Across Kerala, the pitch sounds roughly the same: "Our AI platform will automate your operations, reduce your team's workload by 60%, and give you a 3x return on investment within a year." The brochure is polished, the demo is impressive, the price is ₹4 to ₹6 lakh for the first year, and the salesperson is on a tight quarterly deadline. For many Kerala business owners — in manufacturing, healthcare, education, retail, and services — this conversation is now a monthly occurrence. Before you buy that expensive AI software, there are ten questions you should demand clear answers to, and several red flags that should make you walk away entirely.
Why AI Pitches Are Flooding Kerala Businesses Right Now
AI software vendors are targeting Indian Tier 2 and Tier 3 cities aggressively in 2026. The reasoning is straightforward: enterprise clients in Mumbai and Bengaluru are more sophisticated buyers who ask harder questions and demand longer pilots. SME owners in Kochi, Thrissur, Kozhikode, and Trivandrum are being targeted because the competitive intensity from the buyer's side is lower and deals close faster.
This doesn't mean all AI software sold to Kerala businesses is bad. Some of it is genuinely useful and well-suited for local business processes. The problem is that the buying process is rushed, the demos use polished sample data rather than the buyer's actual messy data, and the post-sale support — especially in Malayalam or in Indian time zones — is often weaker than the pre-sale promises suggest.
Most Kerala businesses that regret expensive AI purchases cite three root causes: the tool solved a problem the business didn't actually have, the team never adopted it because training was inadequate, or the business's own data wasn't in a format the AI could use. All three of these are entirely preventable with the right questions asked upfront.
10 Questions to Ask Before Any AI Purchase
1. Does this solve a specific business problem, or is this AI for AI's sake?
Start here before anything else. Identify the one specific, measurable problem the AI is supposed to solve. "Improve efficiency" is not a problem. "Our accounts team spends 14 hours a week manually entering purchase orders into Tally because our suppliers send them as PDFs" is a problem. If the vendor's product can solve that exact problem, you have a real use case. If the vendor responds to your specific problem with a generic demo of their platform's broad capabilities, that mismatch will not resolve itself after purchase.
2. What data does this need, and do I already have it in usable form?
AI tools run on data. Before a single rupee is committed, ask the vendor exactly what data their system requires — its format, its volume, its history, and its labelling requirements. A demand forecasting tool needs at least 12 to 24 months of clean sales data, broken down by SKU and location. A customer churn prediction model needs structured records of customer behaviour, not a collection of WhatsApp screenshots. If your business data lives in spreadsheets from ten different people, scanned PDFs, or inside a salesperson's phone, factor in the cost of data cleaning before you factor in the cost of the software.
3. Who in my team will actually use this day-to-day?
Name a specific person. Then ask whether that person's current role gives them the time and authority to use the tool consistently. AI tools fail most commonly not because they don't work but because no one in the organisation takes ownership of using them. The vendor will say "your team will be trained" — but training and adoption are different things. If the answer is "the IT person will manage it," ask whether that IT person has capacity beyond their current workload.
4. What does the onboarding and training process look like?
Ask for specifics: how many hours of training are included, whether it is recorded so new employees can access it, whether there are training materials in Indian English (or Malayalam), and what happens if the person who got trained leaves the company. The best vendors provide structured onboarding with check-ins at 30, 60, and 90 days. Vendors who hand you a PDF manual and a support email after sign-up are selling software, not solutions.
5. Does it support Malayalam, or Indian data formats like GST invoices and Aadhaar?
This is a Kerala-specific but critical question. Many AI tools — particularly those built for Western markets — have poor support for Indian invoice formats, GST reconciliation fields, or Aadhaar-based identity verification. If the tool has a customer-facing component (a chatbot, a voice assistant, a document processor), ask specifically whether it handles Malayalam input and output. "We'll add that in a future update" is not a satisfactory answer if you need it now. Confirm this with a live demo using a Malayalam-language input or an actual Indian GST invoice, not a demo template.
6. What happens to my data — and is it stored on Indian servers?
India's Digital Personal Data Protection Act has clear implications for businesses handling customer personal data. If the AI software processes your customers' names, phone numbers, purchase history, or health records, you need to know exactly where that data is stored, who can access it, and how it is deleted when you end the subscription. Ask for this in writing. Many AI vendors route data through servers in the United States or Singapore. While this is not automatically illegal, it does create compliance exposure that you are responsible for, not the vendor. For a Kerala cooperative bank, a hospital, or a school — sectors with heightened data sensitivity — this question is non-negotiable.
7. What is the real total cost: license, implementation, training, and maintenance?
The price quoted in the pitch presentation is almost never the total cost. Get a written breakdown that includes: the annual or monthly license fee, the one-time implementation and setup fee, the cost of integration with your existing software (Tally, Zoho, WhatsApp Business), the cost of training, the cost of support beyond the included tier, and the cost if you need to add users or data volume. It is common for a ₹3 lakh quoted price to become ₹5 to ₹7 lakh once implementation, integration, and year-one support are properly accounted for.
8. Can I see a demo on my data, not generic sample data?
Any AI vendor willing to run a credible demo on your actual data is demonstrating that the product works in conditions that match your reality. Vendors who insist on using their own polished demo environment are protecting themselves, not you. Bring a sample of your real data — 100 invoices, 6 months of sales records, a set of actual customer queries — and ask them to run it through the system live. If the output is messy, incomplete, or requires extensive manual correction, you have just saved yourself a significant amount of money and frustration.
9. What does the exit strategy look like if this doesn't work?
Ask before you sign: how do you export your data if you decide to stop using the platform? How long does the export take, and in what format does it arrive? What is the notice period to end the contract, and are there penalties for early termination? What happens to your historical data after you leave — is it deleted, archived, or retained by the vendor? Vendors who make exit straightforward are confident in their product. Vendors who make exit difficult are protecting their revenue at your expense.
10. Who are three comparable businesses in Kerala or India using this successfully?
Not a generic case study on the vendor's website — an actual reference you can call. Ask for businesses that are comparable to yours in industry, size, and geography. A ₹10 crore textile manufacturer in Thrissur and a ₹200 crore logistics company in Pune face fundamentally different implementation challenges. If the vendor cannot provide a single Indian reference willing to take a 15-minute call with you, treat that as a significant gap in their track record.
Red Flags That Should Make You Walk Away
Beyond the ten questions, watch for these specific patterns in vendor behaviour during the sales process.
- Unrealistic ROI claims without underlying assumptions: "300% ROI in 12 months" is a marketing claim, not a financial projection. Ask them to show you the model: which specific costs will drop by how much, which revenues will increase by how much, and what assumptions underpin those numbers. If they can't produce a calculation, the number is invented.
- No pilot option: A vendor who won't allow even a limited 4-week pilot on a subset of your real workflows is either protecting a weak product or is under pressure to book full contracts immediately. Both situations are bad for you.
- No Indian references willing to be contacted: As described above. Generic testimonials from companies you can't verify are not references.
- Support is entirely offshore with no Indian time zone coverage: If you run a retail operation in Kozhikode and your AI inventory system breaks down at 9 PM on a Thursday before a Vishu sale, a support team that operates only during US Eastern hours is not functional support. Confirm explicitly whether there is India-based or at least Asia-Pacific timezone support available.
- Contract requires annual prepayment with no milestone clauses: Paying a full year upfront before the system is even implemented transfers all the risk to you. Push for a payment structure that ties at least the second and third tranches to successful implementation milestones.
How to Run a Proper AI Pilot
A pilot is not "let us try it for a month and see what happens." A proper pilot has four specific components before it starts.
A pre-agreed success metric: Define exactly what success looks like in measurable terms before the pilot begins. "Invoice processing time drops from 14 hours per week to under 5 hours per week, with an error rate below 2%" is a success metric. "The team feels like it's saving time" is not.
Real data, not demo data: The pilot must run on your actual data, including your messiest, most edge-case data. If the system handles your cleanest records well but fails on the 30% of records that are exceptions, you need to know that during the pilot.
Your team operating the tool, not the vendor: The vendor can provide support during the pilot, but your own staff should be the ones running the system. If your team cannot operate it without constant vendor hand-holding after three weeks of training, adoption will collapse the moment the vendor's attention moves to the next sale.
A clear go/no-go decision date: Agree on a specific date when you will make the purchase decision based on whether the success metric was met. Without a deadline, pilots drift indefinitely — which benefits the vendor, not you.
When to Build Custom AI vs. Buy Off-the-Shelf
For most Kerala SMEs, buying an off-the-shelf AI tool is the right starting point. The cost and timeline of custom AI development — building a model trained on your specific data, deploying it, maintaining it — is substantial, and off-the-shelf tools have improved significantly in their ability to handle Indian business contexts.
Consider custom AI development when: your business process is genuinely unique and not served by any existing tool (a specialised grading algorithm for a Wayanad spice exporter, for example), when the off-the-shelf tool requires so many workarounds that implementation cost exceeds the cost of custom development, or when your data volume and business scale justify the investment. If you are at that stage, a proper AI and machine learning consultation — before any vendor selection — will help you define the right architecture and avoid costly false starts.
The distinguishing question is this: does the off-the-shelf tool require your business to change its process to fit the software, or does the software adapt to your process? If every vendor demo requires explaining "well, in India we do it differently" and the vendor has no clear answer, that is a signal that their tool was not built for your market.
Frequently Asked Questions
Does the DPDP Act apply to AI software used by Kerala businesses?
Yes. India's Digital Personal Data Protection Act requires that any software handling personal data of Indian citizens — customer names, phone numbers, purchase history — must comply with data localisation and consent requirements. Before purchasing any AI software, ask the vendor explicitly whether your customer data will be stored on Indian servers, who has access to it, and how it is deleted if you end the subscription. Many offshore AI vendors have not yet adapted their data handling for DPDP compliance, which puts the liability on you as the business using the software.
What is the difference between buying an off-the-shelf AI tool and building a custom AI solution for my Kerala business?
Off-the-shelf AI tools are pre-built software products — think AI-powered CRMs, inventory forecasting modules, or chatbot platforms — that are configured to fit your business without custom development. Custom AI means building a model or system specifically around your own data and processes. For most Kerala SMEs, off-the-shelf tools are the right starting point because the cost and timeline of custom AI are significant. However, if your business has a genuinely unique process — like a specialised grading system for spices or a custom loan approval workflow for a cooperative bank — a custom solution may deliver better ROI than forcing your process into a generic tool.
How do I run a proper AI pilot before committing to a full purchase?
A proper AI pilot lasts 4 to 8 weeks, runs on a limited but representative subset of your real data, and has a pre-agreed success metric before it starts — not after. For example: "We will run the AI invoice processing tool on our last 3 months of invoices. Success means processing time drops by at least 40% with fewer than 2% errors." Without a pre-agreed metric, vendors can claim any result is a success. Also insist that your team — not the vendor's team — operates the tool during the pilot. If your staff can't use it without constant vendor support, that's critical information about adoption risk.