How do you identify and prioritize data quality issues and improvement opportunities?

Data quality assessment has to be directly tied to improving business outcomes. So step #1 is to define which business outcomes are the most important to achieving your business strategy. Step #2 is to understand which business processes produce these business outcomes. For example, if your business strategy is focused on improving customer retention, what are the key business processes that improve it? Online ordering process? Customer service? Or maybe order fulfillment? To get the most value out of data quality discipline in your company, it's imperative to start with these business strategy and key business process alignment steps lest it becomes a purely technical exercise with no appreciable value to the business.

To identify and prioritize data quality issues, start with profiling data sources. If using an AWS Ecosystem; use Glue for metadata analysis. Assess data quality dimensions like accuracy, completeness, consistency, and timeliness. Leverage Athena for SQL querying to detect anomalies and discrepancies. DataBrew can help in cleaning and transforming data. For prioritization, evaluate the impact on critical business processes and regulatory compliance. Use AWS Lake Formation for centralized data management. AWS Data Quality Monitoring with CloudWatch provides real-time insights. Iterate improvements using AWS Step Functions for orchestration. AWS Glue ETL ensures continuous enhancement. Monitor progress with AWS Quicksight.

Prioritising data quality starts with looking at the intended outcomes of your corporate or organisational strategy and then the data outcomes needed to support that strategy. Getting lost in sources and data governance in some organisations will waste a lot of time and much time disappearing down cul-de-sacs. Once you have clear outcomes the prioritisation becomes much easier and then you can look at quality and governance.

We tend to start by assessing as many data sources we can get our hands and as quickly as we can which can overwhelming. Instead, start cataloging your data sources first and how the data flows between them to spot any issues or weak points. The goal is to trace your issues back to the source, not get lost in the pipes.

Assessing data sources involves a thorough examination of the data’s journey from creation to consumption. Utilize data catalogs, lineage, and dictionaries to gain insights into the source, structure, and metadata of your data. This step is crucial for identifying potential quality issues at their origin. Ask yourself: Are the sources reliable? Is the data collection method consistent? How well does your data governance policy uphold data quality and compliance standards? Implementing robust data governance ensures not only the quality but also the security and privacy of the data. Regularly reviewing and updating these policies in line with evolving data regulations and organizational needs is essential.

Based on my experience, here are three most important properties of a data quality rule: 1. What's the impact on the business outcome when this rule breaks? 2. What's the threshold at which this impact occurs? 3. Is this impact high enough to justify spending money on fixing the problem? Data quality dimension, while making for a fun discussion among data professionals, doesn't help define impact or help make the resulting issues more relevant and actionable - i.e. doesn't help with the goal of data quality program, which is to improve business outcomes.

To identify and prioritize data quality issues, we use a scoring system based on criteria such as timeliness, proximity to public registration, and source confirmation. Timeliness refers to the data's age, with recently updated data being more reliable. Sources from official registries are more trustworthy. Data confirmed by regulations or multiple sources are also rated higher. Additionally, we distinguish between time-sensitive data, like addresses that require regular updates, and time-independent data, like birthdates. For example, an address from a registry is more reliable than one from the internet.

Try not to get carried away trying to hit all six dimensions - work out which ones will give the best returns and start with the top two. Over stretching and reaching for perfection will merely slow you down in reaching the key next goal which is using the high quality data - quality data itself should not be the target. The goal should be to support the corporate strategy, so pick the quality dimensions will have the biggest impact on that strategy.

When trying to define data quality dimensions, whats been helpful for me is taking a ‘diagnostic approach’, where you start broad on the issues that have highest risk to you (aka your business needs), and then drill down to specifics. Think of it as an annual checkup vs testing for a specific illness.

Defining data quality dimensions tailored to your organization’s specific needs is a strategic exercise that aligns data quality with business goals. Accuracy, completeness, consistency, timeliness, validity, and uniqueness each play a role in how data supports decision-making processes. For each dimension, establish clear criteria, expected values, and measurement methods. This ensures that data quality is not only measurable but also actionable. Consider how each dimension impacts your business objectives: Does data accuracy directly influence customer satisfaction? Can data timeliness affect market responsiveness? Setting up metrics for these dimensions enables targeted improvements and prioritizes efforts aligned w the most value…

Ask your data consumers about their data pain points openly - like a public call to action. Doing that has multiple benefits: 1) You identify data quality issues more easily 2) Your stakeholders can help you prioritize the issues by telling you the business impact 3) It’s a great start for a collaboration because people impacted by data issues love to help fix them By following the pain, you will gain speed, impact and commitment for your data quality management journey.

Tools like data profiling, validation, and cleansing play a critical role in analyzing data for errors, inconsistencies, and anomalies. However, the true impact of data quality issues often surfaces through user experiences. Encourage stakeholders to report discrepancies that affect their decision-making or operational efficiency. Implementing a feedback loop where users can easily report issues ensures that data quality improvement efforts are user-centric and focused on areas that directly impact business operations. Regularly scheduled reviews of reported issues, coupled with comprehensive data analyses, will highlight recurring problems and emerging patterns, guiding prioritization and remediation efforts.

Only focus on the data quality issues that have a significant impact on what you are trying to achieve. This focus requires hard prioritisation and clarity of purpose….cleaning everything will set unrealistic expectations, slow your team down and you will not get closer to your objective.

Technology alone isn’t enough—engage your data users and stakeholders, as their feedback often reveals issues that tools might miss, particularly those affecting decision-making or operations. Regularly soliciting input ensures that you’re addressing the most impactful problems. The key question to consider is: ‘Are we just fixing what’s easy to detect, or are we truly addressing the data issues that hinder our business performance?’ Effective identification combines technical analysis with real-world user insights, ensuring a comprehensive approach to data quality.

Tools can be useful in sampling data quality issues, but they will always be bounded above by their technical scope. There could be many more issues tools can't capture, but people can. Thus, starting with scanning the business environment, holding unstructured (e.g., rich-picture) workshops with resources, reviewing expressed symptoms or identified challenges, etc., are some useful early techniques to corral issues. The next step is to identify patterns of quality issues and to assess -- with resources -- their impact, priority, etc.

Use commercial and stakeholder impact analysis - don't waste time fixing data quality issues that aren't affecting and won't affect the rest of the the org (why do you even have that data?). Consider a second dimension of "complexity to resolve" or similar - this may include financial, procedural or cultural challenges. Leverage stakeholders' willingness to drive the changes necessary: if they're not motivated to act, you can effectively deprioritise. Whichever route you take, make sure it's a collaborative effort between the D&A office and the rest of the org / leadership. Keep proactively engaged with the strategic direction of the org - less impactful data quality issues may grow in importance as new requirements emerge.

Once identified, data quality issues must be prioritized based on criteria such as impact on business operations, frequency of occurrence, and complexity of resolution. Utilizing a data quality scorecard or dashboard provides a visual representation of issues, facilitating easier prioritization and tracking. Consider adopting a weighted scoring system that takes into account both the quantitative and qualitative impact of each issue. This method allows for a more nuanced approach to prioritization, ensuring that resources are allocated to fixing problems that will have the most significant positive effect on business outcomes.

Quantifying impact at both the organizational and operational unit level can be helpful in building support for remediation efforts. When an operational unit can see the direct benefit of resolving a data quality issue, they are more likely to voluntarily resolve the issue. However, if the remediation doesn't directly benefit a single unit, the organizational view can be used to explain to a broader decision making body why it is important to the organization and should get priority. This dual-impact approach leverages the strength of both grass-roots discussions and top-down centralized decision making approaches.

Identifying stakeholders in various business areas who can articulate and quantify the impact of poor data quality on their processes is a crucial step. Fortunately, opportunities to pinpoint these instances are plentiful within any organization. Pain points related to data quality are ubiquitous across business functions, making their identification relatively straightforward. Recognizing these challenges allows for targeted interventions and strategic improvements in data quality management.

To prioritize data quality issues, assess their impact on key business objectives and operations. Consider factors like the severity of the impact on decision-making, compliance risks, and customer satisfaction. Issues directly affecting revenue, regulatory requirements, or critical business processes should take precedence. Also, evaluate the frequency of issues and the effort required for resolution. This helps in allocating resources effectively, focusing first on high-impact, high-frequency problems, and those with a feasible path to improvement, ensuring strategic alignment of data quality initiatives with business goals.

The next step involves actionable remediation. This includes deploying tools for data transformation, enrichment, and deduplication to rectify identified problems. But, how does your organization ensure these actions lead to sustainable improvements in data quality? Establishing clear roles and responsibilities is crucial; designate data stewards for overseeing quality improvements. Addtly, integrating data quality processes into daily workflows prevents future issues. Utilize automation wherever possible to maintain consistency and accuracy. For lasting change, embed data quality checks and controls into your data management systems. This proactive stance ensures data quality is maintained continuously, not just addressed periodically.

Para implementar as a??es, é necessária uma combina??o estratégica de ferramentas e uma governan?a de dados sólida. E algo a considerar é que essas a??es n?o s?o solu??es pontuais e sim partes de um ciclo contínuo de aprimoramento dos dados. E para ter sucesso na implementa??o dessas a??es na organiza??o, é preciso processos e fluxos de trabalho otimizados, e escopos de responsabilidade bem definidos para os responsáveis, para n?o haver brechas no processo que permita output com erro.

Data Quality improvements require time and investment. So, this cannot be a side project and should be prioritized with the overall portfolio. Otherwise we will only see minimal improvements due to lack of resources or change in priority. This is what I have seen in the transformations I have driven in enterprises. Hope this helps.

This is often formulated as a type of data strategy called data quality strategy. Given the pivotal quality challenges identified and prioritized above, what are the actions we will be taking to resolve them? The quality strategy answers that question and orchestrates resources and policies in alignment with other data strategies (ultimately in alignment with business objectives).

Assess data quality: Analyze data sources and flows to identify missing data, duplication, inconsistency, and inaccuracies. Define quality dimensions: Establish completeness, accuracy, consistency, timeliness, validity, and uniqueness to evaluate data quality. Conduct data profiling: Analyze sample datasets for anomalies, outliers, and patterns indicating potential issues. Prioritize issues: Assess impact, severity, frequency, criticality, and compliance to prioritize areas for improvement. Establish metrics: Define measurable metrics aligned with objectives to monitor data quality. Implement controls: Develop validation rules, cleansing techniques, standardization procedures, and error detection mechanisms.

To evaluate the improvement of data quality, I use the "Data Quality Evolution Assessment" (DQEA) method. This involves specific audits and reports to measure progress. Focusing on feedback, DQEA helps identify gaps and opportunities for improvement, understanding the real impact of data quality enhancement actions, and continuing to optimize data quality for more reliable and precise decision-making.

Don't conduct outcome assessments in your own operational silo. Have a public component your assessment that your data customers can review. Identify contributions that were meaningful; quantify, qualify, describe the impact & get additional feedback. If all you have done is evaluate current quality & current outcomes, you haven't done a very good job. While current exploit activities are the "main thing", you must explore at a minimum 5-10% to keep up with a changing world, to grow your teams' capabilities, & to give your staff chances to hit home runs. Mature Explore's should cycle in to replace stale & less productive efforts. Outcomes should benchmark current goals as well measure to the overall vision.

One of the best way to measure data quality outcomes is by tracking the impact of the DQ issues that were fixed. To define impact, it's important to tie it to the business goals of the data strategy in the organization. For example, if the goals are tied to managing regulatory risk and improving finance and risk reporting, the DQ impacts would be tied to the correctness and timeliness of the regulatory, risk and finance reports. If the business goals are growth related, then the impacts might be tied, as an example, to marketing efficacy. This impact classification can also be used to prioritize both the root cause analysis and remediation. That would ensure the alignment of the DQ program with the goals of the overall Data Strategy.

As mentioned previously, this should be the first step because this will is the most efficient way of calculating what needs to be done…. You don’t want to get this far and realise you are barking up the wrong tree?? Additionally, you may believe that data quality is a one off issue - but the quality degrades over time, hence any data quality initiatives must be sustainable over the mid to long term to prevent downstream issues as a result of poor quality data.

The best place to start is to look for where DQ is driving cost into the organisation. For example, if the quality of customer records is driving repeat contact, therefore having an adverse effect on productivity/cost to serve, it would be sensible to look at the root cause and address this. In this way you can demonstrate genuine RoI on DQ investment. By doing this you can build momentum to fix the next issue. Building up a set of tangible outcomes for better DQ helps the organisation see DQ as part of a wider, strategic, quality programme.

It is crucial to understand the business strategy where the data strategy will be implemented. This involves identifying the business objectives that will be supported by the implementation of the data strategy. The definition of value as any factor that enhances the total value created by the company through the adoption of a data strategy, either by capturing greater value from the current business model or by creating new value, and the data that generates this value has to be prioritized for the quality work.

How to identify and prioritize data quality issues effectively: Data Profiling: Perform extensive data profiling to identify issues like missing values, inconsistencies, and duplications. Monitoring helthiness: Define critical metrics, such as accuracy and consistency, to assess situations objectively. Business Impact: Evaluate the impact of each issue on business operations and decision making. Effort: Estimate the difficulty and time needed to fix each problem to aid prioritization. End user feedback: Implement a process to measure improvement effectiveness and possible enhancements. Improving data quality is iterative and requires ongoing attention for reliable data, improved decisions, and business success.

It is important to consider data quality management as a continuous function rather than one-time initiatives due to changing data landscape. Have a comprehensive framework especially around key data related to important metrics to monitor data quality and completeness, with established roles and responsibilities to act on any inconsistencies identified through the process. Potential impact of a data quality issue is a good metric to prioritise the issues. Impact could be identified in context of decision making, compliance, customer experience and cost of fixing.

Identifying and prioritizing data quality issues and improvement opportunities requires a systematic approach. Begin by conducting thorough data assessments to identify inconsistencies, inaccuracies, or incompleteness. Prioritize these issues based on their impact on business processes, decision-making, or regulatory compliance. Establish data quality metrics aligned with organizational goals and define thresholds for acceptable data quality levels. Implement data profiling and monitoring tools to continuously assess data quality and identify areas for improvement. Finally, collaborate cross-functionally to develop and implement data quality improvement initiatives, ensuring ongoing monitoring and refinement of data quality practices.