Almost every underperforming Meta Ads account we are handed in India has the same shape. The targeting has been rebuilt four times, the creative has not changed in three months, the product page has never been touched, and nobody has reconciled the ROAS in Ads Manager against what actually landed in the bank after returns. The campaign settings are the last place to look, not the first.
We are a Meta Business Partner and we run these accounts daily. What follows is the order we work in, and why.
Start with the offer and the landing page
Paid social does not create demand for a product nobody wants at a price nobody will pay. It buys attention and points it somewhere. If the somewhere is weak, more budget makes the loss bigger, faster.
The product page is the ad
A cold visitor from a feed arrives with roughly zero context and a thumb already moving. The page has to answer, above the fold and without scrolling, what this is, what it costs in rupees, when it arrives, and what happens if they do not like it. Delivery timelines and return policy are not footer material for a first-time buyer from an ad. They are the objection.
Trust signals do real work in this market. Visible reviews with photographs, a working phone number, a delivery estimate by pincode, and recognisable payment marks at checkout all reduce the hesitation that a cold audience arrives with. So does a page that loads quickly on a mid-range Android phone on a patchy connection, which is what most of your traffic is.
Cash on delivery changes the maths of every campaign
This is the part imported playbooks from the US and UK get wrong, and it is the single biggest structural difference in Indian D2C. A COD order is not revenue. It is an option on revenue that the customer can decline at the door, and every declined order costs you forward shipping, return shipping and handling on a sale that never happened.
Two consequences follow. First, your true return on ad spend has to be calculated on delivered orders, not on the purchase events Ads Manager reports, and the gap between those two numbers is a management decision rather than a rounding error. Second, anything that shifts orders from COD to prepaid is an advertising lever even though it lives outside the ad account: a prepaid-only discount, a free shipping threshold on prepaid, a confirmation flow on WhatsApp before dispatch. We treat prepaid share as a campaign metric, because it is one.
If you are optimising towards purchase events that include COD orders you will eventually unwind, you are teaching the system to find you more of the customers who do not pay. That is worth fixing before you touch anything else.
Account structure: fewer campaigns than you think
Consolidate so the system can learn
The instinct to split budget across many small ad sets comes from an era when manual targeting was the lever. It is now mostly a way to starve every ad set of the data it needs to get out of learning. A simple structure, with one prospecting campaign carrying most of the budget and a tightly scoped retargeting campaign alongside it, gives the delivery system enough signal in one place to do the work it is now better at than you are.
Restructuring is not a monthly activity. Every rebuild resets learning, and an account that is rebuilt every few weeks never leaves the noisy phase. Change one thing at a time, and give it long enough to mean something.
Audiences are not the lever they were
Broad targeting with strong creative now beats narrow interest stacking in most accounts we run. The exceptions are real but narrow: genuine first-party lists, high-intent site behaviour, and existing customers for cross-sell. Those belong in retargeting, kept small and kept separate, so you can see what prospecting is actually doing without retargeting flattering the number.
One caution on retargeting in India: a warm audience that converts at a flattering ROAS is frequently an audience that would have converted anyway. Judge the account on blended performance, not on the best campaign in it.
Creative is where the wins are
Volume beats polish
Targeting is largely automated. Creative is the input you still fully control, which makes creative production rate the real constraint on most accounts. A brand shipping a handful of new concepts a month will find winners. A brand shooting one beautiful campaign a quarter will not, however good that campaign is.
Test concepts, not colours. A different hook, a different format, a different objection answered, a different person on camera. Swapping a button colour is not a test, it is a way to spend the testing budget without learning anything.
What to shoot for an Indian feed
Vertical, sound-off-legible, and shot on a phone as often as not. Founder-to-camera explaining why the product exists. Unboxing. The product being used in a recognisably Indian home or street rather than a studio. Customer video, with permission. Text on screen, because a large share of views happen with sound off.
Price and offer on the creative itself is usually worth testing in this market. Buyers here are price-aware and a creative that hides the number attracts clicks that bounce at the product page.
Language
English-only creative is leaving reach on the table in most categories outside the top metros. Hindi, and Hinglish in particular, often outperforms polished English for the same product. If you sell regionally, test the regional language properly rather than running a translated caption over an English video.
Measurement you can defend
The platform number is not your revenue
Ads Manager reports conversions it believes it influenced, inside an attribution window you chose, before returns and before COD failures. Your accounts report money. These two numbers will never match, and the useful discipline is not to make them match but to know the ratio between them for your business and track it.
Run the check monthly: total ad spend against total delivered, paid-for revenue. That blended figure is the one to make decisions on. Campaign-level ROAS is for steering within the account, not for deciding whether the account is working.
Send server-side events, and use the data you already own
Browser-side tracking alone loses events to blockers, in-app browsers and consent choices. Sending conversions server-side from your store, deduplicated against the browser events, gives the system a cleaner signal to optimise against. On Shopify this is close to configuration rather than development work, and it is usually the highest-return technical job in a neglected account.
Your customer list is an asset the platform does not have. Uploading it properly, and keeping it fresh, improves both exclusion and lookalike quality. Exclude recent purchasers from prospecting unless the product is genuinely repeat-purchase on that cycle.
WhatsApp is part of the funnel, so measure it
For a lot of Indian brands the journey is ad, product page, abandoned cart, WhatsApp message, order. If the WhatsApp step is invisible in your reporting, the campaign that started the chain looks like it failed. Tag the source, pass it through to the conversation, and close the loop. Cart recovery over WhatsApp is frequently the cheapest revenue in the whole operation and it is routinely credited to nothing.
The first thirty days on a neglected account
In rough order, because order matters more than the list does.
- Fix tracking first. Server-side events, deduplication, and a purchase event that fires once and means what it says.
- Establish the blended number. Spend against delivered revenue, so there is a baseline that survives contact with the bank statement.
- Audit the product page on a mid-range phone, on mobile data, as a stranger would see it.
- Collapse the structure. One prospecting campaign with real budget, one retargeting campaign, and stop rebuilding.
- Get a creative pipeline running before optimising anything else, because everything downstream depends on it.
- Attack prepaid share and RTO with offers and confirmation flows, and watch what it does to the delivered number.
Frequently asked questions
What budget do I need to start on Meta in India?
Enough that the account can gather conversion data in a reasonable window rather than trickling. Rather than a fixed rupee figure, work backwards: at your current conversion rate and average order value, what daily spend produces enough purchases a week to tell signal from noise? If the honest answer is that no budget you can afford gets there, the problem is upstream in price or product, and more ad spend will not fix it.
Why did my ROAS collapse after I turned off a losing ad set?
Frequently because the ad set was not losing. Attribution assigns a conversion to one touchpoint, and a prospecting ad set that introduces the customer often gets credited to the retargeting ad that closed them. Turn prospecting off and retargeting has nobody to retarget. This is the strongest argument for judging the account blended rather than campaign by campaign.
Should I run ads directly to WhatsApp instead of my website?
For considered purchases, made-to-order products, and categories where buyers want to ask before they buy, click-to-WhatsApp is a legitimate primary objective and often outperforms sending traffic to a page. For a straightforward catalogue product it usually adds a manual step to a purchase that would have completed on its own. Test it as a campaign rather than adopting it as a policy.
Do I need a separate agency for Meta Ads and for the store?
You need the two to talk. Most of the levers in this article, page speed, product page structure, server-side events, prepaid offers, cart recovery, are store work that shows up as advertising performance. When the ad agency and the store team are separate and do not speak, those jobs sit in the gap between them and nobody owns the number.