Question
Name the choice the team must make after the test.
Example: pause, scale or revise.
Marketing analytics / controlled experimentation
Fire one question at a time.
Gaforenix prepares noisy marketing signals, fixes the test conditions and lets evidence cool before anyone names a winner.
- 01
- Decision per cycle
- 03
- Evidence layers
- 00
- Vanity metrics required
01 / Raw input bench
Prepare the material before measurement.
Reliable analysis begins with a narrow operating question, a readable baseline and a known owner. Everything else waits outside the chamber.
Baseline
Use the same event, window and comparison rule throughout.
Document known gaps before launch.Owner
Assign one person to interpret the result and record the action.
No anonymous dashboard decisions.Do not fire yet
If tracking changes during the test, stop and relabel the cycle. A cleaner delay is more useful than false precision.
02 / Firing brief
Set the decision, evidence and cadence.
Choose three operating conditions. The kiln returns a practical testing route, the primary risk and the next decision checkpoint.
Observe before adding heat.
Hold the tracking definition stable for one campaign cycle. Compare qualified behaviour, not platform credit.
- Primary risk: mixed conversion rules
- Checkpoint: end of cycle review
- Owner: growth operations lead
03 / Firing profile
Move through four deliberate stages.
Every stage has one job. Use the controls to inspect the condition, risk and hand-off before the next stage begins.
Stage 01 / Prepare
Lock the question and baseline.
Confirm that every source uses the same conversion definition. Record gaps and choose one decision owner before spend changes.
04 / Glaze evidence
Read the surface, the context and the decision.
Check where the signal adhered.
Compare the planned event with observed behaviour. Mark missing segments and separate directional evidence from attribution claims.
- Inspect conversion consistency
- Label platform disagreement
- Keep uncertainty visible
05 / Channel cones
Apply heat to one variable.
Channel cones keep the comparison legible. Select the variable you can change without rewriting the whole measurement system.
Channel allocation
Hold audience, offer and creative constant. Compare spend distribution against qualified behaviour over the agreed window.
Decision: move budget only after the hold period.
06 / Cooling window
Let the evidence settle.
Early results are volatile. Move the control to choose how much evidence the decision can safely carry.
EarlyWorkingMature
Working window
Use directional evidence for planning, but keep budget changes reversible. Review again after the final lag period.
Action: prepare a bounded next test.07 / Result stamp wall
Stamp the action, not the story.
Signal is directional
Keep the baseline stable and watch one more window before changing spend.
Owner / analyticsEvidence is mixed
Isolate the disputed variable and run a smaller controlled comparison.
Owner / growth operationsAction is supported
Record the limit, name the owner and make the agreed reversible change.
Owner / channel lead08 / Studio protocol
Three rules protect every firing.
Open each rule before a new cycle. If a rule cannot be met, change the question rather than forcing a confident answer.
Use the same conversion event, exclusion rule and observation window across sources. Document unavoidable differences beside the result.
Observed behaviour, controlled tests and modelled credit answer different questions. Compare them, but never merge them into one certainty score.
State the action, owner and review date before interpreting the result. This makes post-hoc storytelling easier to detect and correct.
We choose one decision event, a stable comparison window and the exclusions that materially change the result. Known tracking gaps remain visible in the baseline note.
We keep both readings, identify the definition or collection difference and choose which source can support the current decision. Disagreement is evidence, not a reason to average blindly.
Not in every environment. We distinguish observed association, controlled incrementality and modelled contribution, then state what each layer can and cannot support.
Only when the original cycle is closed and labelled. A mid-cycle change creates a new condition, so it needs a new baseline note and decision checkpoint.
The record includes the question, conditions, evidence layers, known limits, chosen action, owner and review date. It is a decision trace, not a decorative dashboard export.
No. Gaforenix structures measurement and experimentation; it cannot guarantee commercial outcomes or remove uncertainty from changing markets and customer behaviour.
10 / Studio brief
Bring one measurement question.
Share the decision, current evidence and known limits. The brief creates a local confirmation only; it does not send data or promise engagement.
- Describe one decision
- Name the strongest available evidence
- State the timing and owner