If you walk into a mid-sized Indian company this year and ask who’s using AI, you’ll probably get five answers from five departments. Marketing has a subscription to one tool, HR is trialling something else, and there’s usually a person in finance who’s been quietly pasting quarterly figures into a free chatbot on their phone. Ask what the company is spending on all this, or what it’s got back, and people start looking at each other.
What they’ve got is a bunch of experiments running side by side. A strategy is what pulls those into one direction, with one set of rules. In this piece I’ll go through why 2026 is a reasonable year to get serious about it, the seven steps I’d use to build the plan, the mistakes I keep seeing, and how you can tell if the plan is doing its job.
Put simply, it’s a written plan for how your company is going to use AI. Where will it help most? What goes first, what waits? What rules do people follow, and how do you get your staff to actually feel okay using it?
People assume a serious plan has to be long. It doesn’t. I’ve seen 200-page reports that nobody opened again after the board meeting, and I’ve seen a three-page plan run a business for a full year without anyone making a fuss about it. The short ones get looked at every quarter, which is probably why they work. They read like a business plan where technology happens to be involved.
Ownership is the thing that decides whether any of it happens. When a CEO or a senior business head is behind the plan, it moves. Get IT, HR and finance into the room from the first meeting as well, because nobody likes getting a finished document dropped on their desk with a note saying “please follow this.”
A couple of years back, you mostly asked AI tools questions and got answers. That’s still possible, but agents can now handle an entire workflow without someone holding their hand: they’ll read an invoice, check it against the purchase order and flag what doesn’t match. So we’re talking about something that sits inside daily operations now, rather than something you show off in a demo room.
Indian languages used to be a real problem too. These days voice and chat tools manage Hindi, Tamil, Telugu and a handful of others well enough to run a service desk or support a field team. I’d have been pretty sceptical of that claim a few years ago.
The legal side matters as well. Most of the DPDP Act’s main duties start in May 2027, and they apply to any AI touching personal data. If you build privacy controls in now, it’s much cheaper than trying to add them after twenty different tools are already in use.
Money is less of a barrier than people think. You don’t need deep pockets for this anymore, or a contract with some big overseas platform. There are open-source models now, and a growing number of Indian AI providers, so a modest budget can get you going. Meanwhile, chances are your competitors are working through a list very much like this one. Each quarter you sit on it, they get a bit further ahead.
Begin with whatever your board already grumbles about at every meeting. It might be collections taking forever, or a service desk that’s too expensive, or one plant that won’t stop having breakdowns. Choose three. Any AI idea that doesn’t help with at least one of them gets parked, even if the vendor’s demo blew everyone away.
Give each goal a number you can check later on, for example how many days it takes to process an invoice, or how many service calls a month are repeat calls. Without a number you can’t really argue about a goal, and you definitely can’t prove you hit it.
This means looking at your data, your systems, your people’s skills and how your company usually handles change. Is the data you need actually reachable? Do your core systems have APIs? Is there anyone on the team who knows AI beyond what they’ve read in the news? If the answers are bad, it’s far better to find out now than six months into a project that’s going nowhere.
Ask every department for ideas, and score each one twice: once for business value, once for how easy it’d be to deliver. The ones that score well on both go to the top. Don’t throw the others away, though, since something that looks hard this year can turn simple next year.
One thing that helps a lot is speaking to the people doing the actual work and not only to their managers. If you ask a clerk where their afternoons go, you’ll often hear about a problem their manager never knew existed.
Work out who approves a new use case, what data can be used, and where that data will be stored. Write a short policy on public AI tools in plain language so people aren’t guessing about what’s safe to paste into a chatbot. Anything where AI decides something about a customer or an employee needs a human to look at it before it goes live. One extra pair of eyes is enough.
For routine stuff like ticketing or document search, an off-the-shelf tool usually does the job fine and there’s no point reinventing it. Building your own starts to make sense when the way you work is a bit unusual, or when the data is the kind you’d be uncomfortable handing over to someone else. In practice most companies land somewhere in between, buying for a few things and building for the rest.
Two things I wouldn’t compromise on, whichever route you take. The data has to remain yours, and you should be able to walk away from a provider without rebuilding everything from zero. Open-source components help here, and so does keeping your data in India. Both make it much harder to get stuck with one vendor.
Pick your best use case and give the pilot a single clear goal, a fixed budget and roughly eight weeks. The way to judge it is simple: are real users using it in production? A nice-looking demo doesn’t count for this. What you learn during the pilot is also what helps you get backing for the next phase.
When the first pilot works, copy its design for the next use case. Train people according to their role and in whatever language they’re comfortable with, and give each team someone who champions it. Every quarter, go back over the whole plan, stop the things that are stuck and put more into the things that are working.
I’d add a skills plan at this point too, so that a few people in each function learn how to spot new use cases and write prompts that actually get good results.
Before you approve any budget, go through this table. If you’re mostly landing in the warning column, fix the basics first, starting with data and ownership.
| Area | Ready signs | Warning signs |
|---|---|---|
| Business goals | Leadership has agreed on three outcomes | Someone says “we should do something with AI” |
| Data | The data you need is digital, current and reachable | Data lives on paper, in email and in personal files |
| Systems | Core systems share data through APIs or exports | Old systems that can’t connect to anything |
| People | Owners are named and a few people are eager to try | No owner, and staff are openly worried about their jobs |
| Governance | A written policy covers data use and public AI tools | People paste company data into free chatbots |
| Budget | Money is set aside for a pilot and what comes next | Money covers licences, nothing for rollout |
Starting with the tool is number one by a long way. A vendor does a demo, a senior person likes it, licences get bought, and then a few weeks later someone asks what problem it was meant to fix.
After that it’s departments each doing their own thing. You pay twice for similar tools, the data ends up in a dozen places, and teams don’t learn from each other at all. Even a small central team setting some standards would stop most of this.
Leaders also tend to forget about the people side. If your staff are anxious about AI, or nobody ever showed them how to use it, a really good system will just sit there unused.
And then there’s waiting for clean data, which can go on forever because no company has perfect data. Prepare what your first use case needs and tidy up the rest along the way. Remember to budget for support after launch as well, because AI keeps needing attention once it’s live.
You’ll want to measure at two levels. Per project, look at time saved, cost per transaction, error rates and how many people really use the thing. Across the company, count the live use cases, add up the value they’re bringing in, and check what share of your staff use AI in a normal week.
The pilot-to-production ratio tells you a lot. When most pilots never go live, the problem is usually how you’re delivering, not the technology. Put all of these numbers on one page and share it with leadership every month.
In a healthy strategy things get faster as you go, so your third use case should reach production quicker and for less money than your first one did. I’d also run a short confidence survey twice a year to see whether people’s comfort with AI is keeping up with how much they’re being asked to use it.
Swaran Soft has been around for over 25 years, building enterprise systems for more than 350 clients along the way. On AI strategy and roadmap projects, our team looks at every use case a client has in mind and scores it on value and feasibility. We help set up the governance rules and then prepare the investment case that leadership needs before saying yes. In one case, it was that roadmap that convinced a board to sign off on ₹2 crore for AI.
After that, we prove the first use case with an 8-week fixed-fee pilot built on an open-source stack, so there’s zero vendor lock-in. If your sensitive data has to stay in India, our Private AI Lab setup keeps it on your own infrastructure, and our Cybersecurity & Compliance team checks every use case against DPDP rules from the first day.
None of what works here is complicated. Plans that succeed tend to be short, practical, and built around problems the business is already losing sleep over. Settle on three outcomes, take an honest look at how ready you are, and sort your use cases by how much value they bring and how easy they’ll be. Get the governance rules agreed early. Doing it after something has already gone wrong is a lot more painful.
After that, it’s about proof. Run a pilot with a fixed scope, make sure it actually reaches production, and grow from whatever works. Look at the plan every quarter and keep people involved all the way through. I don’t think the winners in 2026 will be the companies with the biggest pile of AI tools. They’ll be the ones who were clear about what they wanted those tools to do in the first place.
Book your free 45-minute AI Pilot Assessment. Email info@swaransoft.com or call +91 9220-313-650 (India) | +971-50-9292-650 (UAE). Visit www.swaransoft.com
AI Architect and Entrepreneur building India's Edge AI ecosystem. 25+ years in enterprise technology. Founder of Swaran Soft, Gignaati, and Copilots.in.