From idea to impact, in five stages

Most people describe an AI content system as one thing that writes. The run log disagrees. Content is the smallest stage in the loop. Research, distribution and follow-up are where the hours actually go, and a system that skips them produces drafts nobody reads.

2,683runs across the five stages

949runs at distribution

849runs at research

339runs at conversion

This is the squad that runs the whole loop, grouped the way the work actually breaks down.

Finding one

The stages are not equal, and the imbalance is the point

Fig. 1 — runs by stage. Nine hundred and forty-nine runs at distribution against 175 at creation. Stages are assigned by job name, so this is an operational map rather than a tidy funnel. The shape is what matters: publishing an idea is a smaller job than getting it in front of people.

I have run this loop enough times to recognize the shape. Writing the piece was never my bottleneck. Finding the thing worth writing, getting it in front of the right people, and following up on the response is where my hours actually go.

Finding two

Research is a stage, not a step

849 runs sat in research: knowledge deep-dives, intelligence sweeps, RSS ingestion, the subject-mastery loop. That is not browsing. Each run has an output, a brief or a memory entry the next stage can read, which is the only thing that separates research from procrastination.

The failure mode is a brief that says "write about AI agents." The useful version names the audience, the claim, the evidence, and the objection the piece has to answer. A research agent that produces the second kind is the entire reason the creation stage stays cheap.

Finding three

Distribution outnumbers creation five to one, and it should

Fig. 2 — share of the loop. Distribution is the biggest stage by a wide margin. Newsletter queue syncs, image updates, archive syncs, platform posting. None of it is creative. All of it is the difference between a piece existing and a piece being seen.

There is a version of this that is pure noise, and it looks like posting every piece everywhere with no adaptation. I have shipped that version. The one that works treats each channel as a different audience with a different opening line, and spends the runs to rewrite rather than crop.

Finding four

Engagement is where attention becomes intent

371 runs went into engagement this cycle: nurture passes on the algorithm report, reply watching, message logging, connection monitoring. This is the stage most teams skip entirely, which is why their pipeline looks like a cliff after the content ships.

An engaged reader is a different person than a reader. They have told me something about what they care about, and that is the cheapest qualification signal I have found in B2B.

Finding five

Conversion is the smallest stage, and that is correct

Fig. 3 — monthly runs across the loop. The engine holds its level. A campaign would spike and vanish. A loop with a schedule produces roughly the same shape every month, which is what makes the output comparable and the pipeline forecastable.

339 runs at conversion, and I want every one of them arguing with a warm human rather than a cold list. A funnel with a wide conversion stage is usually one that skipped qualification and is paying for it with unqualified calls.

The five, in one page. Research finds the story. The creator builds it in every format. Signals find who cares. Outreach starts the conversation. The closer keeps the pipeline honest.

What to copy

Five moves for running the whole loop

Map the loop before you build it. Name the five stages and the one artifact each hands to the next. Gaps in a loop are cheaper to find on paper than in production.
Make research earn its name. A brief with an audience, a claim, evidence and an objection. Anything less is a topic, and topics produce generic pieces.
Budget for distribution at five times creation. If the effort is inverted, the writing is decoration.
Treat engagement as a stage, not an outcome. Someone has to work the replies, on a schedule, with a place to record what was learned.
Keep conversion narrow. If the last stage is wide, the middle stages were not filtering.

What the loop cannot do. Decide that the idea is worth a month of attention. Every stage here can be argued with, improved and automated. The conviction that a topic matters is the input the engine cannot generate for itself.

The engine is the advantage, not the artifact

One piece of content is an event. A loop that researches, produces, distributes, engages and converts on a schedule is an asset that compounds, and the run log is the only honest way I have found to see which stage is actually carrying it.

Five stages, one continuous engine, and the number that matters is not how many pieces shipped. It is whether the loop still turns in a month when nobody is watching.

Sources

Sexton, Koka. “The Idea → Impact Squad.” Agent crew infographic series, 15 Oct. 2026.
Run telemetry, 2,683 completed runs assigned to five operational stages by job name, 6 May – 26 Oct. 2026. Stage assignment is by job name, not by semantic classification.