Practical notes from the engineers who build agents, copilots and AI products.
AI customer service agent projects fail for five recurring reasons: they chase deflection instead of resolution, they launch with no evaluation suite, they cannot reach the systems holding the answer, nobody signs off the actions they take, and nobody owns them after go-live.
The agent itself barely changes between a startup and an enterprise; everything around it does. A startup ships one or two intents on one helpdesk in six to eight weeks. An enterprise spends as long on access control, data residency and change approval as on the agent itself.
Ask for the evaluation suite first. A vendor who has shipped will show a golden question set built from real conversations, name the metrics they report, state who owns the code and prompts, explain how running costs are billed, and describe what happens when a model is deprecated.
Buy when your support work is answering questions on a standard helpdesk, and build when it is completing tasks inside systems only you run. Most companies need both: a bought platform for the long tail of enquiries, and a custom agent for the five intents that carry the volume and the cost.
An AI customer service agent in India costs $12,500 to $42,000, or ₹8,00,000 to ₹28,00,000, to build with a specialist partner, plus your own model API usage and a care plan from $1,000 or ₹68,000 a month. Scope, integration count and language coverage move the number inside that band.
LLM application development costs $21,000 to $84,000, or ₹13,60,000 to ₹56,00,000, for a production build with Eazyware, plus your own model usage and an optional care plan from $1,000 or ₹68,000 a month. Scope, integration count and evaluation depth decide where in that band you land.
Measure an AI customer service agent on resolution rate, seven-day reopen rate, cost per resolved conversation, escalation quality and CSAT on agent-handled contacts. Deflection is not a metric: it counts conversations that ended, not problems solved. Track the five together, because each can be gamed alone.
An AI customer service agent is not compliant or non-compliant by nature; your deployment of it is. Under the DPDP Act, 2023 you need a lawful basis and notice for the conversation data, redaction before it leaves your boundary, a retention clock, erasure on request, and processor terms with every model vendor.
The ROI of LLM application development is the value of one workflow getting faster or more accurate, minus build, inference and ownership costs. It survives review only when the baseline was measured first, the benefit is a number a business owner already reports, and running costs are modelled over two years.