Practical notes from the engineers who build agents, copilots and AI products.
Build in house when the application is your core product and you already have enough engineers to staff a sustainable rotation. Buy application maintenance and support services when the system matters but does not differentiate, or when nights and weekends rest on one tired person. Most companies blend the two.
A first production AI system in banking or insurance typically costs $12,500 to $45,500, or ₹8 lakh to ₹30.4 lakh, to build. The number that breaks budgets is not the build: it is core system integration, the human review desk and the audit evidence a regulated deployment needs from day one.
Choose the partner who can show you a regulated system they built, name the controls it passed, and explain what they refused to automate. In banking and insurance, domain experience means audit trails, data residency and a tested failure path, not a slide about financial services.
A healthcare AI budget has three parts: $6,250 to $10,500 to prove the hardest case, $12,500 to $70,000 to build it, and a monthly running cost dominated by integration and support rather than model calls. In rupees, roughly ₹4 lakh, ₹8 lakh to ₹46 lakh, and ₹68,000 a month upwards.
Modernise. A full replacement of a core banking or policy administration system is justified in roughly one case in ten: when the vendor has ended support, the platform cannot meet a regulatory deadline, or the business model has changed. Everything else is better served by incremental work.
Most retail and ecommerce AI builds land between $12,500 and $70,000, or roughly ₹8 lakh to ₹46 lakh, plus a monthly running bill in the hundreds to low thousands. A support agent sits at the bottom of that range; a personalisation engine across catalogue, email and WhatsApp sits at the top.
Modernise around the clinical systems and replace only the administrative ones. A hospital cannot go offline for a cutover weekend, so the safe pattern is an interface layer over the HIS, LIS and PACS you keep, with replacement reserved for modules where a rollback is survivable.
AI in logistics costs $12,500 to $56,000 to build, or roughly ₹8 lakh to ₹38 lakh, depending on whether you are automating customer updates, dispatch exceptions or multilingual voice. The number that decides whether it was worth it is cost per shipment, not the build price.
Modernise incrementally in almost every case. Retail runs on a chain of systems where the storefront, order management, warehouse and finance all have to agree, so a full replacement means a simultaneous cutover across all of them. Strangle one bounded capability at a time and keep trading.