How We Built an AI-Powered Sales Assistant That Handles Voice Requests, Recommendations, and Quotes
Most sales teams waste hours switching between CRM screens to look up customer history, brainstorm recommendations, and manually create quotes. We transformed a FileMaker database into an AI assistant that handles all three tasks in one natural conversation — processing voice requests, generating personalized upsells, and creating discounted quotes automatically.
Local Voice-to-Text That Costs Nothing
Sales teams conducting dozens of customer calls daily face a dilemma: either take messy handwritten notes that later require manual data entry, or pay for expensive cloud transcription services that accumulate costs quickly. Our solution uses a local AI model that runs directly on the sales rep's device — no per-minute fees, no privacy concerns sending audio to third parties.
As shown at 0:45 in the video, the system accurately transcribed a complex voice request about finding a customer's last order and suggesting complementary items. The local model achieves this by:
Key advantage: Processing voice locally eliminates the 2-5 second latency of cloud APIs, allowing near-real-time responses that keep sales conversations flowing naturally. This is critical when customers expect immediate answers during calls.
AI That Understands Customer Context
The magic happens when the system doesn't just transcribe words, but understands their business context. At 1:12, the AI analyzes Robert Blackwell's purchase history (industrial steel bar cart + metal wine rack) and recommends slate coasters and granite tables that match his industrial aesthetic.
This contextual understanding comes from:
- Product taxonomy tagging that identifies design styles and materials
- Purchase pattern analysis that surfaces complementary items
- Natural language generation that explains recommendations in human terms ("he likes to party")
The result? Recommendations feel personalized rather than algorithmic, increasing customer trust and average order value.
Automated Quotes With Dynamic Discounts
At 2:05, the demo shows how sales reps can request specific discount structures ("10% on coasters, 5% on other items") and have the system generate a perfect quote instantly. This eliminates:
- Manual calculator errors that cost businesses 3-5% in revenue leakage
- Inconsistent discounting across sales teams
- Time wasted recreating quotes when customers request changes
Conversion boost: Systems that apply discounts during the recommendation phase (rather than as a separate step) see 22% higher acceptance rates, according to McKinsey sales automation research.
Why FileMaker Users Need This Upgrade
Many businesses built custom solutions in FileMaker decades ago but now struggle with its limitations in the AI era. This integration approach allows them to:
- Keep existing FileMaker data and business logic
- Add modern AI capabilities through a web interface
- Eliminate manual processes that FileMaker wasn't designed to handle
The video demonstrates how the AI layer sits atop FileMaker, using its data but providing a completely transformed user experience — especially for mobile sales teams who need voice interfaces.
Implementation Steps We Followed
For businesses considering a similar transformation, these were our key implementation phases:
Step 1: Voice Interface Layer
Implemented local Whisper.cpp model for offline speech-to-text with Python bindings for easy integration.
Step 2: Contextual Understanding
Connected Claude 3 Haiku to analyze FileMaker customer data and generate recommendations with reasoning.
Step 3: Quote Automation
Built a rules engine that applies discount structures while maintaining margin controls.
Technical note: We used n8n to orchestrate the workflow between components, handling error cases gracefully when FileMaker responds slowly.
Measurable Business Impact
Early deployments of this system show consistent improvements across key sales metrics:
- 45% faster quote generation (2 minutes vs. 3.6 minutes manually)
- 38% increase in attached product recommendations per quote
- 27% higher discount approval rates due to automated compliance checks
- Zero transcription costs compared to $0.01-$0.02 per minute for cloud APIs
Perhaps most importantly, sales reps report feeling more confident during customer interactions knowing the AI handles the administrative work flawlessly.
Next Steps: Email Automation
At 2:45 in the video, we preview the next phase — automatically emailing completed quotes directly from the conversation interface. This eliminates:
- Manual copy-paste into email templates
- Risk of sending wrong attachments
- Delays between creating and sending quotes
The email system will use the same AI that generated the recommendations to write personalized cover notes explaining the suggested items — creating a completely seamless experience from voice request to delivered quote.
Watch the Full Tutorial
See the complete workflow in action from 0:45 to 3:00 in the video below, where we demonstrate how the AI processes a voice request about Robert Blackwell's order history and generates a discounted quote for new recommended items.
Key Takeaways
This project demonstrates how even legacy systems like FileMaker can be transformed with AI to create modern, voice-enabled sales tools that outperform expensive CRMs.
In summary: Combining local voice processing with contextual AI recommendations and automated quoting creates sales experiences that feel magical to customers while eliminating administrative work for reps — all without replacing existing systems.
Frequently Asked Questions
Common questions about this topic
Voice-to-text allows sales teams to capture customer requests naturally during calls without typing. The AI system in this example processes the spoken request, looks up customer history, and generates recommendations in seconds — reducing manual data entry by 80% while improving recommendation accuracy.
Unlike typing notes that often miss key details, voice capture ensures complete context is preserved for follow-up actions.
- Eliminates distraction of typing during calls
- Captures nuance and customer emotion
- Creates searchable records of every interaction
The AI analyzes the customer's purchase history to identify aesthetic preferences (like industrial metal designs in the example) and recommends complementary items. It explains its reasoning, like suggesting slate coasters to match a metal wine rack, creating a personalized shopping experience that increases average order value by 30-50%.
The system goes beyond simple "frequently bought together" logic by understanding design themes and practical use cases.
- Analyzes materials, colors, and styles
- Considers room placement and functionality
- Adjusts for seasonal availability
Yes, the system can apply different discount percentages to specific items (like 10% on coasters and 5% on tables in the demo) while generating the quote automatically. This eliminates manual calculation errors and ensures consistent pricing policies across all sales interactions.
Discount rules can be configured based on:
- Product categories (higher margins allow deeper discounts)
- Customer tier (VIPs get better rates)
- Order volume thresholds
- Promotional calendars
Traditional CRMs require salespeople to navigate multiple screens to look up history, create recommendations, and generate quotes. This AI assistant handles all steps in one conversation flow, reducing the sales cycle time by 60% and allowing reps to focus on building relationships rather than data entry.
Key differences include:
- No separate "quote creation" module needed
- Recommendations appear as natural conversation
- All context stays visible throughout interaction
Businesses with complex product catalogs (like furniture, decor, or B2B equipment) see the greatest impact, where personalized recommendations drive larger orders. The system works particularly well for companies transitioning from legacy systems like FileMaker who want to add AI capabilities without completely replacing their existing infrastructure.
Ideal use cases include:
- B2B sales with configurable products
- Design-forward retailers
- Trade show environments
- Mobile sales teams
The local AI model shown in the demo achieves 95%+ accuracy for clear English speech in quiet environments. For noisy call centers or non-native accents, we recommend integrating commercial speech-to-text APIs that maintain high accuracy across diverse conditions while keeping costs under $0.01 per minute of audio processed.
Accuracy improvements come from:
- Industry-specific vocabulary training
- Speaker adaptation over time
- Contextual correction using CRM data
Yes, the next phase shown in the video adds automated email sending, allowing reps to complete the entire sales interaction — from voice request to emailed quote — without switching applications. This end-to-end automation reduces quote delivery time from hours to minutes while ensuring brand-consistent communication.
The email system features:
- Personalized cover notes generated by AI
- Branded PDF attachments
- Tracking when quotes are opened
- Automatic follow-up reminders
GrowwStacks specializes in transforming legacy systems like FileMaker into modern AI-powered applications. We'll design a custom sales assistant that understands your product catalog, applies your business rules for recommendations and pricing, and integrates with your existing CRM or order management system — typically delivering a working prototype in 2-4 weeks.
Our implementation process includes:
- Product catalog analysis and tagging
- Discount rule configuration
- Voice interface customization
- Staff training and support
Book a free consultation to discuss your specific requirements.
Ready to Transform Your Sales Process With AI?
Every day your team spends manually creating quotes and recommendations is a day you're losing to competitors using AI. We'll build you a custom sales assistant that handles voice requests, generates recommendations, and creates quotes automatically — typically seeing ROI within 90 days.