How we deployed an AI chatbot that handled 70% of queries without a single employee

The customer distributed industrial parts to around 200 business buyers. Every day they received 30 to 50 queries about order status, spare part availability, pricing and delivery times. Two sales reps were spending an average of three hours a day handling these queries instead of actually selling. They came to us last summer.

Where we started: an audit without sugar-coating

Every project with us begins with an audit. Not to make the process look formal, but because deploying a chatbot without understanding the data makes no sense.

In this case we found that 68% of all incoming queries fell into four categories: order status, availability of a specific part, pricing for repeat orders and delivery time. The rest were sales enquiries or complaints. That remainder belonged to people. Everything else we could automate.

The customer had their data in the Pohoda accounting system and handled communication via email and phone. A chatbot was the right solution. We connected it to Pohoda via API so it could see live order statuses and stock levels in real time.

Deployment: six weeks from audit to go-live

The entire project from the first audit to live production took six weeks. Two weeks for data preparation and integration, two weeks of testing with real scenarios and two weeks of gradually migrating customers to the new channel.

The most demanding part was not technical. It was convincing the sales reps that the chatbot was not there to replace them. It helped when they saw the first reports: the chatbot had taken over the routine, but commercial calls and complex complaints were still coming to them. Suddenly they had time for the customers who genuinely needed their attention.

Results after three months

Numbers after 90 days in production:

• 70% of queries handled by the chatbot without operator involvement
• Average response time dropped from 2.5 hours to 8 seconds
• Sales reps recovered over 40 hours per month
• Customer satisfaction improved: customers praised 24/7 availability

What surprised us: the chatbot started revealing query patterns we had not expected. It turned out that a large share of customers were repeatedly searching for a specific type of part that the customer stocked but was not displaying correctly in the catalogue. That insight cost the customer one hour of work in their system and likely generated new sales.

What this means for other businesses

A chatbot is not a magic button. It is a tool that works exactly as well as the data it runs on is prepared. This project worked well for three reasons: the customer had structured data in their system, they had clearly defined repeating scenarios and they were willing to go through the audit process without trying to shortcut it.

A company that wants to deploy a chatbot in a week without an audit either does not have its data in order and the chatbot will give wrong answers, or has such a simple situation that it does not need a chatbot at all.

Want to know what a similar project would look like for you? Start with a free audit. We will walk through your data and processes and tell you honestly what a chatbot can and cannot handle in your company.