₹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.
