Build a Custom Voice AI for Your Freight Forwarding Team
The best voice AI for a freight forwarder is a custom system built on a large language model. It connects directly to your TMS and ERP to give real-time shipment updates.
We built a voice agent for a 15-person freight forwarder handling 200 daily status-check calls. The system went live in 4 weeks and now automates 85% of those calls. This freed up two operations staff to focus on complex shipments and new bookings.
The project scope depends on the systems it must read from and the query types it handles. A system that only checks shipment status in CargoWise is a simple build. One that also generates quotes by pulling rates from multiple carrier portals is more complex.
What Problem Does This Solve?
Most freight operations start with a standard IVR phone tree. This system can route calls but cannot answer questions. A customer calling for a tracking update must still wait for a human agent, which defeats the purpose of automation and ties up staff.
Off-the-shelf voice AI platforms from major telephony providers seem like the next step. However, they are trained on generic retail or support conversations. They fail when a customer provides a PRO number instead of a house bill of lading number. This ambiguity leads to incorrect lookups and customer frustration. These platforms also charge per minute, and a high volume of 30-second status calls quickly becomes more expensive than a full-time employee.
A 12-person forwarder we worked with tried a well-known AI contact center solution. Its transcription model had a 30% error rate on logistics-specific acronyms. Customers were constantly escalated to human agents, who then had to re-ask for the tracking information. The manual workload barely changed, but the company had a new $1,200 monthly software bill.
How Does It Work?
We start by connecting directly to your Transportation Management System (TMS) and other data sources using their APIs. We use the Claude API's function-calling feature to define the specific actions the AI can perform, like `get_shipment_status(tracking_id)` or `lookup_container_eta(container_id)`. This creates a reliable bridge between natural language and your internal systems.
The core of the solution is a Python application built with FastAPI. It receives calls through a Twilio phone number, which passes the audio stream to our service. We use a real-time transcription service to convert speech-to-text. That text is fed to a Claude 3 Sonnet model with a system prompt engineered specifically for freight forwarding terminology, ensuring it understands the context of your business.
The entire process from receiving audio to generating a text response takes under 500 milliseconds. The model's response is converted back to audio using a low-latency text-to-speech engine. The complete conversational turn completes in under 2 seconds. The application is deployed on AWS Lambda, so it scales automatically and typically costs less than $50 per month to run.
Every call is transcribed and logged to a Supabase database. We provide a simple dashboard to track call volume, automation rates, and common failure points. If the rate of calls escalated to a human agent rises above 15% for more than an hour, an automatic alert is sent via Slack, allowing us to quickly diagnose and fix the issue.
What Are the Key Benefits?
Live in 4 Weeks, Not 6 Months
From TMS integration to your first live customer call in 20 business days. Automate over 80% of routine status calls by next month.
One-Time Build, Not Per-Minute Billing
A single fixed-price project with low, flat monthly hosting costs. Your bill is predictable and does not penalize you for high call volumes.
You Own the Source Code
You receive the complete Python codebase and all system prompts in your company's GitHub repository. There is no vendor lock-in.
Monitors Itself for Errors
The system sends a Slack alert if it fails to resolve an unusual number of calls, allowing for prompt tuning of its logic.
Connects Directly to Your TMS
Direct API integration with systems like CargoWise, Magaya, or custom-built platforms gives the AI the exact same data your team uses.
What Does the Process Look Like?
API Access & Call Review (Week 1)
You provide read-only API credentials for your TMS and 30 days of call logs or support tickets. We use this to define the exact scope and deliverables.
Core Logic & Prompt Engineering (Week 2)
We build the FastAPI service and write the Claude API prompts. You receive sample call transcripts showing how the AI handles your specific queries.
Telephony Integration & Testing (Week 3)
We provision a phone number and connect it to the agent. Your internal team tests the system by calling in with real tracking numbers and requests.
Launch & Handoff (Week 4+)
The system goes live. After a 30-day monitoring period, we deliver the final source code, documentation, and a runbook for maintenance.
Frequently Asked Questions
- How much does a custom voice AI project cost?
- Pricing is a fixed-scope based on complexity. A system for shipment status lookups is a 3-4 week build. Adding dynamic quote generation or booking capabilities can extend that to 5-6 weeks. The final cost depends on the number of TMS or ERP integrations. We provide a fixed-price quote after our initial discovery call.
- What happens if the AI can't answer a customer's question?
- The system is designed to escalate gracefully. If it fails to understand a request after two attempts, or if the caller says 'agent', the call is transferred to your main operations line. A full transcript of the AI's conversation is immediately emailed to your team, giving the human agent who takes over complete context.
- How is this different from using a service like Twilio Flex?
- Twilio Flex is a toolkit for building contact centers; it is not an AI agent itself. We use Twilio for the underlying phone number management but build the actual intelligence with Python and the Claude API. This gives you a more powerful, custom-fit solution and avoids being locked into a single platform's proprietary AI.
- What if our TMS doesn't have a modern API?
- This is common with older industry software. If a direct REST or SOAP API is not available, we can often establish a read-only connection to the underlying database (e.g., SQL Server). If no electronic access is possible, this type of automation is not a good fit. We identify this in the first discovery call before any work begins.
- Can the agent handle heavy accents or different languages?
- Yes. The transcription models we use support over 50 languages and dialects. During testing, if we find that specific accents common among your customers have lower accuracy, we can tune the system's prompts to better recognize industry terms. Modern transcription services handle most accents with over 95% accuracy out of the box.
- How is our sensitive shipment data handled?
- The entire system is deployed within your own cloud infrastructure (AWS) or an account we create and hand over to you. Syntora never stores your customer or shipment data on our systems. All connections are encrypted, and because you own the source code, you are free to conduct your own security audits.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call