The Bigger Picture For Enterprise Data Infrastructure
Vin Vashishta
AI Advisor | Author “From Data To Profit” | Course Instructor (Data & AI Strategy, Product Management, Leadership)
Businesses operate in a multi-technology environment. The move from digital to AI isn’t like upgrading from an iPhone 11 to an iPhone 15. AI doesn’t replace digital. Businesses can’t go straight from digital to AI, like trading in an old phone for a new one. Each technology wave supports and amplifies the next.
One of the most-seen posts of 2023 on LinkedIn is this meme. It perfectly represents the disconnect between how businesses see transformation and how transformation works. It’s an iterative process, not a generational leap. It is continuous, not one time.
In my book, Continuous Transformation is the first framework I introduce, and right behind it comes the culture change frameworks. Rita McGrath explained the implications of rapid technology change in ‘The End of Competitive Advantage.’ I knew Google didn’t read the book after they said, “We have no moat.” Technology alone isn’t a competitive advantage. AI doesn’t guarantee ROI.
Why Transform?
Each technology wave creates opportunities. Digital can support automating logical, easily defined processes. The cloud enables digital technology to scale. Data, analytics, and machine learning manage complexity and reduce uncertainty, something no technology before it was capable of.
However, without data, there is no AI. Without the cloud, there is no data or AI. Without AI, there is no cloud either. Azure, Google Cloud, AWS, Netflix, Meta, and most other businesses that operate infrastructure at scale leverage AI to manage it. There is too much complexity and uncertainty for digital tools to meet the need.
The data science field must take off the t-shirt and put on a collared shirt to gain traction. Buy-in is necessary across the business to support data engineering, infrastructure purchases, and more advanced data science initiatives. We have come to terms with AI requiring enterprise-wide transformation, but the relationship between CxOs and the data team hasn’t been built to support it. The business culture hasn’t evolved either.
‘Adapt or go out of business’ is the usual line, but I have a different perspective. Adapt to take advantage of the green field of opportunities. Each technology wave and transformation phase delivers new products and improvements. It won’t come all at once.
The frustration on all sides is connected to speed. Everyone sees where the opportunities are and where the business must go. But it isn’t happening fast enough. On one side, we must manage expectations. Transformation will take time, and value will be delivered quarterly. On the other side, we must accelerate progress so the increments happen faster and lead to bigger wins.
The Buy Vs Build Evaluation Has Changed
Third-party tools accelerate data and AI maturity. Data teams want to do everything ourselves, but it’s rarely the right approach at the early stages of maturity. CEOs are under pressure to accelerate the business’s adoption of AI. Buying solutions unlocks early returns and gives the business access to partners who can help with the transformation process.
At Meta, machine learning models are available to everyone in the business on a central platform. Development teams can integrate them into new products. Non-technical users can also leverage models using the same platform. Meta realized that making models more accessible significantly increased their ROI. The same is true with data.
If you don’t work at a company like Meta, building the platform internally isn’t feasible. Taking off the t-shirt and putting on a collared shirt means making business decisions around architecture and initiative selection. Assembling a collection of Indie data science tools and platforms is great for the data team but leaves out the rest of the business. Building it all ourselves takes longer and costs more.
领英推荐
It’s an enterprise-wide problem that requires an enterprise-grade solution. The tools we use need to extend to support non-technical users as well. Otherwise, the data team is stuck doing reporting and dashboards instead of high-value model development and data curation.
Adopting a comprehensive third-party solution delivers the infrastructure and tools to get the business’s data in order. Most come with self-service tools built in. The business needs a platform that can manage data from across the business and make it accessible across the business.
Internal Data Marketplaces For Accessibility
I’m using SAP’s Datasphere and Cloud ERP as examples in this article, but the concepts extend to any enterprise-wide third-party platform. SAP Datasphere implements a marketplace for data accessibility. Originally, it was built to provide access to external third-party data, but businesses need an internal marketplace.
Third-party data is decentralized because that model makes the most sense. Each third party has developed access to data-generating processes that enable them to deliver high-quality data sets. Businesses buy data vs. rebuilding the data sets themselves to save time and money.
The same principles apply to internal data. It doesn’t make sense for each business unit to recreate data sets from across the business or for data engineers to build pipelines from system to system. Planning is forward-looking and prescriptive, which fits with the machine learning paradigm. It requires data from across the business, not just within the single team’s silo, because planning shouldn’t happen in a silo.
Business leaders use the marketplace concept to discover the data they need when they have no idea where it all lives or even what’s available. The marketplace balances centralized and decentralized data. The alternative to a third-party solution is building the data pipelines and marketplace platform internally. As expensive as buying tools is, the cost of building the platform internally is much higher.
Integrating Generative AI
Tools like SAP’s Joule and Microsoft’s Copilot are the fastest way to realize returns from Generative AI. Both are Generative Interfaces with a suite of business applications behind them. It’s a quick way to satisfy the CEO’s call for integrating Generative AI into operations at a low cost. Large infrastructure purchases and upgrades don’t sound like a low-cost option, but compared with the costs of LLMs, they are.
Foundational models have revealed that most businesses won’t have an advanced AI R&D organization like Meta or Google. Just like internally developing the platform doesn’t make much sense, internally building foundational models doesn’t either. Talent, model maintenance, and inference serving costs are too high for the unit economics to work.
Developing large foundational models will be owned by the top tier of tech companies. This isn’t just Generative AI but all the foundational model classes to come. Applied AI researchers won’t be necessary at most companies.
Data engineers and analysts will be the data team’s core players. Data scientists and MLOps engineers are necessary, but they add the most value after the business’s data and infrastructure have been prepared for analytics and machine learning. Buying off the shelf accelerates the maturity progression and takes some stress off the data engineers.
Could businesses do this? Yes. Does the ROI justify doing it? No. Will the business lose competitive advantages by buying vs. building in-house? No, again, the technology isn’t the moat or competitive advantage. What is?
Data is the competitive advantage, especially with planning, but only if it’s formatted for analytics and machine learning use cases and accessible across the business. SAP’s approach is interesting because it balances centralized and decentralized data architectures. The company’s thesis is to build the technology around the data instead of the data around the technology.
No matter what direction the business takes, there’s no straight line to data that doesn’t involve digital and cloud. There’s no path to AI without all three. Transformation is an iterative, incremental process, but there are ways to accelerate parts of the journey. Third-party platforms, foundational models, and partners offer shortcuts at each phase.
Super insightful article and digital transformation with best in class technology is a recipe for value for all businesses.
Dad - Lucky Husband- Commercial Building Automation Specialist- AI Solutions Founder - Developer - Writer
11 个月Agreed. We're in a huge gold rush phase where everyone can offer a use-case, but very few are delivering and building a product.
Data Analytics & Science Mentor | Data Science Consultant @ TMLC | EX - U - Smart AI LAB
11 个月Absolutely Vin Vashishta, digital transformation is multi-dimensional!
Building stuff with generative AI
11 个月Enterprise data is their moat for sure. An internal data marketplace would be great to streamline integration into models.