Build the knowledge base.
Add company information, product descriptions, and technical details that the AI can use during the conversation.
USEFUL PROJECT
Sales Helper.
An AI assistant that follows your call plan and helps you answer customer questions as they come up.
v2 · In client testingVersion 2 is with the client for active testing and final refinements.
THE STORY BEHIND THE PROJECT
How it started
The idea for Sales Helper grew out of my work with Layher Kft. Early in our collaboration, they asked how AI could support their sales team.
Their extensive range of scaffolding systems came with a lot of technical detail. When a customer asked about a particular component’s dimensions or specifications, finding the right information during the conversation could be a challenge.
I proposed an assistant that could follow the call, keep track of the conversation plan, and help the salesperson find relevant product information when a question came up. They liked the idea, and I developed it into Sales Helper.
FROM SETUP TO SUMMARY
How it works
Setup
Before the callSet the context
Define call scenarios.
Create a library of call plans, each with its own questions and talking points.
Prepare battle cards.
Create quick-reference material for recurring questions, objections, and other common sales situations.
Live
During the callFollow the conversation
The call plan and customer questions stay in view as the conversation unfolds.
Track questions as they are asked.
The system follows the conversation and updates the call plan when the salesperson asks a planned question.
Recognize answers, even out of order.
If the customer answers a question before it has been asked, the system can recognize that and update the corresponding item.
Help answer customer questions.
It detects questions from the customer and generates suggested answers using the configured knowledge base.
Surface relevant battle cards.
When a matching situation comes up, the corresponding reference material appears in the dashboard.
Review
After the callKeep a useful record
Summarize the conversation.
Get a written summary of the call, with the key discussion points and agreed next steps.
Revisit the details.
Return to saved calls to review the transcript, analysis, and summary.
WHAT MADE IT HARD
Key challenges
A dashboard that runs itself.
During the call, the salesperson shouldn’t have to click a single thing. The interface decides on its own what to bring into focus, what to show, and what to move into the background.
Speech-to-text that understands Hungarian.
Hungarian is the primary language, and most speech-to-text models struggle with it. After extensive experimentation, ElevenLabs came out on top.
Built for more than one language.
The interface is available in Hungarian and English, and calls can run in 26 conversation languages. Call scenarios can be translated into the chosen language, with AI assistance if needed.
THE TECHNOLOGY BEHIND IT
Under the hood
- 1 TRANSCRIBE
Scribe v2
Turns the salesperson’s and customer’s separate audio tracks into text.
- 2 EVALUATE
TypeSafeJev
Evaluates call-plan progress, detects customer questions, and identifies matching battle-card situations.
- 3 WRITE
OpenAIGPT Luna
Writes notes and suggested answers using the conversation and the configured knowledge base.
Find this interesting? Let’s talk.
Whether you’re hiring or exploring what AI could do for your team, I’d be glad to hear from you.
Write to me