One kind of chatbot invents an answer about your refund policy. The other reads the policy. That difference is what separates a demo from something you can put in front of customers.
The word chatbot covers three very different products at three very different prices, and most disappointment comes from buying the cheapest and expecting the most expensive. A scripted bot follows a decision tree. A raw language model answers confidently from general knowledge, including about your business, which it does not have. A grounded agent retrieves your actual documents and answers from those.
Only the third is safe to point at customers. Grounding — retrieval-augmented generation — means the agent looks up your policies, your pricing, your intake forms and your FAQs before answering, and cites what it found. When it does not find an answer, it says so and hands over to a person rather than improvising.
Guardrails matter as much as capability, particularly in regulated or high-stakes contexts. The agent must know what it must never answer, and every one of those cases needs a clean human escalation path. We define that boundary explicitly during scoping, because the failure mode of a confident wrong answer is worse than no answer at all.
Build cost and running cost are two different numbers, and conflating them is the most common budgeting surprise. An agent has a one-off build and an ongoing per-conversation cost that scales with usage. We quote both, because only knowing the first is how projects get abandoned in month three.
Fixed price, fixed scope, agreed in writing before anything starts. You own everything on day one.
What the agent handles, and — just as important — what it must always escalate to a person.
Your policies, FAQs, product data and documents cleaned and structured for retrieval. Most of the quality comes from this step.
pgvector-backed retrieval so answers come from your documents rather than the model's general knowledge.
Anthropic Claude or OpenAI chosen for the job, balancing answer quality against per-conversation running cost.
A clean path to a human with the conversation context attached, so the customer does not start over.
On your site, on WhatsApp, or both — whichever channel your customers already use.
Adversarial testing against the cases the agent must refuse, before it goes anywhere near a customer.
Build cost and projected running cost quoted separately, with usage monitoring after launch.
* Indicative starting prices in USD. Final quotes scoped per project — VAT handled for UAE & EU clients.
An agent earns its cost when the same questions arrive repeatedly and the answers genuinely exist in writing somewhere.
We will talk you out of it in these cases:
AI agents start from $2,000 to build, an indicative starting price in USD. Separately from the build, there is a running cost per conversation that scales with usage. We quote both numbers, because budgeting only for the build is the most common way these projects get abandoned a few months in.
RAG — retrieval-augmented generation — means the agent retrieves your actual documents before answering, rather than generating from general knowledge. It is the difference between an agent that invents an answer about your refund policy and one that reads the policy. Only the grounded kind is safe to put in front of customers.
Grounding plus explicit guardrails is how you prevent that. The agent answers from your documents, and where it cannot find an answer it says so and escalates to a person. We test adversarially against the cases it must refuse before it goes live — that testing is part of the build, not an extra.
Yes. For a lot of businesses in this region WhatsApp is not a channel but the primary sales and support interface, so we deploy there as readily as on a website — and often to both.
It can, and whether it should is a real question rather than an automatic yes. For consumer-facing businesses a meaningful share of customers are more comfortable typing in Tamil, particularly older and first-time online buyers. We will give you an honest read on whether it is worth it for your specific audience.
It hands over to a person with the conversation context attached, so the customer does not have to explain themselves twice. Defining exactly when that handover triggers is part of scoping, and in regulated contexts it is the most important part.