Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Coaching Your Client Calls in Real Time
Transcript
- Lucas: So there's a number that's been stuck in my head all week: thirty-seven percent of enterprise software buyers now expect vendors to use conversation AI during product demos. That's from a Gartner pulse survey published just this month — June 2026. Luna: Wait — they expect the seller to be coached by AI while they're pitching? That feels like a pretty big shift from just recording calls for later review. Lucas: Exactly. And it's not just demos. Real-time AI coaching is quietly becoming a must-have in sales tech, customer success, even recruiting. The idea is that instead of getting a scorecard after the call ends, you get a subtle nudge — a buzz on your wrist, a word on your screen — while you're still talking. Luna: A couple of dollars a month is genuinely what keeps these going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. Lucas: Yeah, it's listener support that keeps this show ad-free and lets us dig into these specific numbers. And speaking of specific numbers, let's talk about how this technology actually works under the hood. Lucas: Most of the major platforms — Gong, Chorus, which is now part of ZoomInfo — have added a live mode. But there are also newer entrants like Second Nature and Replica AI that were built from the ground up for real-time guidance. Luna: So what's the technical architecture? How fast does it have to be to not feel laggy? Lucas: The latency target is under 200 milliseconds for transcription. They use a whisper model — that's OpenAI's speech to text system — running on-device or on a low-latency edge server. Then a smaller large language model, often a fine-tuned Llama or Mistral variant, scores the rep's speech against a predefined playbook. Luna: And the playbook is what — a set of best-practice phrases, objection-handling flows, that sort of thing? Lucas: Right. The company's sales team or enablement team writes it. Things like: after the prospect mentions a pain point, confirm it back before pitching. Or: if you haven't mentioned pricing by minute twelve, prompt. The model checks for those patterns in real time. Lucas: If it detects a deviation — say the rep jumps straight to features without acknowledging the problem — it sends a cue. That could be a haptic buzz on an Apple Watch, a subtle color change on a second monitor, or even an earpiece whisper. The rep then adjusts mid-sentence. Luna: I've seen demos where the cue is literally a glowing dot in the corner of the screen. It's non-intrusive but hard to miss. But I wonder about cognitive load — isn't it distracting to get coached while you're already juggling the conversation? Lucas: That's the big pushback. A Forrester study from late 2025 looked at this specifically. They found a twenty-two percent lift in deal close rates for teams using live cues compared to post-call review only. But they also noted a learning curve — reps who had used the system for fewer than ten hours actually performed worse. Luna: So there's an adaptation period. The brain has to learn to treat that buzz as useful information rather than noise. Lucas: Exactly. The vendors are responding by making cues more contextual. Instead of a generic 'ask about budget' prompt, the system might wait until the prospect uses a phrase like 'we're looking at Q3' and then nudge: 'They just signaled timeline — confirm budget range.' Luna: That's much smarter. Now, what about disclosure? If I'm on a call and the person on the other end is getting AI coaching, don't I have a right to know? Lucas: This is where it gets legally interesting. California's new AB-2876, which took effect January 2026, requires explicit consent for any AI that 'materially alters a conversation.' That includes real-time coaching. So if you're selling to a company in California, you need to say something like 'I'm using an AI assistant to help me serve you better — is that okay?' Luna: And if the prospect says no, does the system just shut off? Lucas: Yes. Most vendors now have a one-click disclosure toggle. If the prospect opts out, the system stops transcribing and coaching immediately. Some even log that as a compliance event. Other states are watching California — New York has a similar bill in committee right now. Luna: So this technology is spreading beyond sales. I've seen it used in customer support, where the AI coaches the agent on empathy statements. And in recruiting — training interviewers to avoid bias by flagging leading questions. Lucas: Recruiting is a fascinating use case. If an interviewer asks 'So you're from Stanford, huh?' in a way that could signal bias, the system can buzz them. Or if they interrupt the candidate too often. Some companies are using it for diversity training in real time. Luna: But you could also imagine it being used to pressure reps — like 'you haven't asked for the close yet' — creating more stress. Lucas: That's a legitimate concern. The best implementations let the rep customize the frequency and type of cues. You can dial it down to only the most critical prompts. And the playbook should be co-authored with the reps, not imposed by management. Lucas: There's also a question of data privacy. The transcription is happening in real time, but where is that data stored? Gong and Chorus both say they retain transcripts for model training unless you opt out. Second Nature offers an on-premise deployment for sensitive industries. Luna: So if you're in healthcare or finance, you'd probably want that on-prem version. Lucas: Exactly. And the EU's GDPR adds another layer — you need a legal basis for processing, usually legitimate interest or consent. So multinational companies are already standardizing on a consent-first approach. Luna: Let's zoom out. How big is this market? Are we talking about a niche tool or something that every enterprise will use in five years? Lucas: IDC estimates the real-time conversation intelligence market at about $2.3 billion in 2026, growing at over thirty percent CAGR through 2030. That includes sales, support, and HR use cases. And Gartner predicts that by 2028, sixty percent of enterprise customer-facing roles will use some form of real-time AI guidance. Luna: That's a rapid adoption curve. What's the biggest barrier right now? Cost? Trust? Lucas: I'd say it's the combination of integration complexity and the creepiness factor. Getting the real-time pipeline to work with existing CRM and telephony systems is non-trivial. And if the implementation feels like surveillance, reps will resist. The successful deployments frame it as a coaching tool, not a monitoring tool. Luna: So culture matters a lot. A trust-based culture might adopt this well; a command and control culture might break it. Lucas: Exactly. The technology itself is neutral — it's how you deploy it. I think we'll see a split: companies that use it transparently and collaboratively will see the twenty-two percent lift. Companies that use it to micromanage will see turnover. Luna: I'm also curious about the future. Could this eventually coach managers during one-on-ones? Like, 'You haven't asked your direct report about career growth in three meetings — bring it up now.' Lucas: That's absolutely on the roadmap for several vendors. And there's even talk of using it for job interview practice — imagine a tool that coaches you in real time during a mock interview, then gives you a full transcript with feedback. Luna: That would democratize access to interview coaching, which is usually expensive and exclusive. Lucas: Right. So the same technology that helps a sales rep close a deal could help a recent grad land their first job. That's the optimistic view. The pessimistic view is that we all end up speaking in corporate-optimized scripts, every conversation a rehearsed performance. Luna: I think the line is in how much agency the human retains. If the AI is a suggestion engine, fine. If it's overriding your natural speech, that's a problem. Lucas: Well said. And that's probably the biggest design challenge for the next generation of these tools: how to augment without erasing the person. It's not just a technical problem — it's a philosophical one.