AI chatbots for customer service that do not embarrass you

The demo answers three questions perfectly. Production asks four thousand, half of them badly worded and some of them about refunds. We build the second kind.

Custom software SaaS products AI development Web applications AI chatbots Nearshore teams

Overview

What "production chatbot" has to mean

A support bot is only as good as its grounding, its permissions and its refusal path. Everything decisive here is invisible in a demo.

What a production support chatbot actually is

A support chatbot is not a chat window with a language model behind it. It is a feature that reads your help centre, policies and order data, answers only what that customer is allowed to see, refuses when it does not know, and hands the conversation to a human with context attached.

Everything that makes it survivable — grounding, permissions, evaluation, cost caps — is invisible in a demo and decisive in production. That is where the budget goes.

When it pays for itself

When ticket volume repeats: the same forty questions in different words, all day. Deflection is measurable within a week because the mix is so predictable. If your tickets are all unique, automation will not help and we will say so.

Signals your support queue is automatable

the same questions arrive dozens of times a day in slightly different words;
agents copy answers out of a help centre nobody reads;
answers depend on data your bot could look up — orders, plans, shipments;
first-response time is the metric your customers complain about;
seasonal spikes force you to hire and then let people go.

Signals that a support queue can be automated: repeating questions, lookups, first-response pressure

Price

From $10,900

Timeline

3–5 weeks

Payment

50/50 by milestone

Anatomy

What separates a deflection engine from a toy

Five things, none of which show up in a demo — and all of which decide whether you can leave it switched on.

Built in from the start

Grounding. Answers come from your help centre, policies and order data, with citations an agent can check.

Permissions. The bot sees exactly what that customer is allowed to see — the same rules as your app, enforced server-side.

A refusal path. "I do not know, here is a human" is a correct answer, and we measure how often it fires.

Clean handover. Full conversation, detected intent and customer context land in your helpdesk, not a fresh blank ticket.

Evaluation in CI. A labelled set built from your real tickets; a prompt change that lowers accuracy does not merge.

Price

From $10,900

Timeline

3–5 weeks

Inference

At cost, capped by your budget

Where it pays

The boring 40–60% of tickets

Volume repeats: the same questions, in different words, all day. That is exactly what this is good at.

E-commerce

Order status, returns, sizing, stock. Deflection here is measurable within a week because the ticket mix is so repetitive.

SaaS

Onboarding questions, billing, plan limits. The bot can read the account it is talking about, which is what makes the answers useful.

Logistics

Shipment status and exception explanations — the questions that currently interrupt a dispatcher twenty times a day.

Real estate

Listing questions and viewing scheduling, with handover to the agent who owns the deal rather than whoever is on shift.

Pricing

What it costs

Published, because a support bot is a budget line and you need the number before the business case.

Criterion

Agency quoteOur engagement

Cast & ShipTypical agency

Ticket analysis and feasibility

$8,000–15,000 discovery

$4,490, one week

Support chatbot on your data

$25,000+

From $10,900, 3–5 weeks

Chatbot with actions inside a product

$40,000+

From $18,900, 5–8 weeks

Evaluation harness handed over

Rarely included

Always included

Inference costs

Marked up

At cost, capped

Cost

What an AI chatbot costs

Inference is billed at cost against a cap you set — it never turns into a margin we quietly keep.

Pilot on one flow

$4,490

One intent end to end, with an evaluation set, so deflection is measured

1–2 weeks

Why teams choose us

Four things we can prove, not four adjectives.

Evaluation, not vibes

A labelled set built from your real tickets, scored in CI. A prompt change that lowers accuracy does not merge — you get the harness, not just the bot.

Refusal designed first

"I do not know, here is a human" is a correct answer. We measure how often it fires and tune the threshold with you.

Costs capped in code

Token budgets and model routing are enforced by the application, so the monthly bill cannot surprise you.

Clean handover to humans

Full conversation, detected intent and customer context land in your helpdesk — not a blank ticket that makes the customer repeat themselves.

Industries

What we buildfor your industry

Support automation pays fastest where ticket volume repeats: e-commerce, SaaS, logistics and property management.

18 sectors

IT companies and digital agencies
Insurers and brokers
Lenders and microfinance
E-commerce
Wholesale and distribution
Manufacturing
Freight and forwarding
Clinics and medical centres
Dental
Car dealers and service centres
EdTech
Legal and consulting firms
Equipment rental
Real estate agencies
Property developers
Travel and events
Beauty and wellness
Construction and renovation

FAQ

Chatbot questions

What support and product leads ask before committing.

How much does an AI chatbot cost?

From $10,900 to build. Running cost is inference plus hosting — typically tens to low hundreds of dollars a month at moderate volume, and we model yours during the feasibility week.

How accurate will it be?

We do not guess: accuracy is measured on a labelled set built from your own tickets, and we show the number with the refusal rate next to it before launch.

What happens when it does not know?

It says so and hands over to a human with the full context. That path is designed first, not last.

Can it take actions, not just answer?

Yes — order lookups, refunds within policy, scheduling — with approval gates and an audit trail. That is the $18,900 tier.

Will our data be used for training?

No. Your data is used for grounding, not training, with zero-retention configurations by default.

How is this different from the bot in our helpdesk?

Off-the-shelf bots answer a FAQ page. This one answers questions that need your data and your rules, and it gets evaluated like software rather than trusted like a demo.

2–4×

faster on the code-writing share

in 3 weeks

from idea to MVP

230+

projects delivered

17 years

of engineering experience

Work