AI's real ROI will go to companies willing to refound themselves around it, not just bolt it on. That takes real work, and real trade-offs in systems, people and habits. In this edition, we look at the AI ROI question you should really be asking, and why the answer starts with how your systems are wired.
Most CXOs and business leaders have the wrong timeline for ROI from their AI investments. There are two fundamental pieces to getting it right.
First, the real returns will only start showing after 3 to 5 years of consistent investment.
Second, they won't come to those who simply spend more on AI. They'll come to those who connect their systems, business context and tacit knowledge holistically, one piece at a time.
If your board keeps asking, "What's the AI ROI today?", that's the wrong question. The question to ask is: are we starting to see the first green shoots of AI working in a connected system, at least in pockets? And what can we do to speed this up?
Here's the journey most companies will need to go through to get there, one stage at a time.
1. Today's gains are small, and they sit in pockets
They come from a few sharp people, one team, maybe one function.
Nearly half of organisations in India have multiple AI use cases live. Only 1 in 10 has scaled them across the business (EY–CII, 2025).
That's not failure. It's stage one, and almost everyone is in it.
2. The blocker isn't AI. It's how your business is wired.
Legacy systems don't talk to each other. Each function guards its own data and its own slice of the process, and nothing joins them end to end.
That's why 78% of organisations in the same EY–CII survey say integrating AI with their existing systems is their biggest hurdle.
Put AI on one slice of an unchanged process and you get the same disconnected process, running faster.
3. Data without context is just noise
AI needs to know how you actually run the business: how decisions get approved, each customer's history with you, and the know-how that today lives only in a few senior and tenured people's heads or systems.
Salesforce didn't start with Agentforce, its AI agents. First it built Data Cloud, one layer that pulls together the data from its sales, service and marketing apps, connected to outside systems through MuleSoft, its integration platform. Agents came two years later, on top of that layer.
4. Don't try to connect the whole company in one go
That's how you end up with a big-bang IT project that gets killed once the excitement wears off.
Connect what you can, where the gain is visible. Two systems and a few customer journeys is a good start. What matters is staying the course.
Show the win, and let that win fund the next connection.
What to do as a founder or org leader
Keep funding AI now, with budgets reallocated from other activities. Don't hit pause.
Back the pockets that already work, then spread them deliberately.
Connect a few journeys end to end, and prove their efficiency.
Use each win to make the case for connecting the next journey.
The companies that win won't be the ones that spent the most. They'll be the ones that kept connecting.
🔥 Worth your time
AWS and Anthropic now admit AI doesn't work without people on site while they race to deploy forward-deployed engineers.
Sridhar Vembu, Zoho, shared that server prices are up 4x and AI usage costs are rising. Zoho is skipping appraisals this year.
What happens when companies rush to adopt AI, only to find it’s creating more work than it saves?
a16z, on what if long running agents could own entire business outcomes, not just tasks.
Ramp AI Index suggests how much companies are spending on AI, and what models they are spending on.
🗒️ Before you go
Got a question about AI in your organisation, or a topic you’d like us to dig into? Just hit reply. We’ll either cover it in an upcoming edition (without naming you or your company) or write back to you directly.
Have a great week, and see you next Saturday.
- PD & Sonali