marina_evening7.75
The Marina evening table
Decides at 18:00 whether to sit outside, and checks the weather first.
Clarity4.0
Specificity10.0
Fit to daypart8.5
Brand voice6.5
Make
Panel
Five personas score every variant. It is a filter, not customer research — it applies a rubric someone wrote down, it does not observe a reaction.
not customer researchdeterministic under seed
Personas5each publishes its own bias
Best scoringeid-evening7.07 / 10
Worst scoringsummer-detox3.69 / 10
Criteria4clarity, specificity, daypart fit, voice
The panel’s most useful single behaviour is discounting a creative aimed at the wrong hour: a morning persona drops a well-written evening ad by three points on daypart fit, which catches a good ad pointed at the wrong slot before a human spends attention on it.
Scores by slot
| slot | marina evening | value | resident | brand manager | commuter | mean | spread |
|---|---|---|---|---|---|---|---|
| eid-evening | 7.8 | 7.2 | 7.1 | 7.0 | 6.2 | 7.07 | 1.52 |
| morning-deira | 6.6 | 7.4 | 7.0 | 7.0 | 6.6 | 6.91 | 0.78 |
| summer-detox | 3.5 | 4.0 | 4.2 | 2.4 | 4.3 | 3.69 | 2.00 |
Spread matters as much as the mean. A creative every persona scores 6 is a different thing from one scored 9 and 3 — the second has found an audience and lost another.
Who is on the panel, and what each gets wrong
value7.22
The Deira value shopper
Price-led, early, and sceptical of an offer that does not state its terms.
Clarity4.0
Specificity10.0
Fit to daypart5.5
Brand voice6.5
resident7.12
The Al Barsha regular
Comes with family at the weekend and plans around opening hours.
Clarity4.0
Specificity10.0
Fit to daypart5.5
Brand voice6.5
brand_manager7.03
The brand manager
Not a customer. Reads for voice breaches and for claims that will need defending.
Clarity4.0
Specificity10.0
Fit to daypart8.5
Brand voice6.5
commuter6.23
The 07:00 commuter
Buys the same thing every weekday and will not read a second line.
Clarity4.0
Specificity10.0
Fit to daypart5.5
Brand voice6.5
What this is not
It does not observe a reaction; it applies a rubric. Treating a panel score as evidence about real customers would be the single worst misuse of this application — so every persona states what it over- and under-weights, the panel reports its spread as well as its mean, and the score is computed in code rather than by a model, so it is the same number every time it is asked.