AI chatbot development
Krikir handles AI chatbot development for support and staff teams, answering from your own content.
Fixed quote after one callReply within one working day
Example: one customer asking about a return
What happens to one question
An AI chatbot answers from your approved content and passes the rest to a person. AI chatbot development suits teams with written policies and repeat questions, and it is the wrong choice if chats are few or a product already passes your test.
A question arrives
A customer asks in the chat box on your site, or a member of staff asks an internal knowledge assistant.
It searches your content
It reads approved pages, policies and documents at answer time, and only what that person may see.
It answers with a source
The reply points to the page it came from. Signed-in customers can ask about their own order.
It hands over
Anything unclear, upset or on your never-answer list goes to a person, with the chat summary attached.
It shows what it missed
Every chat is logged, and the questions it could not answer tell you which content to fix.
Buy a chatbot or build one?
Try a product on your own questions first, and build only for a reason you can name. Product details are as of November 2026.
You run on Shopify and most questions are about products, delivery, returns and order status.
It works in the store's chat widget only and needs new customer accounts.
Your help centre is well written and your team already works in a tool such as Tidio, Zendesk or Fin.
Billing runs per conversation or resolution, so read what counts. Fin treats 24 quiet hours as an assumed resolution.
Staff questions sit in documents you already keep there.
It follows existing permissions, so old oversharing comes through too.
Answers need your own data or several sources, or it must sit inside your site or product.
You carry running costs, content upkeep and retesting.
It must change something: a refund, a cancellation, a customer record.
A wrong action costs more than a wrong answer, so this is a separate build.
A sample test sheet
This sample was written for this page and is not from a client. Each row is one kind of hard question, and you approve the pass mark before testing starts.
Talk it through on a call
We look at what you have now and what you want it to do.
You hear how we would approach it and who would work on it.
You get a fixed quote in writing after the call, with nothing to sign until you are ready.
How a chatbot project runs
Founder and CEO Krishnkant Patidar leads a team of 10, and AI and ML developer Deepak Patidar brings ten years in this field. Earlier client work stays under NDA, so this page shows the method. Review notes reach you in writing, and call times show in your own time zone.
Read the real questions
We sort your tickets and chats by reason, then write the never-answer list with your team.
Try the cheapest route
Your real questions go through the product you already pay for. If it passes, the project can stop there.
Fix the content
Conflicting pages, old PDFs and unwritten rules get sorted out, and each source gets an owner.
Build the assistant
Search over your content, sourced answers, the AI notice, handover, logs and access rules, on a preview you can try.
Test on the sheet
Each row runs several times, because a model can answer the same input differently. Money, privacy and handover rows must all pass.
Release in stages
Staff use it as a drafting aid first, then it takes a share of real chats, then all of them.
Rules for every conversation
Set before launch, with hard limits in code.
It says it is AI
As of November 2026, EU AI Act Article 50 applies, and Commission guidelines name helpdesk chatbots as in scope. A line in your terms, or the word "assistant" alone, is not enough. We build the notice in, and your lawyer confirms the wording.
A person is always reachable
Asking for a person always works. In a Gartner survey released in August 2026, 87% of customers called that option essential.
It admits what it does not know
If your content holds no answer, it stops and offers a person. It never invents a date, a code or a promise.
The chat box stays light
A small button loads the full chat on demand, so your pages stay fast. It works by keyboard and screen reader, and it is designed to sit in your brand.
What you keep
Wrong answers, data and model changes
Vendor terms were read on 7 November 2026 and can change.
When an answer is wrong
We log it, correct it with the customer, fix the source and add the question to the test sheet. In Moffatt v Air Canada (2024), a Canadian tribunal held the airline to its chatbot's wrong advice, even though the right page was linked.
What is sent
The question, the passages found and the fields that answer needs. Data handling is agreed in writing before any real data is used.
Who keeps it
OpenAI keeps API abuse logs up to 30 days. Anthropic deletes API data within 30 days, and flagged content can stay up to 2 years. Google's paid Gemini API keeps it 55 days. None of the three trains on this data by default.
When the model changes
OpenAI states at least six months' notice for generally available models and less for some others, Anthropic at least 60 days, and Google no fixed period. The test sheet reruns before any switch.
Other pages for this work
Questions before you add a chatbot
Buying, wrong answers, people and data.
Book a 30-minute call
Start with a product if it passes your own questions, because that is the cheaper route. Build a custom AI chatbot when answers need data a product cannot reach, or when it must sit inside your site or product.
An AI chatbot can make up answers, so it is limited to your approved content and shows its source. It says when it does not know, and the test sheet includes questions it must refuse. We give no accuracy figure.
Yes: a Krikir chatbot hands over whenever someone asks, with the chat summary attached. When nobody is online, it says so and gives the real alternative.
In the EU, as of November 2026, yes in most customer-facing cases, under Article 50 of the AI Act. Rules elsewhere vary, and US rules differ by state. We build a clear notice into the chatbot, and your lawyer confirms the wording.
A chatbot sends each question, the passages found and the fields needed to your chosen model vendor. As of 7 November 2026, OpenAI, Anthropic and Google's paid API keep it for up to 55 days in normal use, depending on the vendor. Data handling is agreed in writing before any real data is used.
A chatbot is retested on its test sheet before any model switch, so answers do not change without your approval. Who runs the retest after handover is agreed in writing before launch.
Let's talk about your next project
Book a 30-minute call or send a brief. A founder replies within one working day.