I got tired of listening to 50+ call recordings a week on CallRail and Twilio so I built a tool to do it for me

I run rank-and-rent lead gen sites in Australia and the USA. Basically I build a website, rank it for something like "locksmith North Sydney" or "hot tub repair [city]", and sell the calls to a local tradie. I've got 3 sites doing about $800/week combined right now at $50 per booked job.

The SEO and site building part is great. The part that nearly made me quit was the invoicing.

Because my partners only pay for calls where a job actually gets booked, I had to listen to every single recording to figure out which ones counted. At 10 calls a week that's fine. At 50-60 it's genuinely awful. You're sitting there with headphones on for hours listening to voicemails, wrong numbers, people asking for a quote then ghosting, and 10-minute calls where someone describes their problem and then goes "I'll think about it." All to find the maybe 40% that actually booked.

I tried CallRail's keyword spotting and it was terrible — flags "appointment" when someone's complaining about a no-show, misses the real booking when the customer just goes "yeah Thursday works" at the end. Looked at VAs but couldn't get comfortable trusting someone else's accuracy when I'm billing partners per lead.

So I built CallOutcome (https://calloutcome.com?utm_source=indiehackers&utm_medium=organic&utm_campaign=launch_2026_03). It hooks into Twilio or CallRail (takes about 2 minutes to connect), transcribes every recording with Whisper, then GPT reads the whole transcript and classifies whether a job was actually booked. Not keyword matching — it understands context. Also pulls out customer name, service type, urgency, booking date.

Stack is Flask, Postgres on Supabase, deployed on Railway. AI costs about 3-4 cents per call. Whole thing runs automatically now — I just check the dashboard.

The thing that surprised me most was how much it helped with partner trust. I set up proof dashboards where partners can log in and see every call with the transcript and classification. No more awkward "what am I paying for" conversations.

I've run about 500 calls through it. Accuracy is around 95% on classification — I've spot-checked against my own manual review. The 5% it's unsure about get flagged with a low confidence score so I can check those manually, which takes like 2 minutes.

Pricing: free tier does 50 calls/month, paid starts at $29. I'm also doing a Founding 50 deal — $149 one-time for lifetime Pro access (500 calls/month forever) for the first 50 users. Figured early users deserve something for taking a chance.

Right now I've got basically zero external users. Built it for myself, it works, now trying to figure out distribution. Reddit ads were a waste — spent A$91 on 51 clicks and got nothing. Organic Reddit posts have actually worked better — got a signup from one already.

Here's what the dashboard looks like — every call gets a classification, summary, and confidence score:

If anyone here does lead gen, pay-per-call, or anything where you need to know what happened on phone calls, I'd love for you to try it and tell me what's broken. Free tier doesn't need a credit card. I want feedback more than revenue at this point.

Happy to answer questions about the build, the rank-and-rent model, or anything else.