calvin.goh
/ aam · 02 / conversational-analytics-agent
← All projects
Gartner Analytics Ascendancy Model · stage 02 of 04 · on Sooka
01
Descriptive
what's happening?
02 · YOU ARE HERE
Diagnostic
why is it happening?
03
Predictive
what will happen?
04
Prescriptive
what should we do?
Diagnostic · why is it happeningSooka · content analyticsMultilingual libraryAgentic workflow

A conversational analytics agent for Sooka's content team.

Descriptive watchers tell you something moved. Diagnostic work starts the moment the Head of Content asks why. This agent takes the question in plain English, picks the right slice of Sooka's content-performance scoring framework, runs the comparison across languages and titles, picks a chart, writes the narrative — and ships a one-page brief to the exec who asked, before standup.

Question → exec brief
< 90s
Plan + run + write + send
Languages in scope
5
MS · ZH · TA · KR · EN
SQL written by the asker
0
Plain-English questions
Output format
1-page
Headline · chart · so-what · ask
01 · PROBLEM
Diagnosis is rate-limited by the analyst

Every “why did Korean drama retention drop?” lands in the analyst queue. By the time the answer arrives — three hours later in deck form — the content team has moved to the next licensing call.

02 · APPROACH
Plan, slice, narrate

The agent decomposes the question into a plan: which library, which cohort, which comparison, which chart. It runs the slice across the multilingual content scoring framework and writes the narrative in the team’s internal voice.

03 · RESULT
The 80% case answers itself

Routine “why” questions never reach a human analyst. The team gets back to commissioning calls and licensing strategy — the upper half of the AAM curve, where their judgement actually compounds.

// THE EXCHANGE

One question. Three artifacts. No analyst.

// CONVERSATION · 09:14 GMT+8
CG
You09:14:02
Sooka Q1 completion is down vs Q4. Why? Slice by language library — I want to know if it's a Korean-drama story or platform-wide.
Sooka Agent09:14:11
Got it. I'll pull content-perf · completion-rate by language for Q1 vs Q4, hold viewing-hours constant to rule out volume, and flag where the drop is significant.
Sooka Agent09:14:38
Top finding: it's a Korean library story. Korean completion fell 54% → 38%(−30%). Chinese also softened (−8%). Malay, Tamil and English are flat to slightly up — so it isn't platform-wide.
COMPLETION % · BY LIBRARY · Q1 vs Q4auto-picked · grouped bars
Korean-16Chinese-4Malay2Tamil1English-2
Q4 Q1
CG
You09:15:09
Is it the new licensing batch?
Sooka Agent09:15:14
Plausible. Korean titles licensed in Q1 have a median content score 19% lower than the Q4 cohort. Trial-to-paid on Korean entry-cohorts also softened. Want me to check thumbnail-CTR and time-to-first-play by title?
// LIVE RUN

Now ask it yourself.

Pick a question — or paste your own (4 KB cap) — and watch the agent plan, slice, narrate, and emit the same one-page brief format you just saw above. Live model call. Cached replay if the rate-limit fires.

pick a fixture_
or

waiting for verification token…

~/demo/conversational-analytics-agent.log
idle · waiting for input
model_
pick a fixture, then press [ run demo ].
01 · plannumbered steps
waiting for run…
02 · chartauto-picked from shape
waiting for run…
03 · briefexec one-pager
waiting for run…
── agent log ──
no log entries yet
Diagnostic analytics is where most teams get stuck — answers exist, but only the analyst can reach them. Up next on the model: predictive.
NEXT IN THE MODEL →
Predictive
Sports-season churn forecasting · Sooka