Cost guide

How much does an AI chatbot cost in 2026?

A chatbot build on one queue runs $500–$8K at the direct rate, quoted fixed against a written scope: the questions in bounds, the sources it answers from and the systems it touches. The monthly tuning and hosting retainer is priced on its own below, and every figure becomes a written proposal within 48 hours.

Request this option See the full rate card

Direct rate, in ₹ or $ · Itemised proposal in 48 hrs

The range

What it costs,
at the direct rate.

Priced as a build on one queue, grounded in your own documentation, and a monthly support retainer that keeps it grounded as the documentation changes.

ProjectChatbot build$500–$8Kfixed quote, one-time
RetainerMonthly tuning and hosting support$200–$1Kper month

Indicative: every figure is refined into an itemised proposal within 48 hours.INR figures shown excl. GST.

Every option includes a Prism workspace for requests, approvals, tasks and the weekly review.

What moves the price

Why the range
is a range.

How far up the ladder it has to climb

Chatbot questions sit on a ladder. The bottom rung is answerable from your documents alone: a refund window, a delivery policy. The next rung needs this customer's own record, so the assistant must verify who is asking before it reads a live system. The rung above changes something, which needs write access, a confirmation step and a reversal path. Every demo answers the bottom rung. The build is priced on the highest rung your queue needs.

Biggest lever

The state of your documentation

The assistant answers only from what your documentation says, so its condition sets how much of the build is preparation. A current help centre with one voice indexes cleanly. A policy library that contradicts itself, or answers that live in your agents' heads, must be reconciled first. The build's audit finds the gaps. Writing what is missing is your team's work or its own line in the proposal, never a surprise inside the fee.

Hidden lever

Which systems it must reach

The build wires the assistant into the helpdesk, CRM, website or messaging channel you already run. A helpdesk with a clean API and a queue already tagged by topic is the bottom of the range. A system with no API, or one where a customer must be identified before anything can be read, adds a layer to design and test. We size this from your stack in discovery.

Integration lever

What the queue could get wrong

Before a customer sees it, the assistant runs against an evaluation suite built from your real tickets: grounding, tone, refusals and resistance to prompt injection, the industry's term for a user talking it out of its rules. A queue of shipping questions needs a short suite. A queue that touches refunds, health, money or contracts needs a longer one with stricter refusal rules. The difference is visible in the proposal.

Risk lever

Channels and queues

The build is one queue on one channel, the smallest scope that can be evaluated properly. A second channel is not a copy: each has its own conversation shape, handover path and tests. Further channels and queues are priced after the first has earned them, so the pilot never carries the cost of a rollout you have not decided on.

Scope lever

Volume and how often the docs change

The monthly retainer is sized on conversation volume and how often the documentation underneath it changes. Every change means re-indexing and a re-run of the evaluation suite, so a policy library revised quarterly and a help centre edited daily are different retainers. Model usage is never inside either number: the provider bills it, and the retainer passes it through at cost.

Retainer lever

The options

Which option
fits your case.

Each option below is one way to buy AI chatbots. Read what it is and who it fits, then request the one that matches your case; the proposal itemises it.

Project

Chatbot build

$500–$8Kfixed quote, one-time

What it is

A support or sales assistant on one queue, answering from your help centre, policies, manuals and past tickets. Every answer shows its source, and the assistant hands over to a person when confidence drops. The layer underneath is retrieval-augmented generation, or RAG: find the relevant passages first, then write only from them. The fee is fixed against a written scope you approve before code ships.

Who it fits

Teams with a queue of questions their documentation already answers, and a helpdesk to plug into. It does not fit a business with no documentation to ground in, which needs the writing first, or a buyer who wants the assistant to improvise beyond its sources.

Included

Retrieval over help centre, policies, manuals and tickets; no source, no answer
Escalation on confidence thresholds, handing over with the full thread attached
Evaluation suite from real tickets: grounding, tone, injection resistance, refusals
Integration with the helpdesk, CRM, website or messaging channel you already run
Conversation logs, runbook, admin training and full ownership of the code

Not included, quoted separately

Model and API usage
Hosting
Additional channels or queues, priced after the pilot has earned them
Ongoing tuning (see the support retainer)

Turnaround: 3 to 6 weeks; working pilot on your docs within 14 days of NDA
Moves the number: How far up the containment ladder it climbs: answering from documents, reading a verified customer's record, or writing to it.

Request this option →

Retainer

Monthly tuning and hosting support

$200–$1Kper month

What it is

The retainer that keeps a live assistant accurate after launch: re-indexing as documents change, escalation thresholds tuned on what the logs show, model upgrades tested before they go live, hosting and monitoring managed, and a monthly report of what it answered, what it escalated and why.

Who it fits

Any team whose documentation changes, which is nearly every team. It does not fit a build you will run in-house from the runbook, and it is not a way to buy new channels, which are priced as their own lines.

Included

Re-indexing as your documentation changes, with the evaluation suite re-run each time
Escalation thresholds tuned on what the conversation logs show
Model upgrades tested before they go live
Hosting and monitoring managed, inside your own cloud where required
A monthly report: what it answered, what it escalated, and why

Not included, quoted separately

Model and API usage, passed through at cost
Hosting fees
New channels or queues

Turnaround: Monthly, 3-month minimum
Moves the number: Conversation volume and how often the documentation changes underneath it.

Request this option →

Cheaper or dearer

What moves a quote
down, or up.

The proposal itemises scope, integrations and evaluation, so each item below is a line you can see move before you sign.

Brings the quote down

Questions on the bottom rung: answerable from documents, no customer record needed
A help centre that is current, speaks with one voice and says what your agents say
A helpdesk with a clean API and a queue already tagged by topic
One channel to start: the web widget or the helpdesk, not both at once
Real tickets to build the evaluation suite from

Pushes the quote up

Answers that need a verified customer's record, which adds an identity boundary
Write actions such as cancelling or refunding, each with a confirmation and a reversal path
Policies that contradict each other, so the audit ends in a rewrite before the index
Regulated subjects with strict refusal rules and a longer evaluation suite
Deployment inside your own cloud, with residency and retention rules to honour

What you get

What the fee
actually buys.

The fee buys a working assistant on one queue and the evidence that it behaves. No deflection rate, containment figure or ticket count is promised, because your queue is not an average; what ships and how it is measured are.

Answers with their sources

Every reply links the passage it came from, so anyone can check it. When retrieval comes back empty, the assistant says so rather than guessing.

Handover with the thread attached

Confidence thresholds decide where it stops. Refunds outside policy, legal questions and upset customers reach a person with the full conversation and document attached.

An evaluation report you sign

Scores on grounding, tone, refusals and injection resistance, from a suite built on your real tickets. You see them before launch and again whenever the documents change.

Integration with what you run

The helpdesk, CRM, website or messaging channel you already use, with the model chosen on fit, cost and residency and swappable without a rebuild.

Logs you can audit

Every conversation is recorded with the source it used and the reason it escalated, so a complaint can be traced to the exact answer and fixed at the root.

Ownership at handover

The code, prompts, pipelines and runbook are yours, with admin training included. Updating the knowledge base is your team's job, not a change request.

AgenciesAgency partners buy at the partner rates on agencies.pixelcrayons.com; volume terms are agreed under NDA.

Request the partner card →

Questions

Frequently
asked.

Because the work is in the queue, not the conversation. A queue is one body of documentation, one set of escalation rules, one integration and one evaluation suite, built once whether the assistant handles a handful of chats or a flood. Per-conversation pricing would charge you for volume that costs us nothing and hide the real work. Volume shows up where it belongs: the provider's usage bill and the size of the retainer.

No, and be wary of a quote that says it is. Model and API usage is billed by the provider on your own account, so you can see it, cap it and switch models without asking us. The build fee buys the engineering: retrieval, escalation, integration, evaluation and handover. The retainer passes usage through at cost and reports it beside the conversation figures.

No. The handover includes the runbook, admin training and ownership of the code, so a team can run the assistant in-house. Most keep the retainer because documentation changes, and every change should trigger a re-index and a re-run of the evaluation suite. If your documents are stable and your team can own that loop, the proposal says so.

It is priced as its own line, after the first queue has earned it. A new channel brings its own conversation shape, handover path and tests; a new queue brings its own documentation and escalation rules. The proposal shows the addition separately, so you can sequence it on the pilot's evaluation results rather than pay for a rollout the first queue has not yet justified.

Because the step from documents to records is the real step change. An assistant answering from documents touches no live system. One that answers about this customer's plan must verify who is asking, read the system of record with scoped access, and know what to do when identity is uncertain. That is a login boundary, an integration and a set of failure rules to design and test, and the layer a vendor demo never shows.

Often, for the bottom rung, and discovery says so in writing when it is true. A helpdesk vendor's own assistant can answer from your help centre with little setup. What it usually cannot do is read customer records with scoped access, escalate with the full thread attached, or be evaluated against your real tickets. If your queue stays on documents and your helpdesk has a good native tool, we tell you to use it.

Other cost guides
and the service itself.

How much does an AI agent cost in 2026?Cost guide: the options, what moves the price and what the fee buys.How much does a website cost in 2026?Cost guide: the options, what moves the price and what the fee buys.About AI chatbotsThe service page: what the work is, how it runs and the proof behind it.AI chatbots on the rate cardThe same ranges beside every other service, in one card.

Last updated 4 September 2026

What it costs,at the direct rate.

Why the rangeis a range.