Transformation and AI

by Mark Schwartz | on

We’ve all been busily transforming for the last few years. Now there’s this big AI thing. How does it relate to the rest of our transformation? Should we be changing or rethinking our transformation plans?

The easy answer is no, but there are some subtleties. We transform to increase our agility in the face of change, and we accept that the future will see major changes and disruptions. The sudden attention demanded by AI is just a vindication of that belief. You can even test yourself: can your company respond quickly and effectively to the sudden, disruptive intrusion AI represents? If so, you may be further along in your transformation than you thought. As we’ve always said, responding appropriately requires setting up nimble governance processes, an ability to experiment, a culture of innovation, and automation of the good practices, security, and resilience you require.

Digging a bit deeper, I think AI will require exercising some of those transformed muscles that we might not have used yet. I watched a great presentation a few days ago on the wonderful things a company could do with generative AI by combining data from several of their internal databases and using it to tune an LLM. Hmmm—they’ve been trying for years to combine data from those databases to feed analytics without success. Generative AI is not magically going to solve that problem! Generative AI gives us revolutionary ways to use data to capture value. But using it still requires solving the problems we’ve always had—at least pretransformation.

There were probably several issues in the way of combining the data. First, the data probably sits in both technical silos and organizational silos—and organizational silos often hug their data and avoid happily making it available for integration. Second, the organization may not have designed for privacy controls, so a lot of work may be involved in making privacy decisions and establishing guardrails. Third, the data may be locked up in proprietary, legacy databases that make it difficult to extract and combine. Fourth, the quality of the data may not be acceptable, or it may be impossible to match records from one data source to another (lack of common identifiers).

The fifth consideration may be the biggest and most interesting: a simple lack of investment in extracting and combining data. IT departments are busy building new functionality; they can’t do everything! The work involved in combining disparate data might simply not be on the task list, or it might not be a “priority.” It might be falling through governance or prioritization cracks.

With the goal of becoming more agile, transformation may already help with these constraints. As legacy technology is updated, it may become easier (or less time-consuming) to extract data, and therefore not as much of an opportunity cost for the limited IT capacity. Making governance processes nimbler may make it easier for the company to redirect its focus to moving the data now that its potential value is higher due to generative AI. The cultural shift that aligns everyone behind important business objectives will reduce data hugging; a well-designed privacy approach will make data broadly available while still controlling access.

Another reason AI doesn’t really change the nature of digital transformation is that transformation already has companies moving toward better management of innovation. By reducing the cost and risk of experimentation and changing culture to encourage it, organizations are on the way to making continuous innovation possible. Generative AI gives us many powerful new opportunities for innovation—and that’s precisely what it will take to create business value with it. Success with generative AI ultimately depends on an organization’s ability to innovate.

One thing that’s different: we need to see AI as not only a set of tools we can use but also an entirely new approach to using technology to deliver business results. We have always known there are tasks we can’t practically do with software, and we have worked within those limitations. AI makes a whole new range of applications possible and practical. Take something as simple as recognizing handwritten digits. None of us could have created an algorithm and written software to do so—at least, not well. For machine learning, it’s a relatively easy task. None of us could have written software to create images of cows smoking cigars in the style of Salvador Dali. But Stable Diffusion can do it well.

So AI opens a range of business problems that IT couldn’t solve before. And it involves a completely different model of programming and operating. It requires insight into whether a problem is amenable to AI solutions and whether they are preferable to traditional IT techniques. It requires different skills, and transforming organizations will need to make provisions for developing those skills.

AI shifts the focus of digital transformations toward data. As in the example above, we need to find ways to make data available across the enterprise—busting silos, discouraging data hugging, ensuring data quality, and freeing data from legacy technologies. You can almost think of data playing the role of code in AI. We used to think about agility mostly in terms of code (being able to adjust what we were coding quickly when business needs changed); now we can think in terms of data and whether we have easy access to the data streams we need.

Data requires different governance controls than code. It requires a well-thought-out approach to privacy (a “shift left” in DevOps). It requires controls to ensure responsible behavior and accuracy (code is “accurate” when it passes its test suite; what about data?). We’ll need to make decisions about data gaps, data selection for training, and the tools we use for data management.

As a profession, we’ve gotten somewhat lazy about data. We’ve allowed relational databases to hold all our data—even in cases where we weren’t using it relationally. Today, it makes more sense to use key-value databases, time series, document, or graph databases for certain types of data or just use plain old unstructured data. Data comes in many media—with AI, we can make good use of image data and sound data.

Digital transformation offers the opportunity to correct many of our legacy practices around data management. When moving to the cloud, companies often structure their projects in “workloads.” But perhaps they need to orient their initiatives more around data and its location and characteristics. In addition to refactoring code when moving to the cloud, the rising importance of AI suggests that companies should also consider refactoring data.

We can assume that AI may be built into everything we do and every tool we use. Does that change our plans for digital transformation? Hardly. Transformation is about building nimbleness and responsiveness to change into the organization. But AI reminds us that transformation is not just about code and infrastructure but also about data.