Category Archives: Data Quality
In 2012, Forbes published an article predicting an upcoming problem.
The Need for Scalable Enterprise Analytics
Specifically, increased exploration in Big Data opportunities would place pressure on the typical corporate infrastructure. The generic hardware used to run most tech industry enterprise applications was not designed to handle real-time data processing. As a result, the explosion of mobile usages, and the proliferation of social networks, was increasing the strain on the system. Most companies now faced real-time processing requirements beyond what the traditional model was designed to handle.
In the past two years, the volume of data and speed of data growth has grown significantly. As a result, the problem has become more severe. It is now clear that these challenges can’t be overcome by simply doubling or tripling their IT spending on infrastructure sprawl. Today, enterprises seek consolidated solutions that offer scalability, performance and ease of administration. The present need is for scalable enterprise analytics.
A Clear Solution Is Available
Informatica PowerCenter and Data Quality is the market leading data integration and data quality platform. This platform has now been certified by Oracle as an optimal solution for both the Oracle Exadata Database Machine and the Oracle SuperCluster.
As the high-speed on-ramp for data into Oracle Exadata, PowerCenter and Data Quality deliver up-to five times faster performance on data load, query, profiling and cleansing tasks. Informatica’s data integration customers can now easily reuse data integration code, skills and resources to access and transform any data from any data source and load it into Exadata, with the highest throughput and scalability.
Customers adopting Oracle Exadata for high-volume, high-speed analytics can now be confident with Informatica PowerCenter and Data Quality. With these products, they can ingest, cleanse and transform all types of data into Exadata with the highest performance and scale required to maximize the value of their Exadata investment.
Proving the Value of Scalable Enterprise Analytics
In order to demonstrate the efficacy of their partnership, the two companies worked together on a Proof Of Value (POV) project. The goal is to prove that using PowerCenter with Exadata would improve both performance and scalability. The project involved PowerCenter and Data Quality 9.6.1 and x4-2 Exadata Machine. Oracle 11g was considered for both standard Oracle and Exadata versions.
The first test conducted a 1TB load test to Exadata and standard Oracle in a typical PowerCenter use case. The second test consisted of querying 1TB profiling warehouse database in Data Quality use case scenario. Performance data was collected for both tests. The scalability factor was also captured. A variant of the TPCH dataset was used to generate the test data. The results were significantly higher than prior Exabyte 1TB test. In particular:
- The data query tests achieved 5x performance.
- The data load tests achieved a 3x-5x speed increase.
- Linear scalability was achieved with read/write tests on Exadata.
What Business Benefits Could You Expect?
Informatica PowerCenter and Data Quality, along-with Oracle Exadata, now provide the best-of-breed combination of software and hardware, optimized to deliver the highest possible total system performance. These comprehensive tools drive agile reporting and analytics, while empowering IT organizations to meet SLAs and quality goals like never before.
- Extend Oracle Exadata’s access to even more business critical data sources. Utilize optimized out-of-the-box Informatica connectivity to easily access hundreds of data sources, including all the major databases, on-premise and cloud applications, mainframe, social data and Hadoop.
- Get more data, more quickly into Oracle Exadata. Move higher volumes of trusted data quickly into Exadata to support timely reporting with up-to-date information (i.e. up to 5x performance improvement compared to Oracle database).
- Centralize management and improve insight into large scale data warehouses. Deliver the necessary insights to stakeholders with intuitive data lineage and a collaborative business glossary. Contribute to high quality business analytics, in a timely manner across the enterprise.
- Instantly re-direct workloads and resources to Oracle Exadata without compromising performance. Leverage existing code and programming skills to execute high-performance data integration directly on Exadata by performing push down optimization.
- Roll-out data integration projects faster and more cost-effectively. Customers can now leverage thousands of Informatica certified developers to execute existing data integration and quality transformations directly on Oracle Exadata, without any additional coding.
- Efficiently scale-up and scale-out. Customers can now maximize performance and lower the costs of data integration and quality operations of any scale by performing Informatica workload and push down optimization on Oracle Exadata.
- Save significant costs involved in administration and expansion. Customers can now easily and economically manage large-scale analytics data warehousing environments with a single point of administration and control, and consolidate a multitude of servers on one rack.
- Reduce risk. Customers can now leverage Informatica’s data integration and quality platform to overcome the typical performance and scalability limitations seen in databases and data storage systems. This will help reduce quality-of-service risks as data volumes rise.
Oracle Exadata is a well-engineered system that offers customers out-of-box scalability and performance on demand. Informatica PowerCenter and Data Quality are optimized to run on Exadata, offering customers business benefits that speed up data integration and data quality tasks like never before. Informatica’s certified, optimized, and purpose-built solutions for Oracle can help you enable more timely and trustworthy reporting. You can now benefit from Informatica’s optimized solutions for Oracle Exadata to make better business decisions by unlocking the full potential of the most current and complete enterprise data available. As shown in our test results, you can attain up to 5x performance by scaling Exadata. Informatica Data Quality customers can perform profiling 1TB datasets, which is unheard before. We urge you to deploy the combined solution to solve your data integration and quality problems today while achieving high speed business analytics in these days of big data exploration and Internet Of Things.
30% or higher of each company’s businesses are unprofitable
According to Jonathan Brynes at the MIT Sloan School, “the most important issue facing most managers …is making more money from their existing businesses without costly new initiatives”. In Brynes’ cross industry research, he found that 30% or higher of each company’s businesses are unprofitable. Brynes claims these business losses are offset by what are “islands of high profitability”. The root cause of this issue is asserted to be the inability of current financial and management control systems to surface profitability problems and opportunities. Why is this the case? Byrnes believes that management budgetary guidance by its very nature assumes the continuation of the status quo. For this reason, the response to management asking for a revenue increase is to increase revenues for businesses that are profitable and unprofitable. Given this, “the areas of embedded unprofitability remain embedded and largely invisible”. At the same time to be completely fair, it should be recognized that it takes significant labor to accurately and completely put together a complete picture on direct and indirect costs.
The CFO needs to become the point person on profitability issues
Byrnes believes, nevertheless, that CFOs need to become the corporate point person for surfacing profitability issues. They, in fact, should act as the leader of a new and important role, the chief profitability officer. This may seem like an odd suggestion since virtually every CFO if asked would view profitability as a core element of their job. But Byrnes believes that CFOs need to move beyond broad, departmental performance measures and build profitability management processes into their companies’ core management activities. This task requires the CFO to determine two things.
- Which product lines, customers, segments, and channels are unprofitable so investments can be reduced or even eliminated?
- Which product lines, customers, segments, and channels are the most profitable so management can determine whether to expand investments and supporting operations?
Why didn’t portfolio management solve this problem?
Now as a strategy MBA, Byrnes’ suggestion leave me wondering why the analysis proposed by strategy consultants like Boston Consulting Group didn’t solve this problem a long time ago. After all portfolio analysis has at its core the notion that relative market share and growth rate will determine profitability and which businesses a firm should build share, hold share, harvest share, or divest share—i.e. reduce, eliminate, or expand investment. The truth is getting at these figures, especially profitability, is a time consuming effort.
KPMG finds 91% of CFOs are held back by financial and performance systems
As financial and business systems have become more complex, it has become harder and harder to holistically analyze customer and product profitability because the relevant data is spread over a myriad of systems, technologies, and locations. For this reason, 91% of CFO respondents in a recent KPMG survey said that they want to improve the quality of their financial and performance insight from the data they produce. An amazing 51% of these CFOs, also, admitted that the “collection, storage, and retrieval financial and performance data at their company is primarily a manual and/or spreadsheet-based exercise”. Think about it — a majority of these CFOs teams time is spent collecting financial data rather than actively managing corporate profitability.
How do we fix things?
What is needed is a solution that allows financial teams to proactively produce trustworthy financial data from each and every financial system and then reliably combine and aggregate the data coming from multiple financial systems. Having accomplished this, the solution needs to allow financial organizations to slice and dice net profitability for product lines and customers.
This approach would not only allow financial organizations to cut their financial operational costs but more importantly drive better business profitability by surfacing profitability gaps. At the same time, it would enable financial organizations to assist business units in making more informed customer and product line investment decisions. If a product line or business is narrowly profitable and lacks a broader strategic context or ability to increase profitability by growing market share, it is a candidate for investment reduction or elimination.
Strategic CFOs need to start asking questions of their business counterparts starting with their justification for their investment strategy. Key to doing this involves consolidating reliable profitability data across customers, products, channel partners, suppliers. This would eliminate the time spent searching for and manually reconciling data in different formats across multiple systems. It should deliver ready analysis across locations, applications, channels, and departments.
Some parting thoughts
Strategic CFOs tell us they are trying to seize the opportunity “to be a business person versus a bean counting historically oriented CPA”. I believe a key element of this is seizing the opportunity to become the firm’s chief profitability officer. To do this well, CFOs need dependable data that can be sliced and diced by business dimensions. Armed with this information, CFOs can determine the most and least profitability, businesses, product lines, and customers. As well, they can come to the business table with the perspective to help guide their company’s success.
Solution Brief: The Intelligent Data Platform
CFOs Discuss Their Technology Priorities
The CFO Viewpoint upon Data
How CFOs can change the conversation with their CIO?
New type of CFO represents a potent CIO ally
Competing on Analytics
The Business Case for Better Data Connectivity
When it comes to cloud-based data analytics, a recent study by Ventana Research (as found in Loraine Lawson’s recent blog post) provides a few interesting data points. The study reveals that 40 percent of respondents cited lowered costs as a top benefit, improved efficiency was a close second at 39 percent, and better communication and knowledge sharing also ranked highly at 34 percent.
Ventana Research also found that organizations cite a unique and more complex reason to avoid cloud analytics and BI. Legacy integration work can be a major hindrance, particularly when BI tools are already integrated with other applications. In other words, it’s the same old story:
The ability to deal with existing legacy systems when moving to concepts such as big data or cloud-based analytics is critical to the success of any enterprise data analytics strategy. However, most enterprises don’t focus on data integration as much as they should, and hope that they can solve the problems using ad-hoc approaches.
You can’t make sense of data that you can’t see.
These approaches rarely work as well a they should, if at all. Thus, any investment made in data analytics technology is often diminished because the BI tools or applications that leverage analytics can’t see all of the relevant data. As a result, only part of the story is told by the available data, and those who leverage data analytics don’t rely on the information, and that means failure.
What’s frustrating to me about this issue is that the problem is easily solved. Those in the enterprise charged with standing up data analytics should put a plan in place to integrate new and legacy systems. As part of that plan, there should be a common understanding around business concepts/entities of a customer, sale, inventory, etc., and all of the data related to these concepts/entities must be visible to the data analytics engines and tools. This requires a data integration strategy, and technology.
As enterprises embark on a new day of more advanced and valuable data analytics technology, largely built upon the cloud and big data, the data integration strategy should be systemic. This means mapping a path for the data from the source legacy systems, to the views that the data analytics systems should include. What’s more, this data should be in real operational time because data analytics loses value as the data becomes older and out-of-date. We operate a in a real-time world now.
So, the work ahead requires planning to occur at both the conceptual and physical levels to define how data analytics will work for your enterprise. This includes what you need to see, when you need to see it, and then mapping a path for the data back to the business-critical and, typically, legacy systems. Data integration should be first and foremost when planning the strategy, technology, and deployments.
In my last blog I promised I would report back my experience on using Informatica Data Quality, a software tool that helps automate the hectic, tedious data plumbing task, a task that routinely consumes more than 80% of the analyst time. Today, I am happy to share what I’ve learned in the past couple of months.
But first, let me confess something. The reason it took me so long to get here was that I was dreaded by trying the software. Never a savvy computer programmer, I was convinced that I would not be technical enough to master the tool and it would turn into a lengthy learning experience. The mental barrier dragged me down for a couple of months and I finally bit the bullet and got my hands on the software. I am happy to report that my fear was truly unnecessary – It took me one half day to get a good handle on most features in the Analyst Tool, a component of the Data Quality designed for analyst and business users, then I spent 3 days trying to figure out how to maneuver the Developer Tool, another key piece of the Data Quality offering mostly used by – you guessed it, developers and technical users. I have to admit that I am no master of the Developer Tool after 3 days of wrestling with it, but, I got the basics and more importantly, my hands-on interaction with the entire software helped me understand the logic behind the overall design, and see for myself how analyst and business user can easily collaborate with their IT counterpart within our Data Quality environment.
To break it all down, first comes to Profiling. As analyst we understand too well the importance of profiling as it provides an anatomy of the raw data we collected. In many cases, it is a must have first step in data preparation (especially when our raw data came from different places and can also carry different formats). A heavy user of Excel, I used to rely on all the tricks available in the spreadsheet to gain visibility of my data. I would filter, sort, build pivot table, make charts to learn what’s in my raw data. Depending on how many columns in my data set, it could take hours, sometimes days just to figure out whether the data I received was any good at all, and how good it was.
Switching to the Analyst Tool in Data Quality, learning my raw data becomes a task of a few clicks – maximum 6 if I am picky about how I want it to be done. Basically I load my data, click on a couple of options, and let the software do the rest. A few seconds later I am able to visualize the statistics of the data fields I choose to examine, I can also measure the quality of the raw data by using Scorecard feature in the software. No more fiddling with spreadsheet and staring at busy rows and columns. Take a look at the above screenshots and let me know your preference?
Once I decide that my raw data is adequate enough to use after the profiling, I still need to clean up the nonsense in it before performing any analysis work, otherwise bad things can happen — we call it garbage in garbage out. Again, to clean and standardize my data, Excel came to rescue in the past. I would play with different functions and learn new ones, write macro or simply do it by hand. It was tedious but worked if I worked on static data set. Problem however, was when I needed to incorporate new data sources in a different format, many of the previously built formula would break loose and become inapplicable. I would have to start all over again. Spreadsheet tricks simply don’t scale in those situation.
With Data Quality Analyst Tool, I can use the Rule Builder to create a set of logical rules in hierarchical manner based on my objectives, and test those rules to see the immediate results. The nice thing is, those rules are not subject to data format, location, or size, so I can reuse them when the new data comes in. Profiling can be done at any time so I can re-examine my data after applying the rules, as many times as I like. Once I am satisfied with the rules, they will be passed on to my peers in IT so they can create executable rules based on the logic I create and run them automatically in production. No more worrying about the difference in format, volume or other discrepancies in the data sets, all the complexity is taken care of by the software, and all I need to do is to build meaningful rules to transform the data to the appropriate condition so I can have good quality data to work with for my analysis. Best part? I can do all of the above without hassling my IT – feeling empowered is awesome!
Use the right tool for the right job will improve our results, save us time, and make our jobs much more enjoyable. For me, no more Excel for data cleansing after trying our Data Quality software, because now I can get a more done in less time, and I am no longer stressed out by the lengthy process.
I encourage my analyst friends to try Informatica Data Quality, or at least the Analyst Tool in it. If you are like me, feeling weary about the steep learning curve then fear no more. Besides, if Data Quality can cut down your data cleansing time by half (mind you our customers have reported higher numbers), how many more predictive models you can build, how much you will learn, and how much faster you can build your reports in Tableau, with more confidence?
Last time I talked about how benchmark data can be used in IT and business use cases to illustrate the financial value of data management technologies. This time, let’s look at additional use cases, and at how to philosophically interpret the findings.
So here are some additional areas of investigation for justifying a data quality based data management initiative:
- Compliance or any audits data and report preparation and rebuttal (FTE cost as above)
- Excess insurance premiums on incorrect asset or party information
- Excess tax payments due to incorrect asset configuration or location
- Excess travel or idle time between jobs due to incorrect location information
- Excess equipment downtime (not revenue generating) or MTTR due to incorrect asset profile or misaligned reference data not triggering timely repairs
- Equipment location or ownership data incorrect splitting service cost or revenues incorrectly
- Party relationship data not tied together creating duplicate contacts or less relevant offers and lower response rates
- Lower than industry average cross-sell conversion ratio due to inability to match and link departmental customer records and underlying transactions and expose them to all POS channels
- Lower than industry average customer retention rate due to lack of full client transactional profile across channels or product lines to improve service experience or apply discounts
- Low annual supplier discounts due to incorrect or missing alternate product data or aggregated channel purchase data
I could go on forever, but allow me to touch on a sensitive topic – fines. Fines, or performance penalties by private or government entities, only make sense to bake into your analysis if they happen repeatedly in fairly predictable intervals and are “relatively” small per incidence. They should be treated like M&A activity. Nobody will buy into cost savings in the gazillions if a transaction only happens once every ten years. That’s like building a business case for a lottery win or a life insurance payout with a sample size of a family. Sure, if it happens you just made the case but will it happen…soon?
Use benchmarks and ranges wisely but don’t over-think the exercise either. It will become paralysis by analysis. If you want to make it super-scientific, hire an expensive consulting firm for a 3 month $250,000 to $500,000 engagement and have every staffer spend a few days with them away from their day job to make you feel 10% better about the numbers. Was that worth half a million dollars just in 3rd party cost? You be the judge.
In the end, you are trying to find out and position if a technology will fix a $50,000, $5 million or $50 million problem. You are also trying to gauge where key areas of improvement are in terms of value and correlate the associated cost (higher value normally equals higher cost due to higher complexity) and risk. After all, who wants to stand before a budget committee, prophesy massive savings in one area and then fail because it would have been smarter to start with something simpler and quicker win to build upon?
The secret sauce to avoiding this consulting expense and risk is a natural curiosity, willingness to do the legwork of finding industry benchmark data, knowing what goes into them (process versus data improvement capabilities) to avoid inappropriate extrapolation and using sensitivity analysis to hedge your bets. Moreover, trust an (internal?) expert to indicate wider implications and trade-offs. Most importantly, you have to be a communicator willing to talk to many folks on the business side and have criminal interrogation qualities, not unlike in your run-of-the-mill crime show. Some folks just don’t want to talk, often because they have ulterior motives (protecting their legacy investment or process) or hiding skeletons in the closet (recent bad performance). In this case, find more amenable people to quiz or pry the information out of these tough nuts, if you can.
Lastly; if you find ROI numbers, which appear astronomical at first, remember that leverage is a key factor. If a technical capability touches one application (credit risk scoring engine), one process (quotation), one type of transaction (talent management self-service), a limited set of people (procurement), the ROI will be lower than a technology touching multiple of each of the aforementioned. If your business model drives thousands of high-value (thousands of dollars) transactions versus ten twenty-million dollar ones or twenty-million one-dollar ones, your ROI will be higher. After all, consider this; retail e-mail marketing campaigns average an ROI of 578% (softwareprojects.com) and this with really bad data. Imagine what improved data can do just on that front.
I found massive differences between what improved asset data can deliver in a petrochemical or utility company versus product data in a fashion retailer or customer (loyalty) data in a hospitality chain. The assertion of cum hoc ergo propter hoc is a key assumption how technology delivers financial value. As long as the business folks agree or can fence in the relationship, you are on the right path.
What’s your best and worst job to justify someone giving you money to invest? Share that story.
Sometimes, the choice of a name has unexpected consequences. Often these consequences aren’t fair. But they exist, nonetheless. For an example of this, consider the well-known study by the National Bureau of Economic Research study that compares the hiring prospects of candidates with identical resumes, but different names. During the study, titled a “Field Experiment on Labor Market Discrimination,” employers were found to be more likely to reply candidates with popular, traditionally Caucasian names than to candidates with either unique, eclectic names or with traditionally African-American names. Though these biases are clearly unfair to the candidates, they do illustrate a key point: One’s choice when naming something can come with perceptions that influence outcomes.
For an example from the IT world, consider my recent engagement at a regional retail bank. In this engagement, half of the meeting time was consumed by IT and business leaders debating how to label their Master Data Management (MDM) Initiative. Consider these excerpts:
- Should we even call it MDM? Answer: No. Why? Because nobody on the business side will understand what that means. Also, as we just implemented a Data Warehouse/Mart last year and we are in the middle of our new CRM roll-out, everybody in business and retail banking will assume their data is already mastered in both of these. On a side note; telcos understand MDM as Mobile Device Management.
- Should we call it “Enterprise Data Master’? Answer: No. Why? Because unless you roll out all data domains and all functionality (standardization, matching, governance, hierarchy management, etc.) to the whole enterprise, you cannot. And doing so is a bad idea as it is with every IT project. Boiling the ocean and going live with a big bang is high cost, high risk and given shifting organizational strategies and leadership, quick successes are needed to sustain the momentum.
- Should we call it “Data Warehouse – Release 2”? Answer: No. Why? Because it is neither a data warehouse, nor a version 2 of one. It is a backbone component required to manage a key organizational ingredient – data –in a way that it becomes useful to many use cases, processes, applications and people, not just analytics, although it is often the starting block. Data warehouses have neither been conceived nor designed to facilitate data quality (they assume it is there already) nor are they designed for real time interactions. Did anybody ask if ETL is “Pneumatic Tubes – Version 2”?
- Should we call it “CRM Plus”? Answer: No. Why? Because it has never intended or designed to handle the transactional volume and attribution breadth of high volume use cases, which are driven by complex business processes. Also, if it were a CRM system, it would have a more intricate UI capability beyond comparatively simple data governance workflows and UIs.
Consider this; any data quality solution like MDM, makes any existing workflow or application better at what it does best: manage customer interactions, create orders, generate correct invoices, etc. To quote a colleague “we are the BASF of software”. Few people understand what a chemical looks like or does but it makes a plastic container sturdy, transparent, flexible and light.
I also explained hierarchy management in a similar way. Consider it the LinkedIn network of your company, which you can attach every interaction and transaction to. I can see one view, people in my network see a different one and LinkedIn has probably the most comprehensive view but we are all looking at the same core data and structures ultimately.
So let’s call the “use” of your MDM “Mr. Clean”, aka Meister Proper, because it keeps everything clean.
While naming is definitely a critical point to consider given the expectations, fears and reservations that come with MDM and the underlying change management, it was hilarious to see how important it suddenly was. However, it was puzzling to me (maybe a naïve perspective) why mostly recent IT hires had to categorize everything into new, unique functional boxes, while business and legacy IT people wanted to re-purpose existing boxes. I guess, recent IT used their approach to showcase that they were familiar with new technologies and techniques, which was likely a reason for their employment. Business leaders, often with the exception of highly accomplished and well regarded ones, as well as legacy IT leaders, needed to reassure continuity and no threat of disruption or change. Moreover, they also needed to justify their prior software investments’ value proposition.
Aside from company financial performance and regulatory screw-ups, legions of careers will be decide if, how and how successful this initiative will be.
Naming a new car model for a 100,000 production run or a shampoo for worldwide sales could not face much more scrutiny. Software vendors give their future releases internal names of cities like Atlanta or famous people like Socrates instead of descriptive terms like “Gamification User Interface Release” or “Unstructured Content Miner”. This may be a good avenue for banks and retailers to explore. It would avoid the expectation pitfalls associated with names like “Customer Success Data Mart”, “Enterprise Data Factory”, “Data Aggregator” or “Central Property Repository”. In reality, there will be many applications, which can claim bits and pieces of the same data, data volume or functionality. Who will make the call on which one will be renamed or replaced to explain to the various consumers what happened to it and why.
You can surely name any customer facing app something more descriptive like “Payment Central” or “Customer Success Point” but the reason why you can do this is that the user will only have one or maybe two points to interface with the organization. Internal data consumers will interact many more repositories. Similarly, I guess this is all the reason why I call my kids by their first name and strangers label them by their full name, “Junior”, “Butter Fingers” or “The Fast Runner”.
I would love to hear some other good reasons why naming conventions should be more scrutinized. Maybe you have some guidance on what should and should not be done and the reasons for it?
Total Quality Management, as it relates to products and services has it’s roots in the 1920s. The 1960’s provided a huge boost with rise of guru’s such as Deming, Juran and Crosby. Whilst each had their own contribution, common principles for TQM that emerged in this era remain in practice today:
- Management (C-level management) is ultimately responsible for quality
- Poor quality has a cost
- The earlier in the process you address quality, the lower the cost of correcting it
- Quality should be designed into the system
So for 70 years industry in general has understood the cost of poor quality, and how to avoid these costs. So why is it that in 2014 I was party to a conversation that included the statement:
“I only came to the conference to see if you (Informatica) have solved the data quality problem.”
Ironically the TQM movement was only possible based on the analysis of data, but this is the one aspect that is widely ignored during TQM implementation. So much for ‘Total’ Quality Management.
This person is not alone in their thoughts. Many are waiting for the IT knight in shining armour, the latest and greatest data quality tools secured on their majestic steed, to ride in and save the day. Data quality dragon slayed, cold drinks all round, job done. This will not happen. Put data quality in the context of total quality management principles to see why: A single department cannot deliver data quality alone, regardless of the strength of their armoury.
I am not sure anyone would demand a guarantee of a high quality product from their machinery manufacturers. Communications providers cannot deliver high quality customer services organisations through technology alone. These suppliers have will have an influence on final product quality, but everyone understands equipment cannot deliver in isolation. Good quality raw materials, staff that genuinely takes pride in their work and the correct incentives are key to producing high quality products and services.
So why is there an expectation that data quality can be solved by tools alone?
At a minimum senior management support is required to push other departments to change their behaviour and/or values. So why aren’t senior management convinced that data quality is a problem worth their attention the way product & service quality is?
The fact that poor data quality has a high cost is reasonably well known via anecdotes. However, cost has not been well quantified, and hence fails to grab the attention of senior management. A 2005 paper by Richard Marsh[i] states: “Research and reports by industry experts, including Gartner Group, PriceWaterhouseCoopers and The Data Warehousing Institute clearly identify a crisis in data quality management and a reluctance among senior decision makers to do enough about it.” Little has changed since 2005.
However, we are living in a world where data generation, processing and consumption are increasing exponentially. With all the hype and investment in data, we face the grim prospect of fully embracing an age of data-driven-everything founded on a very poor quality raw material. Data quality is expected to be applied after generation, during the analytic phase. How much will that cost us? In order to function effectively, our new data-driven world must have high quality data running through every system and activity in an organization.
The Total Data Quality Movement is long overdue.
Only when every person in every organization understands the value of the data, do we have a chance of collectively solving the problem of poor data quality. Data quality must be considered from data generation, through transactional processing and analysis right until the point of archiving.
Informatica DQ supports IT departments in automating data correction where possible, and highlighting poor data for further attention where automation is not possible. MDM plays an important role in sustaining high quality data. Informatica tools empower the business to share the responsibility for total data quality.
We are ready for Total Data Quality, but continue to await the Total Data Quality Movement to get off the ground.
(If you do not have time to waiting for TDQM to gain traction, we can help you measure the cost of poor quality data in your organization to win corporate buy-in now.)
That second question is a killer because most people — no matter if they’re in marketing, sales or manufacturing — rely on incomplete, inaccurate or just plain wrong information. Regardless of industry, we’ve been fixated on historic transactions because that’s what our systems are designed to provide us.
“Moneyball: The Art of Winning an Unfair Game” gives a great example of what I mean. The book (not the movie) describes Billy Beane hiring MBAs to map out the factors that would win a baseball game. They discovered something completely unexpected: That getting more batters on base would tire out pitchers. It didn’t matter if batters had multi-base hits, and it didn’t even matter if they walked. What mattered was forcing pitchers to throw ball after ball as they faced an unrelenting string of batters. Beane stopped looking at RBIs, ERAs and even home runs, and started hiring batters who consistently reached first base. To me, the book illustrates that the most useful knowledge isn’t always what we’ve been programmed to depend on or what is delivered to us via one app or another.
For years, people across industries have turned to ERP, CRM and web analytics systems to forecast sales and acquire new customers. By their nature, such systems are transactional, forcing us to rely on history as the best predictor of the future. Sure it might be helpful for retailers to identify last year’s biggest customers, but that doesn’t tell them whose blogs, posts or Tweets influenced additional sales. Isn’t it time for all businesses, regardless of industry, to adopt a different point of view — one that we at Informatica call “Data-First”? Instead of relying solely on transactions, a data-first POV shines a light on interactions. It’s like having a high knowledge IQ about relationships and connections that matter.
A data-first POV changes everything. With it, companies can unleash the killer app, the killer sales organization and the killer marketing campaign. Imagine, for example, if a sales person meeting a new customer knew that person’s concerns, interests and business connections ahead of time? Couldn’t that knowledge — gleaned from Tweets, blogs, LinkedIn connections, online posts and transactional data — provide a window into the problems the prospect wants to solve?
That’s the premise of two startups I know about, and it illustrates how a data-first POV can fuel innovation for developers and their customers. Today, we’re awash in data-fueled things that are somehow attached to the Internet. Our cars, phones, thermostats and even our wristbands are generating and gleaning data in new and exciting ways. That’s knowledge begging to be put to good use. The winners will be the ones who figure out that knowledge truly is power, and wield that power to their advantage.
A mid-sized insurer recently approached our team for help. They wanted to understand how they fell short in making their case to their executives. Specifically, they proposed that fixing their customer data was key to supporting the executive team’s highly aggressive 3-year growth plan. (This plan was 3x today’s revenue). Given this core organizational mission – aside from being a warm and fuzzy place to work supporting its local community – the slam dunk solution to help here is simple. Just reducing the data migration effort around the next acquisition or avoiding the ritual annual, one-off data clean-up project already pays for any tool set enhancing data acquisitions, integration and hygiene. Will it get you to 3x today’s revenue? It probably won’t. What will help are the following:
Hard cost avoidance via software maintenance or consulting elimination is the easy part of the exercise. That is why CFOs love it and focus so much on it. It is easy to grasp and immediate (aka next quarter).
Soft cost reduction, like staff redundancies are a bit harder. Despite them being viable, in my experience very few decision makers want work on a business case to lay off staff. My team had one so far. They look at these savings as freed up capacity, which can be re-deployed more productively. Productivity is also a bit harder to quantify as you typically have to understand how data travels and gets worked on between departments.
However, revenue effects are even harder and esoteric to many people as they include projections. They are often considered “soft” benefits, although they outweigh the other areas by 2-3 times in terms of impact. Ultimately, every organization runs their strategy based on projections (see the insurer in my first paragraph).
The hardest to quantify is risk. Not only is it based on projections – often from a third party (Moody’s, TransUnion, etc.) – but few people understand it. More often, clients don’t even accept you investigating this area if you don’t have an advanced degree in insurance math. Nevertheless, risk can generate extra “soft” cost avoidance (beefing up reserve account balance creating opportunity cost) but also revenue (realizing a risk premium previously ignored). Often risk profiles change due to relationships, which can be links to new “horizontal” information (transactional attributes) or vertical (hierarchical) from parent-child relationships of an entity and the parent’s or children’s transactions.
Given the above, my initial advice to the insurer would be to look at the heartache of their last acquisition, use a benchmark for IT productivity from improved data management capabilities (typically 20-26% – Yankee Group) and there you go. This is just the IT side so consider increasing the upper range by 1.4x (Harvard Business School) as every attribute change (last mobile view date) requires additional meetings on a manager, director and VP level. These people’s time gets increasingly more expensive. You could also use Aberdeen’s benchmark of 13hrs per average master data attribute fix instead.
You can also look at productivity areas, which are typically overly measured. Let’s assume a call center rep spends 20% of the average call time of 12 minutes (depending on the call type – account or bill inquiry, dispute, etc.) understanding
- Who the customer is
- What he bought online and in-store
- If he tried to resolve his issue on the website or store
- How he uses equipment
- What he cares about
- If he prefers call backs, SMS or email confirmations
- His response rate to offers
- His/her value to the company
If he spends these 20% of every call stringing together insights from five applications and twelve screens instead of one frame in seconds, which is the same information in every application he touches, you just freed up 20% worth of his hourly compensation.
Then look at the software, hardware, maintenance and ongoing management of the likely customer record sources (pick the worst and best quality one based on your current understanding), which will end up in a centrally governed instance. Per DAMA, every duplicate record will cost you between $0.45 (party) and $0.85 (product) per transaction (edit touch). At the very least each record will be touched once a year (likely 3-5 times), so multiply your duplicated record count by that and you have your savings from just de-duplication. You can also use Aberdeen’s benchmark of 71 serious errors per 1,000 records, meaning the chance of transactional failure and required effort (% of one or more FTE’s daily workday) to fix is high. If this does not work for you, run a data profile with one of the many tools out there.
If standardization of records (zip codes, billing codes, currency, etc.) is the problem, ask your business partner how many customer contacts (calls, mailing, emails, orders, invoices or account statements) fail outright and/or require validation because of these attributes. Once again, if you apply the productivity gains mentioned earlier, there are you savings. If you look at the number of orders that get delayed in form of payment or revenue recognition and the average order amount by a week or a month, you were just able to quantify how much profit (multiply by operating margin) you would be able to pull into the current financial year from the next one.
The same is true for speeding up the introduction or a new product or a change to it generating profits earlier. Note that looking at the time value of funds realized earlier is too small in most instances especially in the current interest environment.
If emails bounce back or snail mail gets returned (no such address, no such name at this address, no such domain, no such user at this domain), e(mail) verification tools can help reduce the bounces. If every mail piece (forget email due to the miniscule cost) costs $1.25 – and this will vary by type of mailing (catalog, promotion post card, statement letter), incorrect or incomplete records are wasted cost. If you can, use fully loaded print cost incl. 3rd party data prep and returns handling. You will never capture all cost inputs but take a conservative stab.
If it was an offer, reduced bounces should also improve your response rate (also true for email now). Prospect mail response rates are typically around 1.2% (Direct Marketing Association), whereas phone response rates are around 8.2%. If you know that your current response rate is half that (for argument sake) and you send out 100,000 emails of which 1.3% (Silverpop) have customer data issues, then fixing 81-93% of them (our experience) will drop the bounce rate to under 0.3% meaning more emails will arrive/be relevant. This in turn multiplied by a standard conversion rate (MarketingSherpa) of 3% (industry and channel specific) and average order (your data) multiplied by operating margin gets you a benefit value for revenue.
If product data and inventory carrying cost or supplier spend are your issue, find out how many supplier shipments you receive every month, the average cost of a part (or cost range), apply the Aberdeen master data failure rate (71 in 1,000) to use cases around lack of or incorrect supersession or alternate part data, to assess the value of a single shipment’s overspend. You can also just use the ending inventory amount from the 10-k report and apply 3-10% improvement (Aberdeen) in a top-down approach. Alternatively, apply 3.2-4.9% to your annual supplier spend (KPMG).
You could also investigate the expediting or return cost of shipments in a period due to incorrectly aggregated customer forecasts, wrong or incomplete product information or wrong shipment instructions in a product or location profile. Apply Aberdeen’s 5% improvement rate and there you go.
Consider that a North American utility told us that just fixing their 200 Tier1 suppliers’ product information achieved an increase in discounts from $14 to $120 million. They also found that fixing one basic out of sixty attributes in one part category saves them over $200,000 annually.
So what ROI percentages would you find tolerable or justifiable for, say an EDW project, a CRM project, a new claims system, etc.? What would the annual savings or new revenue be that you were comfortable with? What was the craziest improvement you have seen coming to fruition, which nobody expected?
Next time, I will add some more “use cases” to the list and look at some philosophical implications of averages.