Executive Summary
For the last two years, most of us have been experimenting with Copilots and chat interfaces that look great in a demo but don’t actually impact P&L. The problem isn't that the models are getting "smarter", it’s that our enterprise foundations are messy. We’re seeing a massive gap between companies that are just "doing AI" and those that are actually leveraging AI to drive consistent, measurable value.
1. Why most AI initiatives stall
We’re currently seeing a lot of activity but very little measurable outcomes. If you look under the hood of most enterprise AI programs, you’ll find the same three challenges:
Pilots are easy to spin up in a vacuum, but they die the moment they need to scale.
A standalone chatbot is fun until it needs to talk to a legacy database that hasn't been touched in a decade.
Most "Copilots" generate a bit of curiosity for a month, then the usage charts fall off because they don’t solve a core workflow problem.
2. What Enterprise Leaders Care About
The leadership team doesn’t care about how many parameters a model has. What they care about is whether it can actually integrate with the existing technology stack without compromising security and how quickly it will pay for itself.
How quickly are people enabled to take the right decisions is where the value lies. For example, if your team can get to the right answer in two minutes instead of two hours, that has a huge positive impact – both on the P&L as well as the culture. The challenge is none of this happens unless the underlying systems are connected properly.
3. Where the real wins are happening today
The enterprise architecture determines how AI is leveraged. The ROI is currently concentrated in areas where workflows are repeatable and data is accessible. Here are some examples:
Customer Support: Summarizing massive case histories, so a human doesn't have to.
Operations: Automating the "paperwork" of HR and Finance that usually drags productivity down.
Sales Ops: Getting a proposal out in hours and not days.
These are not “AI-first” initiatives. They are workflow problems where AI happens to be the best tool for the job given the current architecture.
4. The Real Constraint: Enterprise Architecture
This is the part that has been ignored because it’s hard. The real bottleneck for AI isn't the model; it’s the architecture. Most enterprises have built AI at the "interaction layer" (the chat box). But they haven't built it at the execution layer. If your AI can suggest a solution but is unable to execute it across your systems, you’re only halfway there.
5. Our Approach at Mirketa
We’ve spent a lot of time thinking about how to bridge this gap. We don't believe in "black-box" solutions that you just plug in. Instead, we’ve been building accelerators; specifically, around agent orchestration and MCP-based data layers. Our philosophy is pretty simple:
Keep it in your house: AI should live in your environment, not ours.
Connect everything: If it doesn't work with your legacy systems and your SaaS stack, it's not a solution.
Security isn't an afterthought: It has to be baked into the architecture from day one.
6. Path Forward
If you’re tired of "experimenting" and want to see actual ROI, here is my advice:
Address Architecture Early: Orchestration and data access should not be afterthoughts.
Stop building from scratch. Use standardized frameworks (like MCP) to avoid reinventing the wheel.
Embed AI Into Existing Systems to ensure adoption. If it’s not embedded in the tools your team already uses, they won’t use it.
Measure productivity, and decision speed—not just usage stats.
Bottom Line
The gap between enterprises that successfully operationalize AI and those that remain in experimentation mode will widen over the next 6-12 months. The winners won't be the ones with the flashiest demos but the ones who fix the architecture gap.
About the Author
Ajay Jalali is the VP Delivery and Operations at Mirketa, a global IT consulting and services firm specializing in Salesforce, Oracle, enterprise integration, security monitoring, Site Reliability Engineering (SRE), data engineering, and AI enablement.
Ajay spends his days helping enterprises stop "playing" with AI and start embedding it into the core of their business. Whether it’s Salesforce, data engineering, or AI orchestration, his focus is always on one thing: making sure the technology works for the business.
Top comments (0)