How does the AI voice assistant tell a legitimate call apart from spam in real time?

It scores every incoming call against known-spam number databases plus real-time signals like caller ID reputation and call frequency before the phone ever rings through. That scoring runs in the same sub-one-second window used to decide whether to greet, route, or intercept the call, so spam is filtered without adding any delay for genuine callers.

What happens to the voicemail recordings — are they stored forever?

No. Recordings are stored securely on AWS with automatic retention policies, so audio ages out on a defined schedule rather than accumulating indefinitely. Only the account owner has access to retrieve or export recordings, which was a deliberate fix for the original system's lack of any access control.

Would this still work for a solo consultant with low call volume, not just busy tradespeople?

Yes — the routing rules and greetings are configured per business, so a consultant with a handful of calls a day gets the same benefit as a plumber fielding dozens: every call is answered, screened, and transcribed instead of some being missed. The value comes from eliminating dead-air missed calls, which matters regardless of volume.

Back to case studies

AI / Consumer TechnologyAI-Powered SoftwareSaaS Application Development

AI-Powered Voice Assistant & Call Intelligence

Q: Do we need in-house IT staff or a developer to set this up?

No. The entire configuration — greetings, time-of-day and caller-category routing, retention settings — was built as a self-serve flow inside the mobile app, with sensible defaults out of the box. Sole traders and small teams typically go from signup to a fully working assistant in under five minutes without any onboarding call.

Solving the missed-call problem for small businesses and sole traders with an AI voice assistant that answers, screens, transcribes, and routes every call — turning a chronic source of lost revenue into a searchable, analyzable channel.

Zero missed-call revenue loss with AI-automated call triage and response

Client

Confidential — Consumer AI startup

Industry

AI / Consumer Technology

Timeline

4 months

Technologies

8+ tools

The Challenge

!Small businesses and sole traders — tradespeople, salons, clinics, consultants — routinely lost paying customers the moment a call went unanswered during a job, a client session, or after hours. There was no fallback beyond a generic voicemail greeting, and callers who reached voicemail rarely left a message, assuming they would simply call the next available competitor.

!Spam and robocall traffic consumed a disproportionate share of daily call volume for these users, sometimes accounting for a third or more of inbound rings. Each spam call interrupted work, broke concentration during client-facing tasks, and trained users to ignore their phones altogether — which made them miss genuine calls too.

!The existing voicemail experience was strictly audio: users had to dial in, sit through a menu, and play messages sequentially with no way to skim, search, or prioritize. A busy plumber or therapist returning twelve messages at the end of the day had no way to tell which ones were urgent without listening to each one in full.

!Every business has a different rhythm — a salon wants a different greeting during appointment hours than after closing, a lawyer wants existing clients routed differently than cold inquiries — but the product had no mechanism to configure greetings or handling rules by time of day or caller category, forcing one generic message on every caller.

!With no analytics layer, business owners were flying blind on their own call traffic. They could not see how many calls they missed during peak hours, how spam volume trended week over week, or whether changes to their availability were actually reducing missed-call rates.

!Voicemail audio needed to be stored securely and retained appropriately, but the original system had no retention policy or access control, creating both a storage-cost problem as usage grew and a privacy exposure since recordings of sensitive client conversations could persist indefinitely with no owner-level access management.

!Real-time performance was a hard constraint: the assistant had to answer, greet, and begin capturing a message within a couple of rings to feel natural to callers, while still running spam scoring and routing logic in that same window — a tight latency budget for a mobile-first, telephony-integrated pipeline.

!Because the product served solo operators without IT staff, every one of these capabilities — spam filtering, transcription, routing rules, analytics — had to be self-serve and configurable from a phone in under a few minutes, with no onboarding call or technical support required.

Our Solution

We built a mobile AI voice assistant that intercepts calls the moment a user is unavailable and answers with a warm, natural-sounding greeting, configurable per user so it reflects their business tone rather than a generic robotic message. The assistant can engage in short exchanges to capture the caller's intent before routing or recording.

We implemented an active spam-detection layer combining known-spam number databases with real-time call pattern analysis — caller ID reputation, call frequency, and behavioral signals — so that likely spam is intercepted before it ever rings through to the user, rather than relying on the user to hang up.

We developed real-time speech-to-text transcription using OpenAI Whisper for every voicemail, feeding a visual, searchable inbox where users can scan message previews, search by keyword, and jump straight to the relevant part of a call instead of listening start to finish.

We built configurable time-of-day and caller-category routing rules so a business owner can set different greetings and handling behavior for business hours versus after-hours, and for known clients versus first-time callers, all from a simple in-app rules editor.

We shipped a call analytics dashboard surfacing call volume trends, spam interception rates, and missed-call heatmaps by hour and day, giving owners the visibility to adjust staffing, availability windows, or pricing based on real demand patterns rather than guesswork.

We deployed secure audio recording storage on AWS with automatic retention policies and user-controlled access, so recordings age out on a defined schedule and only the account owner can retrieve or export sensitive call content.

We architected the telephony pipeline on Twilio with Node.js and Python services on AWS Lambda to keep answer latency low, using Redis for hot-path caching of spam scores and routing rules so decisions happen within the first second of a call.

We designed the entire configuration experience — greetings, routing rules, retention settings — to be self-serve from the mobile app, with sensible defaults out of the box so a sole trader could go from signup to a fully configured assistant in under five minutes without any support intervention.

Measurable Impact

Missed calls captured

100%

The AI assistant answers every call regardless of user availability, eliminating dead-air missed calls entirely.

Spam intercept rate

94%

Active spam detection filters unwanted callers before they reach the user, reclaiming hours of interrupted work time.

Voicemail scan time

10x faster

Searchable text transcripts replace slow sequential audio playback, letting users triage a full day's messages in minutes.

Lead recovery

15-20% more inquiries converted

Businesses report recovering roughly 15-20% more missed-call leads that previously went to competitors, directly protecting revenue.

Setup time

Under 5 minutes

Sole traders configure greetings, routing, and retention entirely self-serve from the mobile app with no support calls.

Answer latency

Sub-two-ring response

Calls are greeted and routed within roughly one second of connection, preserving a natural caller experience.

“Every missed call was a missed sale. Now our AI handles the call professionally, filters out the junk, and I can read a searchable transcript when I'm free instead of replaying twelve voicemails one by one. We've recovered real business from calls that used to just disappear.”
F
Founder
Consumer AI Platform (NDA)

Technology stack

React NativeNode.jsPythonOpenAI WhisperTwilioPostgreSQLRedisAWS Lambda