Most "coaching AI" products take you into a scripted nowhere-land: the AI invents a hypothetical situation, you play along, and afterwards you hope some skill transfer happens in your real life. PPL Spark made a different bet.
The bet in one line
Conversations in defined systems — organizations, call centers, team meetings — have predictable structure. That makes them tractable for AI to observe and improve.
Two things follow from this:
- The right material is the conversations people are already having. Not hypotheticals.
- The AI's job is to observe, support, and raise the quality of dialogue — not to generate answers in place of the participants.
Why this is the right moment
Three things converged:
- Transcription got good. Speaker diarization, language detection, and sub-second latency are now commodity. Five years ago, "AI that sits in your meeting and understands who's talking" was a research project. Today it's an API call.
- LLMs can do nuanced real-time coaching moves. Not replace a coach — but notice, for instance, that one participant hasn't spoken in five minutes, and quietly suggest to the host that it's time to bring them in.
- Remote and hybrid work made meetings the dominant unit. The conversation used to be one thing that happened alongside "real work." Now, for knowledge workers, much of the work is the conversation.
The technology caught up to a problem that has been hiding in plain sight.
Two product lines, one bet
The Live Companion. An AI that participates in a meeting while it's happening — transcribing, identifying speakers, tracking the agenda, noticing silence, prompting the host: "Reiner has been quiet for 5 minutes — bring him in." It's a facilitator's second brain.
Agent-Based Modeling. Feed in a situation (a regulation, a lobbying campaign, a family decision), and the system extracts relevant actors, builds their personas from public information, simulates how they'd react, and produces a report. Applied to policy analysis (e.g. MiCA regulation with 49 actors) and personal scenario planning.
Both land on the same bet: start from something real — a live meeting, a real situation — and use AI to see it better.
The proof case
The reference customer is Zurich Kundencenter Lernpilot — an AI pilot inside a Zurich insurance call center, mirroring and improving customer-service conversations, measured against existing NPS. This shape — defined conversations, measurable quality, real business stakes — is the signature of the customer PPL Spark is built for.
If you take one thing from this lesson: we're not betting on "AI that helps you have better imaginary conversations." We're betting on AI that helps you have better real ones.
What's next
The next lesson introduces RAIner, the flagship product — what happens in the 20 minutes a user spends with it, and why the output isn't a personality label but a trajectory.