Why agent sprawl is the wrong problem to solve — and what retail leaders should focus on instead
Here’s a stat that should keep retail executives up at night: the average enterprise now runs 12 AI agents, and that number is expected to hit 20 by 2027. But here’s the kicker — only 27% of the applications those agents rely on are actually connected to each other.
That’s according to the 2026 Salesforce Connectivity Benchmark Report, which surveyed over 1,000 IT leaders. And the headline finding? More than four in five believe this proliferation of AI agents will create more complexity than value.
If you’re a retail CEO, COO, or CTO reading that and feeling a knot in your stomach, you’re not alone. But here’s the thing: agent sprawl isn’t the real problem. It’s a symptom. The real issue is what’s underneath — or more precisely, what isn’t.
What Is Agent Sprawl, Really?
AI agents are software programs that can act autonomously — think of them as digital team members that monitor inventory levels, route orders, flag supply chain disruptions, or answer customer questions without someone clicking a button. Unlike a traditional dashboard that waits for you to look at it, an agent notices that SKU-4421 is running low in your Dallas warehouse and triggers a reorder before you’ve finished your morning coffee.
The promise is enormous. But the reality CIO Dive recently highlighted is that most enterprises are deploying these agents in silos. One agent watches inventory. Another handles pricing. A third manages vendor communications. Each pulls from different data sources, operates on different logic, and reports to different teams.
Gartner’s Tom Coshow calls this “agent sprawl,” and it’s fast becoming a top concern for CIOs. The average organization manages 957 applications. When you layer dozens of autonomous agents on top of that fragmented foundation, you don’t get efficiency — you get chaos with a more sophisticated user interface.
The Data Layer Problem Retail Can’t Ignore
Here’s where this gets specific to retail. Your average mid-to-large retailer runs separate systems for e-commerce, in-store POS, warehouse management, order management, ERP, vendor portals, and marketing automation. Each system has its own data format, its own definition of what a “product” or “order” is, and its own update cadence.
Now imagine deploying an AI agent to optimize your fulfillment. It needs to know real-time inventory across every channel, down to SKU level. It needs to understand which warehouse is closest to the customer, what the shipping costs look like, and whether the item is also available for in-store pickup. That data lives in five different systems that weren’t designed to talk to each other.
This is why 96% of IT leaders in the Salesforce study said agentic AI’s long-term effectiveness hinges on data integration. Not better models. Not more agents. Data integration.
The numbers back this up. The inventory disconnection problem alone costs US and Canadian retailers an estimated $350 billion annually — in stockouts, overstocks, markdowns, and write-offs. That’s not a technology problem. It’s a data architecture problem.
Rethinking the Foundation Before Adding More Agents
The instinct right now is to solve agent sprawl with more management tools — agent observability platforms, governance frameworks, management dashboards. Gartner is already calling for “agent management platforms” as a new software category.
That’s not wrong, but it’s treating the symptom. If your agents are pulling from disconnected, inconsistent data sources, adding a layer that monitors them more closely doesn’t fix the underlying issue. You’re just watching bad decisions happen faster.
The more fundamental question is: what does the data layer look like before you deploy agents?
This is the approach we’ve taken at Ekyam. Rather than starting with agents and working backward, we started with the data. Our Retail Knowledge Graph unifies product, inventory, order, vendor, and location data into a single, connected model — a shared vocabulary that every system (and every agent) can understand. When an AI agent built on this foundation asks “what’s the stock level of SKU-4421 across all channels?” it gets one consistent answer, in real time, not five conflicting ones.
It’s the difference between building a house on bedrock versus building it on sand and then hiring a structural engineer to figure out why the walls keep cracking.
What This Means for Retail Leaders Right Now
MIT research has shown that the biggest AI ROI in retail comes from back-office automation — not the chatbots and personalization engines that get the most press. The unsexy work of connecting systems, standardizing data, and giving AI agents a clean foundation to operate on is where the real margin gains live.
Generative AI is projected to unlock between $240 billion and $390 billion in economic value for retailers. But here’s the uncomfortable truth: enterprises will spend $2.5 trillion on AI this year, and Forrester estimates a quarter of that spend may get deferred because companies can’t demonstrate ROI. Only 15% of organizations currently report a positive impact on earnings from AI.
The gap between potential and results isn’t an AI problem. It’s a data readiness problem. And for retail specifically, it’s a data integration problem.
So before your next board meeting features a slide about how many AI agents you’re deploying, ask the harder question: are those agents working from the same playbook? Do they share a single source of truth? Or are they just making disconnected decisions faster?
The retailers who get the data foundation right won’t need to worry about agent sprawl. Their agents will be working together by design — because they’re built on a unified understanding of the business.
That’s the real competitive advantage. Not more agents. Better data.
