₹85,000 per month.
That was the AI API bill sitting in my client's inbox when they called me in a mild panic last quarter.
They run a mid-sized e-commerce operation in Pune — about 4,000 orders a day — and had integrated AI into customer support, product descriptions, and internal reporting.
The AI was working beautifully.
The invoice was not.
Three weeks later, their monthly bill was ₹12,400.
Same tasks.
Same quality.
No corners cut.
Here's exactly what changed.
The real problem: every task was using the most expensive model When I audited their setup, the issue was obvious within five minutes.
Every single API call — whether it was classifying a customer complaint into one of 8 categories or generating a 2,000-word product description — was hitting the same premium model.
It's the most common mistake I see with businesses adopting AI: they pick one model during the proof-of-concept phase and never revisit that decision as they scale.
You wouldn't hire a senior chartered accountant to do data entry.
But that's essentially what was happening — a top-tier reasoning model answering "Is this complaint about shipping or billing?" Fix 1: Model routing — the single biggest cost lever Model routing means sending each task to the cheapest model that can handle it at acceptable quality.
I categorised their ~47 distinct API call types into three tiers. 68% of calls moved to the lightweight tier, 20% to mid-tier, only 12% stayed on premium.
That single change dropped the bill from ₹85K to roughly ₹38K — no quality loss, verified with two weeks of A/B testing on customer satisfaction scores before switching fully.
Fix 2: Prompt caching — stop paying for the same context twice Their support bot sent the same 1,200-token system prompt with every call — policies, tone, catalogue context, all identical across thousands of daily calls.
Caching processes it once and references it cheaply on subsequent calls within the window.
At ~6,000 support interactions a day, this alone saved ₹8,000-10,000 monthly.
Fix 3: Batching non-urgent requests Not everything needs a real-time response.
Their internal reporting made individual calls as each data point arrived — sixty to eighty calls that could be batched into three or four.
We collect through the day and process in windows at 6 AM, 2 PM, 10 PM.
Batch pricing is typically half of real-time, and a few hours of delay is fine for internal reports.
Fix 4: Output token discipline Their product prompts asked for "a detailed, comprehensive description" and got 800-1,000 tokens when 200-300 was the real need.
We rewrote them with explicit length and structure: "Write it in exactly 3 sentences.
What it is.
Key benefit.
Who it's for." Output tokens cost more than input; cutting length 60% across thousands of daily calls compounds.
The final numbers An 85% reduction, ₹85,000 to ₹12,400, same work.
Customer satisfaction actually rose 3% — the lighter models respond faster, and people prefer a quick reply to a marginally more eloquent one.
What most people get wrong about AI costs The instinct is to shop for a cheaper provider.
The real leverage is architectural.
I've seen businesses switch providers three times and still overpay because the pattern — one model for everything, no caching, verbose outputs — never changes.
If your bill is higher than you'd like, ask: how many calls actually need a premium model?
Are you sending the same context repeatedly?
Can any calls be batched?
Are your prompts asking for more output than you use?
The answers usually reveal that 60-80% of the bill is waste hiding in plain sight.
You don't need to spend less on AI.
You need to spend smarter.
I write about automation and the systems I actually run at architmittal.com.
Originally published there.