Most automation projects in India do not fail because the technology did not work. They work in the demo, they work in week one, and then eight months later nobody can say whether they are still running.
We build this stuff, and we have also been the people who inherited someone else's. The failures follow patterns, and almost none of them are technical.
Here are the six questions that actually predict how it goes, roughly in the order worth asking them.
1. Whose name are the accounts in?
Start here, because it is the most expensive thing to get wrong and the least likely to come up in a sales conversation.
Automation sits on top of accounts: a WhatsApp Business Account, a Meta app, an ad account, a Google Cloud project, an OpenAI or Anthropic key, a CRM. Someone has to own each one. If the agency opens them under their own business, you have bought a service you cannot take with you.
The clearest example is WhatsApp, because Meta's own documentation makes the consequence concrete rather than theoretical:
Templates can only be migrated between WABAs owned by the same Meta business.
Read that carefully. If your provider owns the WhatsApp Business Account, moving to a different provider does not mean a slow template migration. It means the templates cannot move at all, and every one has to be resubmitted from scratch and re-approved. Businesses discover this at the exact moment they want to leave, which is the worst possible moment to discover it.
The same shape applies elsewhere. An ad account in the agency's name takes your pixel history with it. An AI key in the agency's name takes your usage history and your billing relationship.
What to ask: "Which accounts will be registered to my business, and which to yours?" A good answer is "all of them to yours, we get admin access". Anything else needs a reason.
2. What happens on the day it breaks, at 9pm?
Every automation breaks. Tokens expire, Meta changes a rule, a webhook stops arriving, an API adds a required field. The question is not whether, it is what happens next.
Three things worth pinning down:
Who notices? If the answer is "you tell us", the automation is only monitored during the hours you happen to look at it. Real monitoring means the system tells someone when a job stops running, not when a customer complains.
What is the response commitment? Not a promise to care, a number. Same working day is a reasonable ask for a small engagement.
Does anything fail loudly? This is the one people miss. Automation fails silently by default: a message that does not send produces no error anywhere a human will see. Ask specifically how failures surface.
We wrote about this from the WhatsApp side in why "sent" means nothing. A system reporting success while delivering nothing is the normal failure mode, not an unusual one.
3. What am I handed at the end?
A pilot that produces a working demo and no documentation is a pilot that dies when one person leaves.
Ask for the list, in writing, before you start:
- Credentials, in your accounts, with you as owner
- A written description of what runs, when, and what it touches. Not a diagram, sentences.
- Where the logs are and how to read them
- What to do when X fails, for the two or three most likely X
- Who to call if the agency relationship ends
If the answer is "we'll walk you through it on a call", that is not a handover, that is a memory test.
4. How is it priced, and what happens at volume?
Three models in the Indian market, and the differences matter more than the headline number.
| Model | Works when | Watch for |
|---|---|---|
| Fixed project fee | Scope is genuinely fixed | Anything not in the document is a change request |
| Monthly retainer | It needs ongoing care, which most automation does | What the retainer actually includes, in hours or in outcomes |
| Per-message or per-usage markup | Rarely in your favour | This is the one to interrogate |
The third deserves attention. On WhatsApp, Meta charges the same rate whoever you buy through. A provider adding 10 to 30% on top of each message is charging you for volume they did not create, and it is often not on the pricing page. At ten thousand messages a month a 15% markup is roughly ₹13,000 a year for no additional service, which is usually more than the difference between the platform fees you were comparing.
Ask directly: "Is there any per-message or per-usage markup, and what is it?" The answer being "no" is worth confirming in writing.
There is a full breakdown in how to choose a WhatsApp marketing tool in India and a pricing calculator if you want to model your own volume.
5. Can you show me something that has been running for a year?
Not a demo. Not a case study PDF. Something live, that has survived a platform change.
A demo proves someone can build. Something running a year later proves someone can maintain, which is the harder and more valuable skill. Anyone who has kept a Meta integration alive through twelve months of Meta changing its mind has learned things that do not appear in a portfolio.
If the answer is a screenshot, ask what broke in that year and how they found out. The specificity of the answer tells you everything. People who have genuinely operated something will have a story about a token expiring at a bad time.
6. What will you tell me not to build?
The most useful signal in the whole conversation.
Every automation brief contains something that should not be built: a feature the platform does not allow, a workflow that will break the first time reality intervenes, or a nice-to-have that will consume half the budget. An agency that says yes to all of it either has not thought about it or is not going to tell you.
Concrete example from our own world. People regularly ask for automation that sends a DM to every new Instagram follower. It cannot be built. Instagram publishes no webhook for a new follower, so no software is ever told the follow happened, and Meta's documentation states a business cannot start a conversation at all. Any tool offering it is driving a hidden browser session against your account, which is what actually gets accounts restricted. We wrote the evidence up in what Instagram automation genuinely cannot do.
An agency that quotes for that feature is either going to fail or going to endanger your account. One that explains why it is impossible is one you can trust with the rest of the brief.
Red flags
Shorter list, all of them things we have watched cost someone money:
- "We'll handle the WhatsApp account for you" without specifying whose name it is in
- A demo on their own data rather than yours, at every stage
- No mention of monitoring, only of building
- Pricing quoted per message without the Meta rate broken out separately
- Guaranteed results on a channel they do not control, particularly "guaranteed leads" from organic social
- Nobody technical in the room after the first meeting
What good actually looks like
Modest, and mostly boring:
- A small first scope that produces something you can judge in weeks, not quarters
- Your accounts, your data, admin access granted rather than ownership transferred
- Something that alerts a human when it stops
- Written handover before the invoice
- Someone who told you not to build something
None of that is about how clever the automation is. It is about whether it will still be running when you have stopped paying attention to it, which is the only test that matters.
A note on who we are
Kredoo is a CRM with WhatsApp and Instagram automation built in, made by ElevAIte Labs, an automation company based in Hyderabad. So we are not a neutral party on question 4, and you should read this with that in mind.
The reason we are firm about question 1 is that it is the thing we get asked to fix. Every WhatsApp account a Kredoo customer uses is registered to their business, sending through their own Meta credentials, because we have seen what the alternative costs when someone wants to leave. That is a design decision made from watching it go wrong, not a feature.
If you are choosing between providers, the switching guide is worth reading before you sign anything, because it describes exactly what leaving involves.
And if you are shortlisting locally, who actually appears when you search for AI automation companies in Hyderabad shows where each name on those "top companies" lists comes from, including why the map pack ordering reflects review counts rather than any judgement about engineering.
Questions people ask
What should I ask an AI automation agency before hiring them?
Start with whose name the accounts will be registered in, because it is the most expensive thing to get wrong. Then: who notices when it breaks, what documentation you are handed, whether there is any per-message markup, whether they can show something that has been running for a year, and what they will tell you not to build.
Why does it matter whose name my WhatsApp Business Account is in?
Meta's documentation states that templates can only be migrated between WhatsApp Business Accounts owned by the same Meta business. So if your provider owns the account, moving to another provider means your approved templates cannot transfer at all and must be resubmitted and re-approved from scratch. Most businesses discover this only when they try to leave.
How much should AI automation cost in India?
The headline fee matters less than the model. A fixed project fee suits genuinely fixed scope, a retainer suits work that needs ongoing care, and a per-message markup usually favours the provider rather than you. On WhatsApp specifically, Meta charges the same rate whoever you buy through, so a 10 to 30% markup is a charge for volume the provider did not create.
Why do automation pilots fail even when the technology works?
Because nobody owns them after handover. Automation fails silently by default: a message that does not send produces no error anywhere a human sees. Without monitoring that alerts someone, and documentation that survives one person leaving, a working pilot quietly stops and nobody notices for months.
What is a red flag when hiring an automation agency?
The clearest one is an offer to handle your WhatsApp or ad accounts without specifying whose business they are registered to. Others are demos only ever run on the agency's own data, no mention of monitoring, per-message pricing that does not separate out Meta's own rate, and guaranteed results on a channel nobody controls.
Should an automation agency ever tell me not to build something?
Yes, and it is the most useful signal available. Most briefs contain something the platform does not permit or that will break on contact with reality. An agency that quotes for an impossible feature, for example DMing every new Instagram follower, either has not checked or intends to do it in a way that risks your account.
Stop losing leads to the follow-up gap
Kredoo captures leads from Meta and Google Ads, then follows up on WhatsApp automatically. Live in 5 minutes.
Get Started