Method
How the numbers are read.
Anyone can point at a chart that went up. The harder question is whether anything you did caused it. This is how each pilot is set up so that question has an honest answer.
Record the baseline first
Before anything changes, we write down where your channel stands: the metric we care about, its value across the last 90 days, and the date. You get a copy.
Without that, any result is just a story told afterwards. With it, both of us can see exactly what moved and by how much.
Four rules we hold to
- One variable at a timeChange three things at once and nothing can be attributed to any of them. Each avenue is tested on its own, and held long enough to read.
- Rates, not totalsMedian views per post, click-through rate, follows per 1,000 reach. Totals rise simply because you posted more, which says nothing about whether the approach worked.
- Medians, not averagesOne video that takes off will drag an average upwards and make a weak strategy look strong. The median shows what a typical post actually does.
- Matched windowsSix weeks compared against the six weeks immediately before, or the same point in last year's game cycle. Launches and seasons move everyone's numbers, and that is not a result.
Four problems, four different answers
Most channels have one real bottleneck. They look similar from the outside and need completely different responses, so the first job is working out which one you have.
| Problem | What it looks like in the data |
|---|---|
| Discovery | Low impressions, or little reach beyond existing followers |
| Packaging | Plenty of impressions, but few people click |
| Retention | People click, then leave in the first few seconds |
| Conversion | Strong views, but few follows, subscribes or returns |
Results
One creator, anonymised at their request. These are public view counts on their own channel, before and after a change of game and a change in how videos were titled, based on a read of what their earlier videos were doing.
Roblox creator, 2025. Each bar is one video.
What changed. A read of the channel's earlier videos showed which formats were holding attention and which were competing in a crowded space. On that basis we recommended a different game to build around, and a different approach to titling. The two videos that followed are shown above.
What this does and doesn't show. These are the videos from that period, not a controlled comparison. A change of game brings a different audience with it, so the jump reflects both the choice of game and the packaging. We've published the raw numbers rather than a percentage so you can judge it yourself.
More results will be added as current pilots finish, published only with the creator's written approval.
What the read contains
The main deliverable is a written document, two to three pages, delivered within 7 days of your baseline. Every pilot's read has the same four parts.
- 1. Where you standYour baseline numbers, the window they cover, and the date they were taken. Nothing interpreted yet.
- 2. What the data showsWhich of the four bottlenecks above your channel actually has, and the numbers that point to it. Where the data is thin or ambiguous, we say so rather than filling the gap.
- 3. Three to five avenuesEach one states what in your data prompted it, the idea itself, how you would test it, and what would count as it working or not working.
- 4. What we would not doThe obvious moves we considered and rejected, and why. This is usually the most useful part, because it is the advice you would otherwise get from everyone else.
A worked example
Choosing what to build around is the decision that sets up everything else. Here is the kind of picture we put in front of a creator before that call gets made, using public Roblox player counts.
Why it matters for a channel. A record-breaking spike is a short window with enormous competition, and when it passes the audience leaves. A game holding six figures every day is a smaller audience that is still there next month. Which one suits a channel depends on whether the creator is chasing a moment or building something that compounds.
Read the limits. Peaks and current counts here come from different sources and dates, and live counts move hour to hour. The chart is for the shape of the pattern, not the precision of any single number.
Want this run on your channel?
Six weeks, no fee while we build the next case studies. Three pilots run at a time.
Get in touch