Everyone wants AI to work. May I dare say that people sometimes think of AI as a magic button that you push to get everything to work? But no one wants to get to the grassroots to fix what’s needed to make AI work.
Three years into the GenAI era, most enterprises are already “using AI” in some form. Agents are live. Copilots are piloted. Automations exist in at least one team.
And yet, many leaders I talk to today seem to wonder the same question: Why isn’t AI actually delivering the impact we expected?
The MIT report exposes an uncomfortable truth: 95% of enterprise GenAI initiatives fail to produce measurable business impact. Not because the models are weak. Not because the ideas are flawed. But because AI is being deployed into environments that were never designed to support it. And the story isn’t any different in fast-growing regions like Singapore.
What looks like an AI problem is almost always a data problem.
To understand why AI breaks and what it actually takes to make it work, we need to start with the root cause that enterprises consistently underestimate.
Why do data silos happen when your organization grows?
As companies grow, speed matters more than alignment. Teams adopt new tools to move faster. Over time, integrations become brittle, data models drift, and information spreads across systems that were never designed to work as one.
As Christian Klein, the CEO of SAP, puts it:
“Every company you walk into has their data siloes: there have been trends with collecting data and creating data lakes, but no one has solved the problem of making all the data match. And when it doesn't fit, AI can't do magic immediately to say, "I 100% understand how this data fits together, and I can correlate it to produce good results for the company."
On top of this, two additional forces reinforce silos: organizational culture and regulatory compliance pressure.
- Organizational culture: With teams beginning to treat “their” data as territory, access is limited, ownership is unclear, and sharing becomes slow or political rather than automatic
- Regulatory compliance pressure: The wall gets higher and tighter. Data gets locked behind roles, regions, and policies.
The result isn’t a lack of data. It’s too much data: duplicated, inconsistent, and disconnected, giving leaders the false confidence to say, ‘We already have the data. ’
But what they really have is data that:
- Can’t be easily connected
- Doesn’t stay in sync
- Can’t be queried as one system
- Loses context the moment it moves between teams
Humans compensate by jumping between tools and context stitching, but AI can’t, and that’s where the real problems begin.
How does siloed data break AI in the ways teams don’t see
Although data silos aren’t new in the age of AI, their cost compounds exponentially.
AI doesn’t just need data; it needs context, completeness, and connectedness. When information is locked across CRMs, ticketing systems, engineering trackers, spreadsheets, emails, and conversations, AI is forced to work with partial truth.
Gartner predicts that through 2026, 60% of AI projects will be abandoned, not because the models fail, but because the data behind them isn’t AI-ready.
And, that’s exactly what the MIT report reinforced: most enterprises are piloting AI without fixing the data foundation. As a result, projects appear promising in demos but falter when asked to deliver sustained, measurable ROI.
Three critical and often invisible ways teams miss seeing how it affects AI
1. Fragmented context: AI sees events, not the story
AI fundamentally lacks the ability to connect signals across support, product, sales, finance, or customer journeys when your data lives in different systems. It doesn’t see a narrative; it sees isolated moments like:
- A new support ticket, but it misses the product change that triggered it
- A customer complaint, but can’t link it to a sales promise or cost implication
- A rising trend, but not the series of events that caused it.
That’s why frontline teams still juggle five tabs just to resolve a simple issue. Productivity drops. AI suggestions feel shallow. Decisions rely on tribal knowledge instead of system intelligence.
Consider a billing error reported by a customer. The support team sees it as a ticket, finance sees the adjustment, product is informed about a pricing update, and sales has renewal notes. However, AI sees none of this information together. It recommends generic steps while the real root cause remains hidden.
2. Inconsistent truth: AI receives conflicting signals
As organizations scale, data naturally fragments. Teams have their own set of data, success metrics, severity levels, and ownership, and store that information in systems that evolve independently. What should be a single source of truth becomes multiple versions of reality: incomplete, inconsistent, and difficult to reconcile.
I’m sure you have heard the phrase, “Garbage In, Garbage Out”. When AI is trained on such conflicting inputs, it reflects the inconsistency it has been given. Over time, these inconsistencies compound:
- Product priorities drift
- Fixes are delayed
- Customer experience degrades
- AI never reaches its full potential
The problem isn’t the algorithm. It’s the fractured reality it’s learning from.
3. Scaling friction: AI works in one team, fails across the organization
AI often succeeds in isolated pilots, inside Support, Product, or Operations. But without shared data models, those successes don’t scale.
What works in one function becomes brittle the moment it needs to interact with another. As the organization grows, AI should evolve in tandem. In siloed environments, scaling becomes a liability.
And when AI can’t scale, here’s what actually happens:
- AI fails to deliver consistent outcomes
- Teams lose trust in AI recommendations
- Adoption drops, and AI usage becomes extra work
- The business never sees meaningful ROI.
This makes AI without data unification like assembling a puzzle with half the pieces missing.
Fix the foundation first: what the 5% of AI winners do differently
The companies that scale AI successfully aren’t the ones with the biggest budgets or the most powerful models. They’re the ones who fix their data foundation first.
Rob McAveney, CTO at Aras, agrees: “Many organizations assume they need AI when the real starting point should be defining the decision you want AI to support, and making sure you have the right data behind it.”
That shift in thinking is what separates AI experimentation from real business impact.
This is exactly where the 5% of AI winners diverge. From day one, they focus on a few fundamentals and execute them consistently, like:
- Modernize and connect their data layer across teams and tools.
- Audit their “data reality,” not just what’s documented.
- Create a culture of shared ownership and collaboration.
- Build scalable infrastructure that prevents new silos from forming.
They understand a simple truth: AI success is an architectural decision, made long before the first model is deployed.
Where AI finally works: the foundation DevRev was built on
Most businesses rush into AI experiments with copilots or chatbots sitting atop disjointed systems of record. But without clean, unified, permission-aware data, those agents act blindly. They can answer surface-level questions but can’t reason, prioritize, or take contextual action.
Solving that problem requires more than adding another AI layer. It requires fixing how data, context, and ownership are modelled across teams in the first place. That’s the foundation Computer, by DevRev, is built on, long before AI became a boardroom priority, with the mission: AI will only be as powerful as the data foundation beneath it.
Before breaking down the pieces, see how this foundation for Computer, by DevRev, comes together end to end.
1. AirSync: the end of data silos
As teams work on multiple tools by default, it is normal for data to be created and updated across systems of record: CRM, issue trackers, support tools, collaboration apps, and internal tools.
Computer AirSync, DevRev’s patented technology, bidirectionally connects these systems and keeps data unified and consistent. Changes made in one place are reflected everywhere, instead of drifting into silos.
It also securely accesses enterprise data that’s private, permissioned, and often locked inside legacy systems, keeping teams and AI aligned to a single, current source of truth.
2. Computer Memory: the context layer AI has never had
Data becomes intelligence only when it’s connected.
Computer Memory, DevRev’s patented knowledge graph, links customers, conversations, tickets, product issues, engineering tasks, and documentation to create a business ontology that defines how your business runs.
So when AI responds, it isn’t retrieving text. Its reasoning is based on the context of relationships, urgency, ownership, and impact.
3. Foundational service: AI that works as a true teammate
AI can only be as dependable as the foundation it stands on. By unifying data and context, DevRev’s Computer, your AI teammate, acts as a reasoning engine and dependable teammate, not just a chatbot. It:
- Searches across the organization with complete context.
- Performs root-cause analysis to help decision-makers act faster.
- Recommends next steps and creates tickets directly where needed.
- Automates L1/L2 tasks while escalating complex cases to human experts.
- Operates consistently across every surface - support consoles, apps, product UIs, and internal tools.
As organisations grow, their ability to unify data and context will define how well AI systems perform. The hype around AI will continue. But if there’s one message the MIT GenAI Divide report makes clear, it’s this: AI fails not from lack of intelligence, but from lack of context for understanding.
If you’re considering what that foundation should look like in your own organization, it’s worth seeing how Computer, by DevRev, works in practice.
