TheSalesGhai

Curious. Believer. Doer. Building @UseOcto

[TheSalesGhai @nileshghai]

@smperng That's how we ended up building it. The raw conversation is the record, and anything derived points back to the exact turns it came from. A paraphrase can help you find the moment. It shouldn't decide what was promised.

[TheSalesGhai @nileshghai]

@LKLrai Memory, as long as it covers what was promised and what's still open, not only preferences. Callers repeat themselves because the agent forgot the last call, not because it lacked a routine.

[TheSalesGhai @nileshghai]

@kwindla One classifier I'd love to see on top of this: "is this caller back about something still open?" In customer ops that one signal changed how a call should start.

[TheSalesGhai @nileshghai]

@kaixin_tai @ElevenLabs Useful. One view I'd want next: every call from the same customer in one place. Most voice failures I've seen only make sense across calls, like a promise made on Monday that nobody kept by Thursday.

[TheSalesGhai @nileshghai]

@MTSlive @i_am_holt Tone depends on history too. The same polite answer reassures a first-time caller and irritates someone on their third call about the same issue.

[TheSalesGhai @nileshghai]

@AgentGlassInc Handing off the angry caller is half of it. The other half is what that caller hears next time they ring: does the voice agent know the person already fixed it, or does it start from zero?

[TheSalesGhai @nileshghai]

@rohanbhanot Agreed, the order ID is the better key once the number keeps changing. The gap is the first call, before anyone has the ID, or one agent chasing three issues. We key on the open issue, then the person, so either one pulls in the other.

[TheSalesGhai @nileshghai]

Something from customer ops I never had to plan for: callers who are agents.

A personal AI agent retrying every few hours about the same refund looks like five new calls, unless something links them to one person and one open issue.

Who does that linking in your stack?

[TheSalesGhai @nileshghai]

@SalehAhmad_1 One more change in production: the caller comes back. A demo is always someone's first call. Real traffic is full of people on their second or third call about the same thing.

[TheSalesGhai @nileshghai]

@parloa_ai Repeat contacts are where this shows up most clearly. A call can be fully contained and still be the customer's third attempt at the same issue. Do you track that across calls or only within one?

[TheSalesGhai @nileshghai]

@jthathapudiii That turns repeat call detection into a front door problem, not a QA metric. If five retries about one refund show up as five tickets, the queue math breaks fast.

[TheSalesGhai @nileshghai]

@ThorstenDembski Good format. I'd keep that same note for the next time the customer calls the AI back too. Most handover notes go to the person and never come back to the agent.

[TheSalesGhai @nileshghai]

@pranjalbuilds Agree. One gap I see even after the CRM hookup: the CRM has the fields, but not what the caller actually said last time, like "please don't call before 10". That's often what the next call needs.

[TheSalesGhai @nileshghai]

Outbound and inbound voice agents often get built as separate projects.

So the collections bot calls on Monday, the customer rings back on Tuesday, and the inbound agent has no idea what was agreed.

We're building Octo so both read and write the same caller record.

[TheSalesGhai @nileshghai]

@cyrifi Agree it breaks after the call. Then again on the next one: the customer rings back to ask whether the booking went through, and the agent has no idea there was a booking.

[TheSalesGhai @nileshghai]

@Aiholics_ Recalling language and call times is a good start. The harder part is what changes: a fixed issue, a moved appointment, a new address. Updating after every call only helps if old facts get replaced, not stacked.

[TheSalesGhai @nileshghai]

@muratcan Interesting use. Once you know which language the caller was actually speaking, do you carry that to their next call? In support ops, bilingual callers usually wanted to start in the language they ended in last time.

[TheSalesGhai @nileshghai]

@DominionEleojo Same problem one level up. If a caller cancels or changes something, the next call shouldn't be built from the version they already took back.

[TheSalesGhai @nileshghai]

@summer_ikiki_ @_bingwu Moving deployment work into the product is the right instinct. The piece I see stay manual longest is continuity: what the caller said last week still lives in someone's spreadsheet or prompt, not in the product.