Meet Rakesh. He runs a 60-person distribution firm in Indore. Last month his biggest rival started taking orders through a WhatsApp bot, and since then his sales head has forwarded him six AI proposals, each glossier than the one before.
The first quoted ₹40 lakh for something called a transformation programme. Another promised a working bot in two weeks, practically free. Rakesh built this business up from a single godown and he isn't easily fooled, but he had no idea who to believe.
If that sounds a bit like your week, keep reading. Picking an AI consultant, I've come to think, has more in common with taking on a senior manager than with buying a bit of software. Below I'll go through how I'd weigh up the options, how to catch the bluffers before they get far, and how to sign without lying awake that night.
Take the jargon out and it's not complicated. A decent partner looks at your business, figures out which chores software can take off your plate these days, builds something that does them, and then hangs around until your staff are using it without moaning. That last part is where most of them vanish.
For a company your size the useful projects tend to be dull. Answering the fiftieth where's-my-parcel message of the day. Reading supplier bills. Chasing customers who are 60 days overdue.
None of this needs a PhD. It needs someone who can connect AI to Tally, your CRM or that Excel file your accountant won't let anyone else touch.
You'll meet three kinds of firm. Strategy outfits hand you a handsome roadmap and no software. Then there are the product sellers, who've only got the one hammer, so funnily enough every problem you mention starts to look like a nail. What most small businesses really need is the build-and-run type, one team that designs it, builds it and looks after it afterwards.
A large corporate can bury a failed AI project in next year's budget and carry on. You can't. Your cash is tighter, and the three people who'd sit in vendor meetings are probably the same three keeping the business running.
We've watched bad partners leave the same trail again and again: a tool nobody opened after week five, a subscription that needs 90 days' notice to cancel, and customer data sitting on a server in a country nobody can name. Nothing dramatic. It just drains you slowly.
The good ones behave almost the other way round. They'll suggest starting with something small, put a working version in front of you within a few weeks, and when the job's done they hand everything over without making it awkward.
Before you ring anyone, sit down with a cup of chai and list three jobs that eat your team's hours. Be specific, almost annoyingly so. "Pooja and Imran spend most mornings answering delivery queries" is a brief a consultant can work with. "We want AI" is a brief for a sales pitch.
Then watch the first meeting closely. The good ones lean in and ask about Pooja and Imran. The rest open a laptop and start a demo.
Anyone can knock out a case study PDF in an afternoon. Ask instead to see a system that's live at a client today, how long it has been running and what changed after it went in.
I'd put track record ahead of nice branding every time. AI projects rarely fall apart because of the AI itself. They fall apart at the joins, connecting to your other systems, keeping things secure, sorting out problems after launch, and firms that have been shipping software for years have already been burned by all of that. If you can manage it, ask for ten minutes on the phone with one of their existing clients. One business owner will tell another things a salesperson never would.
However warm the pitch, I'd steer clear of the big contract at the start. What you want is a pilot: one narrow job, a fixed fee, and a proper end date somewhere between 6 and 10 weeks out. By the end of it your own people should be using the thing at work, which is very different from watching a prototype run on the consultant's MacBook.
A proper pilot proposal doesn't need more than two pages. It should say which task, which people will use it, which systems it touches and which numbers you expect to move. I get a bit wary when one runs to 15 pages, because in my experience the length is usually covering for a fuzzy scope.
This is the question that saves the most money, and it's worth asking without any politeness. Who owns what you're building for us? Where's our data going to sit? And if we stop paying in 18 months, what happens then?
If they're building on open-source tools and keeping your data in India, you keep your options open. A closed platform on some foreign server mostly keeps their options open, and that's not the same thing at all.
AI isn't a split AC you install and forget until summer. Prompts need adjusting, staff get stuck, and your pricing rules change every few months anyway. The thing I'd press them on is the day after go-live. Whose phone rings when something breaks, how long until they call back, and can they train your people in Hindi, or in whatever language they actually think in?
Rakesh's customers buy on credit, need GST invoices, and would far sooner send a voice note than type out an email, and plenty of small Indian businesses look much the same. Your partner has to get that. Someone who's only ever built for multinationals might hand you something way too heavy for a company of 60. One question I like to slip in casually is how they'd deal with a customer messaging in Hinglish at 11 at night.
Also, meet the people who'll actually do the work, and not just the charming person who sold you the deal. You'll be on calls with them for two months, so liking them helps more than you'd think.
Print this out and take it into your next vendor meeting, then score each firm quietly as they talk. From what I've seen, you'll usually know your answer by about the third row.
| Question to ask | Good answer | Red flag |
|---|---|---|
| What will we solve first? | 1 specific task with a clear outcome | "We'll transform your whole business" |
| Can we see a live system? | A running solution used by a real client | Only slides and videos |
| How is the pilot priced? | Fixed fee, fixed scope, fixed timeline | Open-ended hourly billing |
| Who owns the code? | You do, with full documentation | It stays locked in their platform |
| Where is our data stored? | In India, under your control | "Somewhere in the cloud" |
| What happens after launch? | A named support team with response times | "Just raise a ticket" |
| How will we measure success? | Numbers agreed before work begins | Vague talk about efficiency |
The cheapest quote is the oldest trap going. Very often a low price means a generic chatbot that can't see your billing or stock system at all, and then you're paying a second firm to wire it in properly.
Watch out for the beautifully formatted 12-month roadmap that turns up on the very first visit. I'd smile, thank them, and ask for one thing that actually works before anyone starts talking about the whole year.
Leaving your staff out of it is a quieter mistake, but it costs just as much. Get two or three of them testing from day one. If you don't, they'll nod along politely at the launch and then carry on doing it the old way.
And data rules aren't something you get to skip. India's DPDP Act applies to small firms just as much as big ones. Last thing, don't rush the decision. I'd give it two or three weeks at least, sit across the table from three firms or more, and only then compare.
The time to decide how you'll judge the project is before anyone writes a line of code. Two or three numbers is plenty. Maybe it's the hours your team saves each week, or how many queries get handled without a person jumping in. Maybe it's days knocked off collections, or fewer billing errors. Whatever you pick, note down today's figure honestly, even if it makes you wince a little.
Then look again at week 4 and week 8. And if nothing's moved by the end of the pilot, that's fine, really. You've paid for a cheap lesson instead of an expensive one.
Keep one eye on usage too, not only savings. If the tool saves 10 hours but only one person ever opens it, that's a toy, not a win. I'd ask for a short note every week during the pilot, and if those notes suddenly stop coming, take that as a warning sign.
We started Swaran Soft back in 2001, and these days we work with 350+ clients across India, the UAE, Europe and the USA. Most of our engagements begin with an 8-week fixed-fee pilot that goes from contract to production, built on an open-source stack that you own outright. We're ISO certified and a NASSCOM member too, which matters when it's your customers' phone numbers on the line.
One project we're quite proud of is an HR AI agent. It brought query resolution down from 2 days to under 4 minutes, and around 70% of questions now get answered automatically on WhatsApp. Swap the HR questions for order tracking or payment reminders and you can probably picture the same thing at Rakesh's firm. Our agents already speak 9+ Indian languages, which helps.
So where does that leave you? Mostly it comes down to three things: proof, ownership and focus. Go with the partner who wants to hear about your problems before they show off their product, and hold out for a small fixed-price pilot with targets you've agreed on up front.
Make sure the code ends up yours, and that you know where every byte of customer data sleeps at night. Bring your team in from the first week, since a tool people actually use will always beat a clever one they ignore. Pick one problem that's been nagging you, fix it properly, and grow from there. That's how Rakesh, and you, can get real value out of AI without betting the godown on it.
Want to talk it through with us? We run a free 45-minute AI Pilot Assessment. Drop a line to info@swaransoft.com, or ring +91 9220-313-650 (India) | +971-50-9292-650 (UAE). You'll find more at 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.