Phase 3: Scale

---

The foundation is set, you've found depth in individual conversations. Now the question: what if you look beyond one session? What if you're not just making patterns visible, but by laying them side by side also creating new insights that weren't there before?

---

Where are you?

You have transcripts from multiple conversations. You've iterated with AI, found patterns, given reflections. Now you're wondering: what if this gets bigger?

What if you could apply the same care that a researcher would bring to the data you already have? Seeing patterns. Drawing connections that were always there but stayed invisible. And sometimes creating new knowledge: by making intuitive things visible and demonstrating connections, insight emerges that wasn't there before. Making undercurrents visible that didn't surface in the conversation itself. Looking back at where choices led and what actions came from them.

And what if you could then share that? See if it works in other contexts?

Parts of this I've done concretely. Parts are still a vision. This is the phase where the standalone becomes collective: where individual wisdom that normally stays with one person can be shared and connected. Where you can lay different perspectives side by side and make visible how they relate to each other, without having to choose who's right.

---

The music metaphor

The skeptic story shows what becomes possible when you look beyond one session. But how do the three phases actually relate to each other? This is where AI helped me find a good metaphor. Music.

Phase 1 teaches you the instruments: recording, transcribing, preserving. Phase 2 teaches you to play: analyzing, reflecting, iterating. Phase 3 isn't a new instrument; it's conducting an orchestra. You combine what you already know, but now with multiple instruments at once.

The conductor doesn't play. The conductor listens, connects, and makes sure the whole becomes more than the sum of its parts. That's exactly what you do here: not analyzing yourself, but bringing analyses together into something that no single analysis could have shown.

That's also what happened in the skeptic story: the individual session analyses each showed their own picture, but only when I laid them side by side did the shift become visible.

---

The core principle: first apart, then together

Phase 3 is really Phase 2 at scale. The same techniques, but applied to multiple conversations, sessions or groups. Being able to follow the full story of a decision: what choices did we make, what actions came from them, and where did that lead?

The principle:

1. First analyze separately: each session or breakout on its own, with the same methodology and the same prompt
2. Then synthesize: lay the analyses side by side, look for patterns, bring together

Why not give everything to AI at once?

The most important reason: first look yourself. Start with one transcript. Does the output match your intuition? Does your prompt work? Are you getting what you're looking for? Only then scale up. That way you know whether you can trust AI before you throw ten transcripts at it.

But there are more reasons:

• Comparability: if you analyze each session with the same prompt, you can lay the results side by side. Patterns become visible that you would never see in separate analyses
• Context loss: AI becomes less precise with large amounts of text at once. By distilling first you preserve the nuance of each conversation
• Error resilience: if the separate analyses are solid but the synthesis doesn't satisfy, you only need to adjust the synthesis. You already have the judgment: the problem isn't the analysis but how it comes together.

The strength is in the comparability. The same prompt on ten sessions yields ten comparable analyses. That's where the real insights are.

---

Four organizational patterns

When you start applying the core principle (first apart, then together), you quickly discover there are different forms. Are you working with transcripts from sessions that happened over weeks or months? Then you're mostly analyzing retroactively. Working with breakouts from the same day? Then you can sometimes feed the analyses back live. The form determines the rhythm.

Below are four patterns I've encountered. The situation determines which pattern fits; this is a menu, not a growth path.

1. Over time: same group, multiple sessions

Session 1 ──► Session 2 ──► Session 3 ──► Session 4
    │             │             │             │
    ▼             ▼             ▼             ▼
 Analysis     Analysis      Analysis      Analysis
    └─────────────┴─────────────┴─────────────┘
                        │
                        ▼
               Synthesis over time

This is the form we apply for instance in a collaboration with a group of volunteers working on improving care in their community. You see development, growth, stagnation. The skeptic story above is an example.

2. Parallel: simultaneous but separate (breakouts, key figures)

      Breakout A ──► Analysis A ──┐
                                  │
      Breakout B ──► Analysis B ──┼──► Synthesis
                                  │
      Breakout C ──► Analysis C ──┘

Three groups talking at the same time. Each group gets its own analysis with the same prompt. Then you bring the insights together: where do the concerns overlap? Where do they differ? You can feed this back live at the end of the session, or work it out afterward.

3. Sequential: different groups, same theme

Group 1 ──► Group 2 ──► Group 3
    │           │           │
    ▼           ▼           ▼
 Analysis    Analysis     Analysis
    └───────────┴───────────┘
                │
                ▼
        What do we learn?

Different groups talking independently about the same theme. You look for patterns: what keeps coming back? What is unique to one group? What does the difference tell you?

4. Rolling forward: building on each other (output becomes input)

Group 1 ──► Analysis ──► Output 1
                            │
                            ▼
            Group 2 ──► Input + Analysis ──► Output 2
                                                │
                                                ▼
                            Group 3 ──► Input + Analysis ──► Output 3

The output from group 1 becomes input for group 2. Each group builds on what the previous group delivered. AI maintains the thread while the content grows. Important: the group's feedback is also processed and sharpens the result. It's not just passing along; it's iteration.

In practice you often combine patterns. Breakouts that roll forward. Sessions over time with changing composition. The core principle stays the same: first analyze separately, then synthesize.

---

Lenses: what you can see

With multiple analyses side by side you can apply different lenses. A lens is a question you ask of your material. I think there are many more than what's listed here. Ownership and patterns over time I've concretely applied in the bottom-up process I described earlier. The other lenses are what I think is possible based on what I saw there. An open invitation.

Following shifts: What changed? In language, in themes, in questions, in energy. What disappeared? What appeared? Shifts are the heartbeat of development.

Drawing connections: Where do people struggle with the same thing without knowing it about each other? Where do they actually agree? What shared choices and visions of the future become visible? But also: where are the differences, and what do they tell you? Finding hooks where connection can emerge, especially across difference.

Reading ownership: Does the language shift from "they should" to "we will"? Are people taking more initiative? Ownership may be the most fundamental shift to track. This is what I've experimented with most concretely: pattern analyses over five sessions, a chronicle we presented back to the group as a narrative overview, and choices and actions we tracked systematically.

Reading energy and consensus: Where is the energy in a conversation? Where does it go quiet? Where does real consensus emerge, and where does everyone just go along? This is still experimental; I haven't found a concrete application for it yet.

Group dynamics: What undercurrents are at play? What lives beneath the surface? Where do coalitions form? This requires an important nuance: group dynamics can be analyzed without singling out individuals. "IS initiative being taken" rather than "WHO takes initiative." The line between insight and surveillance is thin; psychological and social safety always come first. This is also still experimental.

Guarding inclusion: Who is being heard and who isn't? Which perspectives get buried in group dynamics? Experimental; important enough to name, but I don't have my own experience with this yet.

Each lens works on a single conversation. But the real strength is in consistently applying them across multiple sessions. Then meta-patterns become visible that you would otherwise never see.

---

What you do with it: three interventions

The value isn't just in seeing, but in what you do with it.

Mirroring: showing the group their own development. "This is what you said three months ago. This is what you say now. Do you recognize that shift?" In a process where we worked with a steering group, we created a narrative overview of five months of collaboration: the search, the friction, the breakthroughs. When we presented that back to the group, they found it valuable to see how much they had actually achieved.

In a transformation plan session in mental healthcare (GGZ, the Dutch mental health system) we worked with a faster variant: group 1 shared their vision, the input was processed live by AI and put on screen, and group 2 read it and responded. Their first reaction to what was on screen was positive, but the real value was that it could be concretely built upon: the conversation was about refining what was already there, and offered space for more nuanced perspectives. There was something to respond to instead of having the conversation all over again.

Connecting: Bringing people together around shared experiences. "You're struggling with the same thing; did you know that about each other?" This works strongest with parallel sessions and with groups that don't know each other.

Evaluating: showing clients or the people involved what shifted. Not as top-down reporting, but as visible evidence of development in the language of the participants themselves. In an evaluation of a bottom-up process we did this entirely in collaboration with AI: laying out evaluation points, AI helped structure and supplement. Honestly: the client was busy with other things and didn't read it. But the process itself was valuable; it forced us to articulate clearly what had shifted.

---

Do you recognize one of these situations?

The deep dives below help you make Phase 3 concrete. Each describes a situation, an approach, and what you can expect. Where you start depends on what you have and what you want to know.

Same group, over weeks or months

You're following a process over a longer period and want to know what shifts.

You follow a group or process over time. What changes in the language? Where does the energy shift? This is the most fundamental application of scale: making development visible that would otherwise stay invisible.

Breakouts or parallel sessions

Multiple conversations, same day. You want to analyze what groups say and find, and compare them.

Laying multiple conversations from the same day or process side by side. Where do people struggle with the same thing, where do they agree, what shared choices and visions do they hold? Where are the differences? This is the application for breakouts, parallel tables, and sessions with different groups on the same theme.

→ Go deeper: In the session, finding hooks and connecting (coming soon)

Long-running process, ownership question

Floor de Ruiter's formula says: Success = Quality of Idea x Ownership. A brilliant plan without ownership dies. A mediocre plan with lots of ownership lives. Can AI make ownership visible? That was the question I wanted to tackle last fall. This is the most concrete, best documented deep dive.

→ Go deeper: Ownership, can AI make ownership visible? (coming soon)

---

What stays human?

The tension in this phase is the greatest. You're no longer working with a single conversation, but with months of material. The temptation grows: automate everything, scale everything, make everything efficient.

But this is precisely where human judgment becomes crucial. The smarter AI gets, the more we need to trust our own intuition. AI can find connections between people, but are they real connections or coincidental overlap? AI can show patterns over time, but which patterns tell the real story? What I notice is that I increasingly trust my gut feeling precisely because I can now test it against what AI shows.

I think there's also a shadow side to that growth. As we trust AI more, the risk is that we stop taking our intuition seriously. That we stop checking our gut feeling against what the analysis says. While precisely that gut feeling, that human taste and intuition, is the distinction that matters.

The bigger the scale, the more important human judgment becomes. AI can find ten patterns; which three tell the real story?

---

Tensions in this phase

Overlap vs. real connection AI finds similarities; it's getting better at this, but the difference between real connection and coincidental overlap still requires your judgment. Two people who both said "tired" aren't automatically allies.

Showing everything vs. selecting AI can name endless patterns. The art is choosing: which three tell the real story? Too many is overwhelming. Too few misses nuance. Finding the balance is human work.

Sharing vs. protecting Scale makes things visible that previously stayed in the room. Some insights are for this group, this moment. What may be shared, what needs to be protected? That trade-off calls for extra care.

Analyzing separately vs. wanting speed It's tempting to give everything to AI at once. But separate analyses with the same prompt yield comparable results, and that's where the real insights are. Patience pays off.

---

The bigger question

Somewhere in this phase, a different question starts to emerge.

You've seen what AI can do for one session. Then for multiple sessions over time. Then for an entire trajectory within a team or community: patterns becoming visible, ownership growing, dynamics you can now name.

And then:

”

What could this do for a neighborhood? A community? An entire sector wrestling with the same challenges?

What if we, with consent and with ethical oversight, would use AI to make collective patterns visible? To let wisdom accumulate instead of evaporating after every session, every workshop, every project?

I sometimes call this the "Social GitHub" dream. I'm not sure if that's exactly the right name, but the idea won't leave me alone. Developers share code on GitHub: not just their solutions, but also how they built those solutions. What if communities could do the same? Not just "this worked for us," but also "this is how we approached it, these were our struggles, this is what we learned along the way."

A neighborhood that discovers something about how to build ownership around care and caring for each other within the community: that knowledge could be accessible for a neighborhood in Groningen wrestling with the same thing. Not as abstract theory, but as concrete experience: their words, their process, their lessons.

I haven't found software that does this yet. And I don't think it's a far-fetched idea that it will exist.

That's still a vision. But every vision starts with small steps, and those steps are what this field guide tries to describe.