As CEOs and Top Dogs, we get excited about the potential upside of projects designed to deliver better data to the organization.

Unfortunately, problems in data quality projects often reveal themselves late — when much of the costly IT work is already done. This is in part because the new processes needed to improve the data can cause their own critical problems. Here are six ‘challenge questions’ to keep in mind when considering data-quality-driven project. Getting good answers to these will go a long way to preventing data-quality-driven project failures at your company.

1. Why does better data really require buying (or building) new software?

Separate the issue of improved data from the big software project. The knee-jerk reaction is to say, “We need a new system.” Instead, look at making incremental changes to the current system. Possibly build a prototype that proves out some of the anticipated benefits. In other words, make smaller, less risky investments first to prove out the investment thesis.

If improvement in data quality is quoted as a side benefit of another project, make sure that project plan includes complete costs for realizing those benefits. Improvement in data quality seldom is a ‘freebie’ side benefit.

2. Determine what the new process will break.

This question is the one that seldom gets seriously considered in IT proposals. It can be a killer.

Realize that your ‘old’ process is in one way or another a working ecosystem. People communicate, there is a rhythm and dynamic that makes the thing go — even with its imperfections. You see the flaws, but seldom acknowledge how well the current system really delivers.

Therefore, look at how the new-and-improved system might break that pragmatic working. Determining these breaks too late in the game will tempt you to hard-wire changes into the system so you retain the best of the old ‘pragmatic’ workings. This makes IT work more expensive and can it more difficult to realize the anticipated benefits. In the worst case scenario, your department leaders might undermine the project as they better understand the implications of the new process.

3. Who do I hold accountable for data quality?

Who is accountable for data quality? Staff the role realistically. You need a supervision process to ensure the data quality work gets done correctly. Typical mistakes include assigning this to an IT function, or staffing it as a part-time ‘add-on’ job. Result: the business doesn’t accept the results, or the work doesn’t get done in a timely fashion. Thus you don’t realize the value of the investment. Take this seriously! Data cleansing usually requires effort. Withot an owner, it probably doesn’t get done.

4. Does the plan account for the add-on processes related to data quality?

Ensure that your team is looking realistically at what it takes to process the data end-to-end to an enhanced level of quality. Such processes include:

1. Who handles data that is ‘kicked-out’ of the system (e.g. bad addresses)? How fast does it happen?
2. What is the measurement framework that analyzes the data quality process? Have the metrics been thought through? Example: old process: everything counts as a lead — marketing happy, sales not happy. New process: Only certain ‘clean’ items qualify as a lead? marketing loses lead count, sales close percentage goes up. These can devastate existing core marketing and sales metrics, and undermine support unless you plan to address the issue.
3. What is the process for managing disagreement about how data is categorized?

5. Is there a clear process for maintaining data quality?

Cleansed data must be maintained. The investment in data cleansing and enrichment is an on-going investment so long as the data is to be used. The more abstract your objectives, the more carefully data must be reviewed and confirmed — along with change management if changes to the data affect interests within the company. If you fail to budget or organize around maintenance, you find yourself back with bad data in no time.

6. Is the project focused?

Do the minimum as fast as possible. In most cases, there should be a business-relevant deliverable in no more than three months. After the first delivery, additional phases should be expected every six weeks or so. Your team will learn a lot this way early in the investment cycle and this will ensure better outcomes in the long run. If data quality is the key deliverable, focus on reporting and analytics to get the value. Making use of the data as fast as possible will demonstrate other possibilities for using it that can influence the direction of the project. Consider deferring ‘extra’ application functionality so that extra work doesn’t sidetrack getting to the real meat.

Popularity: 83% [?]