Every time I talk about AI to a business owner in Kathmandu, the same question arrives within two minutes: kati kharcha lagcha?
It is the right question. Most answers you find online are American, priced for American margins, and quietly assume you have a credit card that works. So here are my actual numbers — what I pay, what my clients pay, and what each thing returned. Where I give a figure, it is one I can point at.
Nepal ma AI chalauna kati kharcha lagcha?
My own working stack runs about NPR 35,000 a month, all in.
The bulk of that is two subscriptions: Claude at USD 200/month and Google's Gemini plan at USD 20/month. That is USD 220, which lands around NPR 31,000 depending on the day's rate. The remainder is API usage — metered calls I only pay for when a job genuinely needs them.
That number shocks people. It should be weighed against the right thing, though, and the right thing is not the sticker price.
Yo mahango haina ra?
Here is the comparison that actually matters.
I had been running Gemini's free tier and hitting the usage limit within about three days, every time. Work stopped. When I moved to the paid Claude plan, a body of work that had taken Gemini roughly a month came back in two to three hours on the first day.
So the question is never "is NPR 35,000 expensive". The question is: what does a month of your time cost, and what does it cost you to not ship for three weeks? Judge the spend against hours recovered, not against the number on the invoice.
If those hours don't come back, the tool is expensive at any price. That test is honest in both directions.
Ek AI agent ko lagi kati lagcha?
Separate from my tooling, the AI sales agent that handles customer messages costs about NPR 10,000–12,000 per month to keep running.
What it does for that: it answers customer questions on WhatsApp and Instagram, instantly, at any hour. A customer messaging at midnight gets a real answer at midnight. No staff member is awake, and no message sits unseen until morning.
Think of it as a staff member with an unusual contract. It does not take a salary. It does not get tired, does not take chhutti, does not have a bad day. You teach it once and it repeats that a thousand times without drift. It is not smarter than your best employee — it is simply always there.
The reason this matters commercially is boring and well documented: speed to reply drives conversion. The faster you respond, the more you sell. A message left unseen is not a neutral event. That customer goes somewhere else.
Marketing ma AI le kati bachaucha?
The clearest receipt I have is an ad campaign where the creative was written by AI.
Total spend: USD 8.
From that eight dollars I hired 3 staff, and the same eight dollars produced 1,831 website visits. One tiny budget, two entirely different business outcomes.
The lesson is not "ads are cheap". It is that the lever was never the budget. The lever was the creative and the targeting. Before you conclude you need money to test something, run five or ten dollars through it and find out whether the offer works at all. Most people skip that step and then blame the budget.
Mahango computer chahincha?
No. This is the most common excuse and the easiest to dismantle.
My own setup is expensive, and it makes no difference to you. The models do not run on your computer — they run on the provider's servers. Your machine is sending text back and forth. A basic laptop, or a phone with a working internet connection, is enough to start.
If you are waiting to buy hardware before you begin, you are not blocked by hardware.
Kaha bata suru garne?
Pick the single task eating the most of your week and hand that one over. In most Nepali businesses I see, it is one of four:
- Marketing — ad copy, creative, captions
- Customer reply — the messages that go unanswered
- Follow-up — the leads nobody chased
- Paperwork — the admin that quietly consumes evenings
Start with one. Not four. Measure the hours it gives back over a month, then decide whether the spend was worth it using your own numbers rather than mine.
What I would not tell you
AI does not replace an expert. I write code and I run marketing, and when I sit down and do either myself with full attention, the result is better than what the AI produces alone. The gap is real.
Where it wins is volume, patience, and availability — the thousand repetitions, the midnight reply, the first draft that saves you forty minutes. Handing it work that genuinely needs your judgement produces mediocre output with your name on it.
The honest framing is not AI will do your job. It is that someone who uses AI well will out-produce someone who refuses to. That has been true of every tool, and it is true of this one.
I'm Prajwal — a business consultant in Kathmandu. I help Nepali businesses grow with sales, marketing and AI leverage. Everything here comes from client work or my own; if I quote a number, I can show you where it came from.