“Trust, but verify” is a Russian saying which gained worldwide recognition when Ronald Reagan used this in the context of nuclear disarmament discussions with the Soviet Union. We have seen a lot of change in the years since then. Indeed, the only constant is change. This proverb holds as true today as it did then, and your organization’s data is important to frame in this proverb’s message.
The advances in technology are so fast that they can seem to be a passing blur. In the world of data there are commonly said to be “The 3 Vs”. They stand for Velocity (how fast that data is coming at us), Volume (how much data we are aggregating), Variety (all the different forms data takes) and a fourth V often added is Veracity. Is the mass of various, fast moving data to be trusted?
Add to the 4 Vs of data the growth in artificial intelligence (AI) and we have a number of things to ponder. In a recent SAS-sponsored research report performed and published by MIT Sloan Management Review Connections, entitled Data, Analytics, & AI; How Trust Delivers Value it highlights that currently it is estimated that only 7% of organizations apply machine learning (ML) and AI in decision making or production workflows. This number will no doubt grow as organizations realize just a few of the benefits of AI (source: EU Business School’s “5 Benefits of AI for Business”):
- Improved customer communications and cost savings
- Strengthening brand loyalty with personalization
- Streamlining the hiring process
- Increase in forecasting accuracy
- Unlocking opportunities
How to Start Building Trust in Data
Countering the Russian proverb to an extent, in MIT SMR Connections’ research report it is asserted that one should begin with the premise that no data is trustworthy. The report further stresses that organizations need to assure that the data on which decisions are being made is well- structured; “you can’t do it if the foundation is held together with duct tape and staples.”
As organizations begin to adopt ML and AI there are a number of foundational best practices to help assure that data is trustworthy before giving the robots the keys to the kingdom:
- Pair internal data scientists with subject matter professionals in your organization to get a more holistic view of how the data is and/or should be managed, helping data-driven innovation.
- Conduct a thorough data quality assessment before embarking on any data architecture change, perhaps most importantly one encompassing AI.
- Consult data specialists to discuss areas for improvement in your data, develop an approach to correct important areas, cleanse relevant databases in their entirety and then maintain the integrity of the database through an ongoing cleansing schedule.
- Look to regulations as business enablers, like the European General Data Protection Regulation (GDPR) as opportunities to assess organizational readiness and to build trust with customers.
- Data minimalization: Ask, is the data needed? If no value is being created from a particular data attribute, why collect it? In ML/AI pre-processing context this may be referred to as “feature extraction”.
- Data preparation: The key steps to take in pursuit of ML/AI, or arguably any data quality initiative, include data normalization, error correction, deduplication, and other core data quality steps which will come to light specific to your organizational idiosyncrasies as the bulleted items above are undertaken.
The MIT SMR Connections research report addresses the importance of closing what is referred to as the “utility gap”; having access not only to the mass of available data, but also having the right data to help inform good business decisions.
Assessing the trustworthiness of your data may include, to name just a few key areas in a consumer context:
- Data validation
- Identification of contactable customers and prospects
- Matching data across channels, brands and markets
- Identification of customer activity and engagement
What School Did Your Doctor Go To?
Is the data source to be trusted? Whether your organization utilizes AI or more traditional data management and analytics, having trust in the origin and current state of your data assets is of paramount importance. In the context of ML and AI, imagine if you will, you are at your new doctor’s office awaiting an important test result. Now, it has just come to your attention that the study materials from which that doctor gained all their knowledge is inaccurate. Will you trust the test results? Data is the foundation upon which decisions are made. The quality of those decisions depends greatly on the veracity of the source data. Admittedly, an important medical test result based on inaccurate data is more severe than, say, a consumer analytics initiative driving your key business decisions. Both are important in their own context and require a sound, trustworthy foundation of data from which good decisions can be made.
How to Champion Emerging Technologies: The People in AI
A promise of AI is the benefit of revealing to us humans unexpected insights from which sound decisions can be made to one’s advantage. Trust the data, but verify. As concluded in a recent Forbes.com article, How Can Data Quality Enhance Trust in Artificial Intelligence “The human element within AI can also be improved. Enhancing trust between humans and machines can start with the initial hiring process — businesses need to hire data scientists without ‘tunnel vision’ to eliminate potential biases. Data scientists who are on the lookout for data that doesn't conform, rather than data that does, will be more likely to spot biases that may be affecting the quality of the data and, therefore, the effectiveness of artificial intelligence.” The MIT SMR Connections report shines a light on the importance of having champions for driving emerging technologies throughout your organization, with individuals and teams in operating units making up the largest percentage at 26% (see figure below from the MIT SMR Connections report).
Image source: MIT SMR Connections
Getting Budgets Approved: It’s the Tip of the Iceberg
The MIT SMR Connections report offers insight into how to go about getting a better data strategy and its architecture approved by top management. Show them what doesn’t yet exist through visualization of what is being proposed. One of the research report respondents uses an iceberg analogy; “likening the 10% floating on top of the water to the analytics results leaders need, while the 90% that floats under the water is my data architecture. They understand we actually have to have a data strategy to enable analytics and decision making.”
A successful AI initiative—or any data-centric initiative—will be a balance of people, process and technology. We must learn to trust in each of these to be effective. Randy Guard, EVP and CMO at SAS concludes the research report succinctly; “Technology and trust go hand in hand, especially when it comes to the future of data.”
Remember, as the proverb goes, Доверяй, но проверяй. Trust but Verify!