not trustworthy…

When decisions are made against collecting additional data, as there is no confidence in its value…

When baseline operational information is ignored due to suspicion…

It’s already too late.

Start Getting Your House in Order Now

Data does not show up on the balance sheet but, like your staff, you’d be sunk without it. Data governance is the processes and procedures put in place to ensure that this asset is captured, cleaned and stored in a way that a) maintains its validity, and, b) is auditable.

You are being paid to come up with meaningful, valuable, data-informed insights and if your data is in doubt, you can never fulfill your role. If you’re still talking with senior decision makers about the veracity and trustworthiness of your data, you have already lost.


It’s time to take action. It’s time to step up and protect your data against the doubts of insight consumers and thereby protect your most valuable asset: your job.

Start With a Senior Champion

An executive sponsor is nice, but this is going to take a lot of persuasion. Everyone will see this as a cost and interpret it as an imposition, or even an affront. This is a change management process.

You will need someone with sufficient authority and drive. Not only are they in a position to make things happen, they have the disposition to make things happen. They need to be all fired up, because this effort is going to shake a lot of trees.

Oh, and just in case you thought that was the hard part? You’re going to need to make good friends with your legal department, as well. You want their input sooner rather than later. It’s their job to understand the requirements laid out in Sarbanes-Oxley and HIPAA and any other data regulations might be in place (or coming). (Privacy, anybody?)

Work With a Cross Functional Team

This task will take the knowledge, talent and persuasive abilities of technical professionals, business authorities and security specialists. Make sure this triumvirate is equally represented so that issues can be more easily discovered and resolved, and acceptance can be successfully fostered.

Spend Some Time on Taxonomy

Get buy-in across the organization on definitions so the element that is quantified here is actually the same thing as the one quantified there. That way, individuals who create local models can be assured they are comparing apples to apples and benchmarks are meaningful instead of confusing.

Attention to detail and cross-department accord are necessary. Be prepared to spend a fair amount of time here.

Set Proper Expectations

The digital data we collect is not precise. We do not know exactly… well, anything. Get over it. Help those who depend on data-informed insights understand that we’re dealing in probabilities rather than certainties.

This is tricky because you’re asking them to buy into a standardized metrics system that you introduce as flawed. The numbers won’t match. All of the information that seems to describe the same things will never equate. That’s OK, as long as everybody has the same expectations.

The CFO’s “system of record” is bound by legal requirements. But all the spreadsheets used operationally need not be exact; they only need to be useful and directionally in agreement with the gold standard.

Ascribe a level of confidence to different types of data that everyone can agree with:

  • Financial/Transactional- 100 percent confidence
  • Customer Satisfaction- 90 percent
  • Online Behavior- 75 percent
  • Advertising Response Rates- 70 percent
  • Social Sentiment- 65 percent

Control Access to Data

With such a valuable asset, precautions are needed on three fronts: technical, informational and consumable.

First, only those vetted to the hilt should be allowed to modify a database structure. This is a technically challenging task that requires a complete understanding of the interactions between databases throughout the company. One must understand all of the potential consequences of such an action and this requires special training.

Second, while the data schema is a technical task, the ownership of the data itself must reside within a specific business unit. These are the people who can recognize when something looks off kilter. These folks know what the norms look like and can provide some level of quality assurance. They are also the ones most likely to depend on that data directly and so have the strongest desire to maintain its accuracy. Assign a data ambassador for each data source and make them the point person to monitor for out-of-bounds thresholds. Caution: data made public or shown to your C-suite should be owned by Finance, with all its historical, administrative rigor.

Finally, data democratization is wonderful in concept, but difficult in reality. If all of your data is trusted, it may seem like a good idea to encourage all of your personnel to make use of it. Yet data comes with lots of unwritten caveats and your entire staff cannot know all caveats of all datasets. It’s quite easy to derive insights without understanding the detailed limitations on specific numbers.

Data democratization is a good idea, but citizenship should include meeting some important requirements.

Start With the Most Trusted

Once you can be assured changes are controlled, start the clean-up process, but do not start with the most egregiously untrustworthy system. It may seem that fixing that one will win organizational confidence, but the legendarily erroneous systems will take too long to validate. They were made for Don Quixote, not for change management.

Instead, tackle the most relied-upon data sources. When those sources are validated and endorsed, publically identify them as trustworthy. You’ve now established a cleansing protocol and the authority to identify which other datasets are in need of attention.

The Ultimate Goal: Enrichment

Having lots of data is not nearly as valuable as having lots of different types of data. Two terabytes of online behavioral data may be better than one terabyte, but being able to correlate one terabyte of online behavioral data with half a terabyte of sales data and a half a terabyte of social activity is far better.

The more data sources your analysts can trust, the more insights they can derive and that will always be worth the effort. Just don’t wait until it’s too late.

Related reading

Flat business devices communication with cloud services isolated on the light blue background.