The hidden data crisis threatening your AI transformation plans

romaset/Getty Images

Historically, when defining a network architecture, it's been fairly easy to create buckets or silos of data. Whether the data was gathered from one system here and one system there, or from one department here and one department there, data almost naturally bucketized.

When one of IT's biggest challenges was storing that data, buckets or silos seemed good enough. But now that capacity is far less of an issue. Our biggest concerns are understanding and insight. Chunks of data, locked in their silos, prevent an holistic and fully integrated level of understanding.

Also: OpenAI wants ChatGPT to be your 'super assistant' - what that means

Add AI into the mix, and access to data becomes even more critical. For the AI to make logical leaps from one concept to another, it needs to see corporate data as a whole.

According to Rahul Auradkar, EVP & GM, Data Cloud at Salesforce, "Most enterprise data (80%) is unstructured, and 59% of organizations lack unified, easily accessible data for AI. This is a significant hindrance when looking to effectively scale."

It's also a problem Salesforce has been trying to solve with its Data Cloud offerings. A Mulesoft Benchmark Report, summarized in a blog post, suggested the average enterprise now manages 897 applications, of which only 29% are integrated.

Think about it. How many cloud services do you personally subscribe to? How many does your business use? And how many local applications? Can you get at all that data?

Traditional customer data platforms

One way businesses have attempted to solve the problem of silos is through the deployment of customer data platforms (CDPs). CDPs expand on the classic customer relationship management (CRM) environments by attempting to unify and aggregate customer data beyond what's just inside the CRM. CRMs are often call-and-sales-tracking tools, while CDPs aim to create a more fleshed-out customer profile that can later be used to foster engagement.

Also: 30% of Americans are now active AI users, says new ComScore data

But CDPs have been falling short. As the acronym says, the focus is on the customer, which means the tool has been the purview of sales and marketing organizations, often locking that data behind virtual department doors. They have not only not been helping enterprise-wide data unification, but actually impeding it.

Salesforce's Auradkar said: "Traditional CDPs help get primarily marketing data in order, but don't harmonize and unify it across the entire organization." He also said that data is often very fragmented. This lack of cross-functionality has consequences for solutions for business intelligence and automation.

He told ZDNET: "There were lots of solutions for connecting data, but no real solution focused on data actionability." And don't even mention AI. These siloed systems often squelch any potential for deep AI analysis across the enterprise.

The need for unification, especially for AI

"A complete AI system is cohesive and scalable," said Auradkar. The system has to integrate "Data, AI, and action to deliver concrete business outcomes."

As we've come to know, modern large language models rely on a form of unsupervised pre-training. They often use multi-modal, free-form data (text, audio, video, sound), not just the contents of rows and columns assigned to fields. "Unstructured data," said Salesforce, "is critical for providing a 360-view of the customer… but only 18% of organizations can take advantage of it."

Also: ChatGPT can record, transcribe, and analyze your meetings now

For the AIs to be able to train on the data, even if it's just a giant pile of documents in folders, they must be able to ingest and process data. That requirement means they have to find data. That's where Salesforce's Data Cloud comes in. "AI relies on quality data," says Auradkar. "Data Cloud provides the unified foundation for intelligent agents, enabling informed actions without costly 'rip and replace' approaches."

The zero-copy strategy for mitigating data movement challenges

Of course, now that you've found all those virtual data piles, you have to somehow move them pile by pile, silo by silo, into some giant storage facility. Or do you? Salesforce believes data integration across existing lakes, warehouses, and apps is possible, without duplication or data movement. This approach is called zero copy, and can reduce cost, complexity, and delay.

Also: Tech prophet Mary Meeker just dropped a massive report on AI trends - here's your TL;DR

Ironically, one of the proponents of eliminating data movement is a giant of industry that physically moves more than 15 million packages per day to more than 220 countries and territories: FedEx. According to Carlos Gonzalez, FedEx senior programmer advisor: "Zero copy is very attractive to us because it's easier and less expensive than ingesting the data again and landing it in multiple spots."

FedEx uses Data Cloud for this work. Salesforce describes Data Cloud as an intelligent library for business and customer data. This library enables enterprises to work with unified data across every department. The idea is that the approach bridges data silos and harmonizes information from data lakes, warehouses, business applications, and both structured and unstructured formats, using low-code tools.

But does it really work?

The concept aims to enable an AI-ready data platform where existing data can be accessed and where companies can also ingest new data in real-time, perform cross-format harmonization, and let AI agents operate without undue friction.

Also: How Salesforce's 5-level framework for AI agents finally cuts through the hype

Salesforce's Auradkar told ZDNET: "With integrated analytics and action layers (Data Cloud is an action layer), insights are no longer passive but serve as immediate calls to action, whether through automation, analytics or real-time suggestions," fed into Salesforce's AI agent offerings.

Salesforce provided ZDNET with customer experience stories.

Adobe Population Health (a health practitioner, not the Adobe of PDF and Photoshop fame) reported to Salesforce that "Automated patient data retrieval, using Data Cloud and Agentforce, saved nurses 375 hours weekly and reduced chart summarization time by 75%."

Auradkar told ZDNET by email: "BMO wanted to unify data to drive more leads with AI-driven personalization. They adopted Data Cloud to unify customer data. With a more targeted approach that combined efforts between retail banking and wealth management, they achieved a 5x increase in leads, doubled email open rates, and reduced time to set up additional marketing campaigns from 3-4 days to four hours."

Getting from here to there

This integration of data is important, not only for large enterprises but also for small businesses. I just spent the last three hours stubbornly trying to get QuickBooks to tell me how many cloud services my small business subscribes to so I could mention it earlier in this article. But that data is locked behind a set of QuickBooks fields that can't be indexed on their own.

I suspect I might be able to dump the entire transaction history for a year, and ask ChatGPT to dig into it, but otherwise, that's data that I should have at my fingertips that's locked in a silo.

Unfortunately, I can't just bolt on Data Cloud and have it generate a report on my cloud spending. AI modernization will require a clear strategy, cross-team collaboration, and, yes, data access. This modernization isn't just signing up for yet another service. It will require a rethinking of infrastructure.

Also: How much energy does AI really use? The answer is surprising - and a little complicated

Auradkar summed up the situation well. He said businesses "Need a unified data strategy with clear objectives and outcomes. Successful AI will not be possible without a cohesive, comprehensive view of customer and business information."

Yeah. So true.

What about your organization? Are you struggling with siloed data or outdated tools that limit your ability to use AI effectively? Have you tried integrating data across departments, or have you explored real-time data platforms? What kind of impact do you think unified data could have on your business workflows or customer engagement? Let us know in the comments below.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.