For support, sales and ops teams
Custom Chatbot Development Services
A bot that answers from your own docs, knows when to get a human, and is tested on your real tickets before a customer ever sees it.
Custom chatbot development services build a bot around your products and your systems, so it can answer the questions your team hears 50 times a day. Most custom AI bots cost $30,000 to $120,000 with a typical agency. We start smaller, with a fixed-scope pilot that goes live in 2 to 6 weeks.
Book a call and we'll send a 1-page plan. English and Spanish bots.
2–6 wks
Fixed-scope pilot to a live bot
100%
You own the code, prompts, test set and chat logs
EN + ES
Bilingual bots, built by a bilingual team
25%
Less downtime in our shipped manufacturing AI project
The problem
Your best people spend the day answering the same questions
Where's my order. Do you ship to Canada. How do I reset my password. Your team answers these all day, copy-pasting from the same doc. Customers wait. The hard tickets sit at the bottom of the queue.
And the off-the-shelf bot you tried? It answered from a generic script, didn't know a thing about your accounts and sent people in circles until they gave up and called.
Cost of doing nothing
What those repeat questions cost you
Set the sliders to your numbers. The default share is 30%, which is what service teams told Salesforce AI already resolves for them in 2025 (they expect 50% by 2027).
Support savings calculator
Team time freed per year
$32,400
Tickets × minutes ÷ 60 × hourly cost × share resolved × 12. This is time your people get back for harder work. It doesn't count faster replies or sales at night, which are often worth more.
2/3
of Klarna's customer service chats were handled by its AI assistant in its first month, 2.3 million conversations.
34%
productivity gain for newer support agents who got AI help, in a study of 5,179 agents (14% on average).
64%
of customers would rather companies didn't use AI in service. So the bot has to be good, and a human has to be one click away.
What you get
A bot that knows your business, and the proof that it works
Knowledge
Answers from your sources
Actions
Looks things up for real
Handoff
A person, one click away
Testing
An accuracy score before launch
Channels
Where your customers already are
Ownership
Yours to keep
Your chatbot
Finds the right source, answers only from it, follows your rules, and logs every answer for review.
How it works
From your tickets to a live bot in 6 weeks
The order matters. We test on your real questions before anyone outside your company sees the bot, and that one step is where most rushed projects skip ahead and get burned.
- 01Week 1
Collect and score
Pull 300–500 past tickets. Find the questions that repeat. Write down the target accuracy. - 02Weeks 2–3
Build
Connect the knowledge sources and 1–2 systems. Set the rules: what it can say, what it hands off. - 03Week 4
Shadow mode
The bot drafts answers for your agents, privately. We fix what it gets wrong. - 04Weeks 5–6
Go live on one channel
Launch with handoff on. Weekly report on accuracy, handoffs and time saved.
Want to see your own accuracy number first?
The pilot is built around it.
Cost
What a custom chatbot really costs
The build is most of the bill. The AI model itself is cheap, often a few cents per conversation or less, and it's rarely the number that should decide anything.
Running costs add up, though. Hosting, monitoring, updating the knowledge base when your policies change. Budget about 15% to 20% of the build each year for that.
Build cost, 2026 market ranges
Top of each range shown
AI cost per conversation
Our estimate: ~20K input + 2K output tokens per chat
| SaaS bot tool | Typical agency build | Ikasu pilot | |
|---|---|---|---|
| How you pay | Monthly seat or per-resolution fee | Big upfront project | Fixed-scope pilot, then a subscription |
| Time to launch | Days | 2–8 months | 2–6 weeks for a first version |
| Uses your private data | Some, through their connectors | Yes | Yes |
| Looks things up in your systems | Limited to their integrations | Yes | Yes, starting read-only |
| Tested on your real tickets | Rarely | Sometimes | Always, before launch |
| Who owns it | The vendor | Depends on the contract | You, from day one |
Proof
We ship AI that runs in production
We're a small studio in Mexico City, so we won't show you a wall of logos. What we can show is a real AI system we shipped, with real numbers. It's a predictive maintenance system. The hard parts were the same ones a chatbot has (messy data, people who need to trust the output).
25%
less unplanned downtime
$500K
maintenance savings
50
sensors unified
15
operators trained
A mid-sized manufacturer had data from 50 sensors and 10,000+ rows of maintenance history spread across sheets and systems. We pulled it into one place, trained 3 models to spot failures early, and gave the team one dashboard with alerts.
Read the full case studyWatch
How a bot answers from your own documents
Most custom chatbots use a method called RAG, short for retrieval-augmented generation. The bot looks up the right page first, then writes the answer from it. IBM explains it well in about 6 minutes.
Still deciding what to build? A short conversational AI consulting engagement maps your tickets and tools first.
Questions buyers ask
Straight answers before you book a call.
Send us 20 of your most common questions.
That's enough for us to tell you if a custom chatbot is worth it, what it would connect to, and what the pilot would cover.
Book a call and we'll send a 1-page plan. English and Spanish bots.