Customer support · AI business automation
Roof Maker's phone kept ringing after hours. Nobody was there to answer it.
AI voice agent for customer support
Calls came in around the clock, but the team could only pick up in business hours, so after-hours enquiries went to voicemail and some became lost sales. During the day, specialists lost hours to routine product and warranty questions. Softblues automated the front line: a voice AI that answers every call, troubleshoots, logs a HubSpot ticket, and escalates anything urgent to a person. 24/7, in production.
The problem
A support line is only as good as whoever's free to answer
Roof Maker makes premium rooflights and skylights, and its support line was outgrowing the team. Calls arrived at all hours, but they could only be answered in business hours. After-hours enquiries went to voicemail, urgent issues waited, and some callers simply went elsewhere.
In the day it was the opposite problem. Technical specialists spent a large share of their time on routine product, warranty and troubleshooting calls that did not need an expert. The process worked, but only while someone was free to pick up the phone.
A support line that depends on someone being free to answer isn't really open.
What makes a voice agent hard to get right
Sound human, and fast
A phone menu is easy. A natural conversation that answers in under two seconds, without the caller feeling like they are talking to a machine, is not.
Know when to back off
Some calls must reach a person. A reported leak or a frustrated caller cannot be left with a bot, so the agent has to read urgency and emotion, not just the words.
Leave a clean record
Every call has to become a structured, validated ticket, not a free-text note, or the CRM quality the business runs on slips.
Fit the stack already there
It had to work with the existing phone system and HubSpot, alongside the human team, without ripping anything out.
The build
One voice, eight agents behind it
Softblues automated the front line of Roof Maker's support. A voice AI picks up on the first ring, holds a natural conversation, diagnoses common rooflight issues, answers product and warranty questions, and creates a HubSpot ticket automatically. It listens for urgency and emotion, so a leak or a distressed caller is escalated to a person straight away, and in business hours it hands off to a live agent with the full conversation in context.
The voice pipeline runs on Aircall for telephony, Deepgram Nova 3 for speech recognition, GPT-4.1-mini for understanding and conversation, and ElevenLabs Flash v2 for a natural reply, with round-trip latency under two seconds. Behind the single voice is a team of eight specialised agents, each owning one part of the call.
How it works
Eight agents behind one voice
- 1
Conversation flow
Runs the call from greeting to wrap-up and keeps it sounding natural.
- 2
Technical troubleshooting
Diagnoses common rooflight and skylight issues over the phone.
- 3
Information delivery
Answers product and warranty questions from the company's own knowledge.
- 4
Classification
Works out what the call is about and how urgent it is.
- 5
Escalation detection
Listens for urgency and emotion and routes leaks or distressed callers to a person.
- 6
Data collection
Captures the caller's details in a structured, validated form.
- 7
Ticket creation
Logs a complete HubSpot ticket for every call, automatically.
- 8
Follow-up
Closes the loop with the customer after the call, so nothing is missed.
Built with
Telephony & voice
Reasoning
Integrations
Tooling & monitoring
Results
What it delivers
Every inbound call answered, day or night
Specialised agents behind one voice
Speech recognition accuracy
Response time, so the conversation flows
Specialist time freed
Routine and after-hours calls are handled automatically, so technical specialists get their time back for the complex cases.
Urgent calls always reach a person
Emotion and urgency detection makes sure a leak or a frustrated caller is escalated immediately, with the call context attached.
Where else this works
The pattern is general: a voice agent that answers, understands, resolves the routine, logs it to the CRM, and escalates the rest. It fits any business with an inbound support or enquiry line, from field services to equipment, utilities and healthcare bookings. The same agent approach runs across our other builds too.
Frequently asked questions
The front line of their customer support. A voice AI answers every inbound call, troubleshoots common rooflight issues, answers product and warranty questions, logs a HubSpot ticket automatically, and escalates anything urgent to a person. It runs in production, 24/7, alongside the existing phone system.
No. It handles the routine and after-hours calls so specialists do not have to, and it hands off complex or urgent calls to a person with the full conversation in context. The aim is to free the team for the work that needs a human, not to remove them.
It listens for urgency keywords, issue complexity and emotional tone. A reported leak, a complex technical problem or a frustrated caller is routed to a person immediately. During business hours it hands off to a live agent with the call context attached.
Speech recognition runs at around 97% accuracy and responses come back in under two seconds, so the conversation flows rather than stalling. It uses real-time recognition and lifelike voice synthesis, not a rigid phone menu.
Every call creates a structured, validated HubSpot ticket automatically, with a conversation summary stored for the team. CRM data stays consistent and complete without anyone logging calls by hand.
No. The pattern, a voice agent that answers, understands, resolves the routine, logs it to your CRM and escalates the rest, applies to any business with an inbound support or enquiry line: field services, equipment and utilities, healthcare bookings and more.
Does your support line go quiet after hours?
If routine calls tie up your team in the day and go unanswered at night, this is the work we do. Book a walkthrough and hear how it handles a call.
Explore Other Projects
Discover more AI solutions delivering measurable results across industries