Aggregation of Micro Refinements
Run small creative experiments and retain only the changes the data validates
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 98%
Aggregation of Micro Refinements is Maini's continuous-improvement loop for making videos better without expecting one breakthrough tactic. After publishing, he studies audience behavior, including retention, drop-offs, rewatches, countries reached, and end-card clicks. He interprets anomalies relative to the normal baseline for that particular format, because a camera comparison and a top-ten video naturally retain viewers differently. A gut feeling then becomes a controlled creative experiment: change the subscription request, keep his face visible, adjust music, or make the outro feel like continuing content. If the resulting data improves, the refinement becomes a mainstay. Repeating the cycle across many releases compounds numerous small gains into a distinctive production style and stronger audience performance.
Origin
Arun Maini developed this process while shifting from daily uploads toward fewer, higher-quality videos. Extracted from Deep Dive with Ali Abdaal.
Core principles
- 01Major quality gains can emerge from many small validated changes
- 02Each release supplies evidence for the next one
- 03Comparable formats need different performance baselines
- 04Gut instinct proposes experiments while data decides whether they stay
- 05Validated improvements should become standard practice
How to run it
- 1
Establish format baselines
Learn the normal performance range for each recurring type of output. Compare a new release with similar releases rather than with fundamentally different formats.
Pro tip Track a small set of metrics that reveal actual audience behavior, such as retention and next-action clicks.
Watch out A strong metric for one format may be weak or unrealistic for another.
- 2
Inspect the behavioral data
After sufficient data accumulates, review the complete performance curve and identify meaningful spikes, troughs, and exits.
Pro tip Maini reviews each video about two days after publication.
Watch out Do not rely only on total views, which can conceal where the experience succeeds or fails.
- 3
Diagnose a moment
Rewatch the content around an anomaly and propose why viewers stayed, replayed, or left at that exact point.
Pro tip Start with sharp changes because their likely causes are easier to isolate.
Watch out Treat the diagnosis as a hypothesis, not proof.
- 4
Design a micro experiment
Change one relevant element in a future release, such as the timing of a request, visual presentation, music, or transition.
Pro tip Choose experiments that can become reusable production practices if successful.
Watch out Changing many variables simultaneously makes the result difficult to interpret.
- 5
Evaluate against expectation
Compare the experiment's result with the established baseline for that format and consider whether the difference is substantial.
Pro tip Use domain knowledge to distinguish a genuine outlier from normal fluctuation.
Watch out Do not claim certainty from a single noisy result.
- 6
Standardize validated gains
Retain changes that work and incorporate them into future scripts, filming, or editing. Discard or revise changes that fail.
Pro tip Document successful practices so collaborators can apply them consistently.
Watch out Recheck mainstays periodically because audiences and platforms change.
In the wild
Maini observed that asking viewers to subscribe near the start could produce a large retention drop because he had not yet demonstrated value. The behavioral data showed irritation rather than successful conversion, so the early request was removed or delayed.
→ The opening protects fragile early retention instead of interrupting viewers with an unearned request.
A conventional roundup signaled that the useful content had ended, prompting viewers to leave. Maini began discussing and pointing toward related videos immediately after the key content, with his face still visible, so the outro felt like a continuation rather than a closing screen.
→ Clicks on end cards roughly tripled.
Maini experimented with showing his face in a circle while other material appeared on screen. He compared the resulting retention with what that video format normally achieved and kept the treatment when the performance supported it.
→ The visual treatment became a recurring feature of the channel.
Common mistakes
Comparing incompatible formats
Different video or product formats have naturally different behavioral baselines. Treating them as equivalent produces misleading conclusions.
Optimizing total views alone
A headline number does not show which moments helped or hurt the experience. Examine the behavioral curve and downstream actions.
Treating correlation as certainty
Without a perfect control, a promising result remains an informed signal. Repeat successful experiments before declaring a universal rule.
Is it for you?
Best for
It is best for creators and product teams that publish repeatedly and receive granular behavioral data.
Not ideal for
It is not ideal for one-off projects with no comparable releases or reliable audience data.
From the transcript
“It's kind of the the way I describe it is a an aggregation of micro refinements.”
“you have a gut feeling. You try it. It works, or it doesn't work. And then you action it.”
“I look at it for every single video that I post.”
From the episode
How To Build A 10 Million Following - Mrwhosetheboss
Mrwhosetheboss