A practitioner's confession from inside a traditional Indian company

TL;DR

  • Most enterprise AI initiatives fail because they start with tools instead of architecture
  • Document-level AI (chatbots, assistants) is aspirin — the disease is structural disconnection between departments
  • The fix: map your business lifecycle, build a unified data layer, co-build with domain experts, and protect your team's focus
  • An intelligence department reports to the MD and measures outcomes — an AI lab reports to IT and measures demos

I've spent the last year and a half building an AI function inside a real estate and logistics company. Not a tech startup. A traditional Indian company with Excel workflows, WhatsApp decision-making, and departments that have operated independently since before AI was a buzzword.

Here's what I've learned: the technology is the easy part. The organisation is where transformations go to die.

Every company I see announcing their "AI strategy" is making the same mistake we almost made. They're starting with tools when they should be starting with architecture. They're hiring data scientists when they should be redesigning decision flows. They're buying software when they should be building systems.

This is the story of how we caught ourselves — and why what we're building now looks nothing like what we originally planned.

We built the wrong thing first

When we started, our approach was intuitive and wrong.

We took SharePoint documents, fed them into AI, and helped departments retrieve information faster. It felt productive. People were impressed. Leadership nodded approvingly. We processed over a thousand files across eight departments. We cut workshop time by seventy-five percent.

And none of it mattered.

Here's what we missed: making individual tasks faster doesn't make the organisation smarter. You can give every department an AI assistant, and you'll still have the same disconnected decision-making, the same data silos, the same leadership teams assembling board decks manually from five different sources that were already stale by the time anyone read them.

Document-level AI is aspirin. It numbs the symptom. The disease is structural.

We were numbing pain when the company needed surgery.

The question that broke the whole approach

The breakthrough came in a single meeting. Our Managing Director asked three questions back to back: What about the data lake? What about BIM? What about lead generation?

Three different domains. And all three pointing at the same gap — there was no intelligence layer connecting the lifecycle of our assets.

That forced us back to first principles. What does this company actually do?

We acquire an asset. We construct it. We lease or sell it. We operate it. Sometimes we sell it again.

Four stages. The entire business. And every one of them was running on its own island — different systems, different data, different teams who rarely talked to each other about anything beyond the immediate handoff.

Once you see that, the AI question changes completely. It's not "which department gets an AI tool next?" It's "what's the intelligence layer that connects the whole lifecycle?"

So we stopped building an AI lab and started building an Asset Intelligence Department. Not a support function. Not an innovation team that demos cool things once a quarter. A department with four verticals mapped directly to the business: Investment Intelligence, Leasing Intelligence, Construction Intelligence, and Operating Intelligence.

Every vertical tied to a business outcome. No technology for its own sake.

The architecture that actually matters

When consultants sell AI transformation, they talk about algorithms, models, and platforms. When you're actually building it inside a company, you end up talking about something far less glamorous: who owns what, who talks to whom, and where the data actually lives.

The data backbone comes before everything. We had SAP running but disconnected from everything else. A PMO dashboard built by a Big Four firm that required manual data entry. BIM and digital twin capabilities on the roadmap that would launch as yet another silo. The first thing we built wasn't an AI model — it was a data lake. One unified store that feeds all four verticals. Without this, you're building intelligence on quicksand.

This is the part nobody wants to fund. A data lake doesn't demo well. You can't put it in a board presentation with a before-and-after screenshot. But it's the difference between four separate AI projects and one intelligence system. The projects impress people in meetings. The system changes how the company operates.

Every initiative needs a business owner who isn't you. This is where most AI teams quietly fail. They build something brilliant in isolation, then wonder why nobody uses it six months later. We embedded alongside business teams from day one — co-building with domain experts through intensive workshops, not building for them. The goal isn't to impress anyone with what AI can do. It's to make people dramatically better at what they already do, then hand off completely.

Boundaries matter as much as ambition. We explicitly defined what we would not do. We don't build solutions for assets outside our core mandate. We don't execute marketing campaigns. We don't take ad-hoc requests that derail strategic priorities.

This sounds obvious. In practice, it's the hardest discipline in the job. In traditional companies, the AI team becomes everyone's Swiss Army knife. "Can you quickly make a presentation?" "Can you help with this Excel?" "Can you build a chatbot for HR?" Every request is reasonable on its own. Collectively, they destroy your ability to deliver anything transformative.

The pattern I keep seeing

I've now sat in enough boardrooms and watched enough transformation efforts stall to recognise the same failure mode repeating.

Companies confuse adoption with transformation. Getting people to use ChatGPT is adoption. Changing how the company makes investment decisions using intelligence that was previously invisible — that's transformation. One is a training problem. The other is a systems redesign. Most companies declare victory at adoption and wonder why nothing actually changed.

They hire for the wrong skill. The scarcest capability in AI transformation isn't machine learning. It's translation — the ability to sit with a business leader, understand their actual workflow (not the one on the process document, the real one), and design an intelligence layer that makes their existing expertise more powerful. Most data scientists can't do this. Most consultants won't. It requires someone who speaks both languages and has the patience to iterate through the messy middle. This is exactly why I threw out traditional interviews entirely — resumes can't show you this.

They sprint when they should be thinking. When leadership says "look at BIM, look at the data lake, look at lead generation," the instinct is to attack all three simultaneously. The discipline is to step back and ask: what connects these? In our case, the asset lifecycle. Once you see that, the three requests aren't separate projects — they're different views of the same system. That reframe saved us from building three tools that would never talk to each other.

They ignore compounding costs because they're invisible. The cost of building an intelligence function shows up on a budget slide. The cost of not building one compounds in silence. Every month without BIM integration is rework accumulating undetected. Every quarter without market intelligence is tenants you lost to competitors who saw the opportunity first. Every year your data stays in silos, your leadership decisions rest on manually assembled reports that were already outdated when they hit the table.

The uncomfortable truth

Here's something nobody writing LinkedIn posts about AI transformation wants to say: most AI initiatives inside traditional companies are performative.

They exist to signal modernity. They produce demos, not systems. They report to IT instead of the business. They're staffed with talented people who spend eighty percent of their time fighting for access to data and twenty percent actually building anything.

The difference between an AI lab and an intelligence department isn't budget or headcount. It's mandate and reporting structure.

An AI lab reports to IT and asks permission to innovate. An intelligence department reports to the Managing Director and is expected to transform.

An AI lab builds proofs of concept and hopes for adoption. An intelligence department co-builds with the business and measures success by outcomes.

An AI lab has projects. An intelligence department has verticals tied to revenue.

If your AI function doesn't have direct access to leadership, a clear mandate that protects its focus, and business owners embedded in every initiative — you don't have a transformation. You have a hobby.

Where the gap shows up

The companies that build intelligence into their asset lifecycle now won't look different from their competitors tomorrow. The gap takes time. And at the industry level, the same logic applies — the AI companies that survive won't be the smartest, they'll be the ones that own their stack.

It shows up in three years, when one company makes investment decisions in days because their data lake surfaces opportunities automatically — while another is still waiting for an analyst to compile a market report. When one company's leasing team walks into a pitch with personalised competitive intelligence — while another brings a generic brochure. When one company catches a construction overrun in real-time through BIM — while another discovers it six months later in an audit.

This isn't about AI. It's about whether your organisation can learn faster than your competitors.

The technology makes the learning possible. The architecture makes it permanent. The discipline makes it real.

And the discipline is the part almost everyone skips.

For the macro view of where the AI industry is headed — and why infrastructure ownership matters at every level — see The AI Shakeout.