AI for marketing: what a small Indian brand can automate

One marketer wrote about a renewal discount that an automation sent to about 4,000 people, including customers who had cancelled 18 months earlier. The email got a 41% open rate and a terrible reply rate. The tool did its job. Nobody checked who it was told to talk to.

That story is the whole topic of this post. AI is very good at doing a job fast and very bad at knowing whether the job should have been done. Most guides for small businesses are written by software vendors, with US examples and no rules for India. Here is a plainer version for a brand with a small team and a budget you watch every week, including two Indian rules that start in 2027.

What can a small brand automate with AI this week?

Start with work that repeats, that the customer never sees in raw form, and that you can undo. These are the jobs that pass.

Pick one row, not six. A ChatPlace guide for solo creators suggests the DM inbox as the first target and 20 frequent questions as the starting list, which is a sensible size for a small brand too. Be careful with its advice, though: that article is written by a company that sells the tool, and it gives no results data.

Here is the part most guides skip. Automating a message is cheap. Checking it is where the time goes. The marketer above runs a 15 minute audit every week on 20 to 25 random contacts, and reports 0.3% complaints on audited workflows against 1.8% on unchecked ones. That is one person's own count with no sample size, so treat it as a habit worth copying and not as a law. The habit is simple: for every automation you switch on, put a 15 minute check in the calendar on the same day.

Which marketing jobs should a small brand never hand to AI?

Use the rupee, face and claim test. Before anything automated goes out, ask three questions.

1. Does it carry a rupee? Prices, discounts, coupon codes, refund promises. A wrong number here costs you money on the spot. A bot that offers 40% off during a conversation is a bot you have to honour.
2. Does it carry a face or a voice? Your founder, a creator, a customer. If AI makes it up, the person in it did not say it.
3. Does it carry a claim? "Clinically tested", "lasts 12 hours", "works with iPhone 16". A claim needs proof, and AI will happily write a proof-sounding sentence that nobody tested.

One yes means a human approves it. Two or three yeses mean a human writes it. Complaints and refund fights always go to a person, because a tired customer wants to feel heard, and a polite loop of apologies feels like the opposite.

I would add a fourth rule for brand owners: do not let AI decide what your brand looks like. The sameness problem I wrote about in why Indian D2C brands look the same on Instagram gets worse when five competitors feed the same prompt into the same tool. AI is trained on the average of what exists. If your goal is to be remembered, the average is the one place you do not want to live.

Does AI work for product videos and ads?

For ideas, scripts and rough cuts, yes. For showing the actual product doing the actual thing, no, and from 2027 there is a legal reason too.

AI video tools can generate a person holding a product, but they do not know how your hinge clicks, how your serum pours or what your fabric does in light. When I film a product, the first thing I check is the one physical action a competitor cannot copy. A generated clip cannot find that shot, because the tool has never held the product. I covered how to find it in how to shoot a product reel people share.

Pricing for AI video is also cheap enough to tempt you. One Indian vendor page lists plans starting at ₹999 a month, though it is the vendor's own claim and gives no tier list. Cheap drafts are fine for testing a hook. The risk starts when you publish them as your product demo.

What do India's new ad and data rules mean for AI marketing?

Two rules matter for a small brand, and neither is covered by the ranking guides I read.

ASCI's AI labelling rules. The Advertising Standards Council of India published its guidelines on 29 September 2026, and they apply from 1 January 2027. In short:

• You must label AI-made content when it could change a buying decision. Examples: a synthetic influencer, a replica of a real person's voice or face, a fabricated setting, a demo of a product that does not exist yet, and sound effects tied to a core feature (the article's example is headphones).
• You do not need a label for routine edits such as colour correction, noise reduction or small blemish removal, or for AI-written ad copy.
• Some things are banned even with a label: fake testimonials, deepfakes of anyone without consent, and exaggerated results.
• The advertiser stays responsible. A label does not make a misleading ad acceptable.

This is my reading of a news report and not legal advice, so read the ASCI text itself before you plan a campaign. The practical point for a founder is that the label is the small part. The real exposure is a generated clip that shows your product doing something it cannot do.

The DPDP Rules on customer data. The government notified the Digital Personal Data Protection Rules on 14 November 2025. Consent notice duties start 18 months later, so around May 2027. Notices must say in plain language what data you collect and why, and give people a link to withdraw consent. Penalties can reach ₹250 crore per breach. Pasting a customer list into a free AI tool is the first habit to stop now, before the rule starts.

If you read that and thought "we have never once checked what our AI tool does with customer data", send me a message on LinkedIn with what you want to film or automate. Working out which part needs a real camera and which part a tool can draft is the part of the job I enjoy most.

How do you start AI marketing in one evening?

Do these steps in order. Total time is about two hours.

1. Open your DM inbox and your support emails. Copy the 20 questions you answer most.
2. Write the 20 answers yourself, in your voice. This is your source of truth.
3. Set up one tool (an inbox tool, a chatbot or even a saved-replies list) to answer only those 20.
4. Add a rule: anything with a price change, a complaint or the word "refund" goes to you.
5. Ask the tool for 10 caption drafts from your last product launch. Pick one, rewrite the first line.
6. Put a 15 minute check in your calendar for next Friday. Read 20 random replies the tool sent.
7. Write down one number to watch, such as reply time or complaints per 100 chats. Compare it in four weeks.

If your product shoot is the next step, write a one page brief first. How to write a product video brief has the eight questions to answer, and the same answers make your AI drafts better too.

Questions people ask

Can a small business in India use AI for marketing without a big budget?

Yes. Start with a free or low-cost chatbot or writing tool for drafts and repeat replies, and spend your time on the checking habit. The cost that hurts is a wrong message sent to many people, and a weekly 15 minute review prevents most of it.

What should small businesses not automate with AI?

Prices and discounts, complaints, refunds, product claims and anything that shows a real person saying something they did not say. Use the rupee, face and claim test: if a message carries one of the three, a human approves or writes it.

Do I have to label AI-generated ads in India?

From 1 January 2027, ASCI's guidelines ask for a label when AI content could change a buying decision, such as a synthetic influencer or a fake demo. Routine edits and AI-written copy do not need one. A label never excuses a false claim. Read the full ASCI text before you run a campaign.

Will AI replace my marketing team?

For a small brand, AI replaces the first draft and the repeat reply. It does not replace the person who knows your customer and decides what you stand for. Give the drafts to the team so they have more time for the films, offers and relationships that bring in sales.

Is it safe to put customer data into AI tools?

Check what the tool does with the data before you paste anything. Under the DPDP Rules, consent notices become mandatory around May 2027, so start by removing names, phone numbers and addresses from anything you share with a tool, and keep a written list of which tools see which data.