The short answer
A support hire covers about 40 hours a week of one time zone and costs the same whether the questions are hard or repetitive. An AI agent covers 168 hours and only handles the repetitive share. The right comparison is not hire-versus-bot - it is which of the two you buy first, and for most small teams the answer is the agent, because it makes the eventual hire a better job.
You have reached the point where support is eating the day. The obvious next move is to hire someone. Before you write the job ad, it is worth being precise about what the role would actually do - because when small teams write that list down honestly, most of it turns out to be work nobody should be paying a person to do.
This is not an argument against hiring. It is an argument about sequence. The first support hire is a good decision made too early far more often than it is a bad decision - and the reason is that the job, on day one, is mostly repetition.
What the job actually consists of
Export a month of your support inbox and sort it by question. Almost every small business that does this finds the same shape: a long tail of genuinely interesting problems sitting under a very fat head of fifteen questions asked hundreds of times.
Shipping timelines. Return windows. Sizes. Whether you deliver to a particular pincode. Whether COD is available. Whether there is a GST invoice. Where the order is. Those questions have answers you have already written down somewhere - and a person answering them is, functionally, an expensive way to read a PDF aloud.
| Share of inbox | What it looks like | Does it need a person? |
|---|---|---|
| ~60% | The same fifteen questions, reworded | No - the answer is already in a document |
| ~20% | Order status and 'where is my parcel' | Partly - policy answers most of it |
| ~10% | Complaints, damage, disputes | Yes, and urgently |
| ~10% | Bulk, B2B, partnerships, edge cases | Yes - and this is the profitable part |
The bottom two rows are the job you want a person doing. The top two are the reason nobody gets to them before 4pm.
What a salary buys, and what it doesn't
We are not going to print a salary figure, because it varies too much by city and experience for any number we quote to be honest. You already know what the role costs in your market. What is worth spelling out is what that money does and does not cover.
It covers roughly forty hours a week, in one time zone, five or six days out of seven. It does not cover the 8pm-to-midnight window where Indian e-commerce traffic actually peaks, which is the midnight-query problem in its most expensive form. It does not cover the week they are ill, the fortnight they are on leave, or the month after they resign. And it does not scale with a festival sale: the same person faces five times the volume in Diwali week for the same salary, which is not a saving, it is a queue.
- One hire covers about a quarter of the week's hours, and none of the peak ones.
- Recruitment, notice periods and training are real costs that never appear on the salary line.
- Coverage has a single point of failure until you hire the second person.
- Volume spikes do not add capacity - they add backlog, which is a slower and more damaging failure.
The comparison people actually make, and the better one
The framing everyone reaches for is cost: is a subscription cheaper than a salary? It is, obviously and by a wide margin, and that comparison is too easy to be interesting.
The comparison that matters is coverage against judgement. A person brings judgement - the ability to read a furious message and know that the policy answer is technically correct and completely wrong for this customer. Software does not have that and should not pretend to. What software brings is coverage: every hour, every channel, at a volume that does not care whether it is Tuesday or Diwali.
So the two are not substitutes competing for the same budget line. They are complements covering different failure modes, and the sensible order is to buy coverage first, because coverage is what makes judgement affordable.
Hire for judgement. Automate coverage. Doing it in the other order is how you end up paying a person to read your own return policy aloud.
What happens if you automate first
Run the sequence forward. You put a grounded agent on your site and your WhatsApp number. It answers the top two rows of that table from your uploaded documents, with a citation on every reply, at any hour. The bottom two rows arrive in an escalation inbox with the full transcript attached.
Three months later you are hiring. The job you are hiring for now is different: it is disputes, bulk enquiries, and the relationships that come out of them. That is a role a good person will stay in. The version of the job where they answer 'do you ship to Wakad?' four hundred times a month is the version people leave within a year, and replacing them costs you the recruitment cycle all over again.
There is a second effect that is easy to miss. Because every escalation carries the whole conversation, the new hire's first reply picks up mid-conversation instead of starting over. Their ramp-up is shorter, because the context they need is attached to the work rather than living in someone else's head.
What happens if you hire first
The other order is not a disaster - plenty of businesses run it. But it has a predictable shape. The new hire spends their first months on the repetitive share, because it is loud and it is urgent and it arrives all day. The interesting work still waits, because it always waits. Response time improves during business hours and does not change at all outside them.
Then volume grows, which is the outcome you wanted, and one person is no longer enough. The obvious fix is a second hire, and the arithmetic that was already unfavourable doubles. Somewhere in that second cycle most teams automate anyway - having paid for a year of the thing automation was going to do.
When hiring genuinely comes first
There are businesses where this is the wrong advice and it is worth being straight about which.
If your support volume is low but every conversation is high-value and consultative - enterprise sales, bespoke manufacturing, anything where the conversation is the product - there is no repetitive 60% to remove, and a person is simply the right tool. If your answers change daily and are never written down anywhere, an agent has nothing to ground itself in, and the fix is documentation before it is software. And if most of your inbound needs an action rather than an answer - process this refund, change this order - then the constraint is admin access, not information.
The test is the export. If the top fifteen questions are more than half your volume, automation is the first purchase. If they are not, hire.
The hybrid almost everyone lands on
In practice the steady state is not either extreme. The agent takes every conversation first, answers what your documents cover, and hands over anything it is not confident about - along with the transcript, the customer's contact details and which documents it searched. A person handles what arrives, and because only the genuinely human share arrives, one person covers a volume that would previously have needed three.
That is the arrangement worth costing, and it is a very different sum from the one you started with. You are not asking whether to spend a salary or a subscription. You are asking how much support headcount a subscription lets you not hire yet - and for most small teams, the honest answer is about two years' worth.
Where to start: run the export, count your top fifteen, and read what to put in the knowledge base so the agent has something to ground itself in. Our plans start free, which is enough to test the premise on your real inbox before you decide anything about headcount.
Frequently asked questions
Does an AI agent replace a support hire?
Not entirely. It removes the repetitive share of the work - typically the same fifteen questions arriving over and over - so the person you eventually hire spends their time on disputes, bulk orders and relationships instead of quoting the return window.
How many queries actually need a human?
In the inboxes we have looked at, roughly one in ten. The other nine are answerable from documents you have already written. The problem is that the one is buried under the nine, which is what makes response times slip.
What if the AI answers something wrong?
A grounded agent answers only from your uploaded documents and cites the file it used, so a wrong answer is a wrong document - and you fix it by editing the document. It escalates rather than guessing when nothing matches.
Is it cheaper than hiring?
VaaniAPI is a flat monthly plan from $19 with no per-message fee, so the comparison against any salary is not close. The more useful question is coverage: a hire covers one shift, five or six days a week; the agent covers every hour including the 8pm-to-midnight window where most Indian shopping happens.
