“Week three’s empty-state board is still on our wall. Merchandising hates it and uses it anyway.”
Flagship · 11 weeks
Query Intelligence Lab
A working education in in-app search analytics. You instrument a sandbox, taxonomise a dump, forensic an empty state, and write a ranking brief a product lead can use without translating it first.
Modules
What the eleven weeks contain
-
Event contracts without drowning in noise
Search submitted, results rendered, impression, click, downstream action. Session stitching on mobile. You will write a one-page contract and watch it fail on purpose.
-
Query taxonomy and intent clustering
Navigational versus exploratory, known-item versus comparative. Naming clusters so two teams can use the same words. Thai mixed-script examples from the first hour.
-
Zero-result and low-result forensics
Empty states as inventory, synonym, and language problems. A triage board you can take back to merchandising. We do not treat a 0-hit page as a tracking bug by default.
-
Ranking features versus presentation bias
Why position one looks gifted. Card design, badges, and image size as confounders. A modest click model, taught as literacy, not as a research paper.
-
Reformulation and session success
Chains of queries as distress, not engagement. Success definitions that survive a sceptical engineer. Time-to-useful-result alongside click-through.
-
Reporting to PMs without vanity dashboards
One brief, three charts maximum, a named limitation. How to retire a metric that only exists because a vendor template included it.
-
Capstone: instrument, diagnose, recommend
Sandbox catalogue plus an anonymised dump. You deliver a search quality memo. We grade clarity and honesty, not optimism.
Learning outcomes
- Define search success in language your engineering and product leads both accept.
- Instrument in-app search events so impressions and clicks are not interchangeable.
- Build a living query taxonomy, including Thai and mixed-script realities.
- Run an empty-state forensic and separate catalogue holes from analyser failures.
- Write a ranking note that admits presentation bias instead of hiding it.
- Produce a capstone brief a colleague could implement the following sprint.
Lead instructor
Wilawan Chaiyasit
Former head of search quality at a Bangkok super-app. She teaches instrumentation and taxonomy, and is allergic to dashboards that cannot be replayed query by query.
Guest sessions on click models are led by Marcus Ellery, who has evaluated ranking for catalogue apps across SEA and still refuses to call himself a data scientist in marketing copy.
Informational fee
Desk Seat ฿48,500. Signal Pair ฿86,000. Instrument Bench (private, up to six) ฿210,000. VAT extra where applicable. No online checkout — see fees and enquire.
Cohorts run three times a year. Remote seats join live; recordings are not a substitute and are not sold.
Who it is for
Product managers, analysts, and search engineers who already have an in-app search box and suspect the metrics. SQL comfort helps; we do not require you to write production pipelines. If you need cluster operations, this is the wrong school.
FAQ
Plain answers, including a limit
Will you instrument our production app?
No. Labs run on a sandbox catalogue and anonymised query dumps we provide. You may bring screenshots of your own dashboards, but we do not receive production credentials or live PII. That is a real limitation: if your only goal is a vendor to wire events into your app, hire an implementation partner instead.
How much Thai language coverage is included?
Practical, not academic. Mixed script, missing spaces, transliteration, and stopword damage are in the Lab. The two-day Thai Query Normalisation Workshop goes further. We do not ship a tokenizer, and we are not a computational linguistics degree.
Is there a certificate?
Yes, if you want stationery. We grade the capstone memo. Most alumni care more about whether their Tuesday search review got shorter.
Can one person attend without a data colleague?
Yes — Desk Seat is designed for that. Signal Pair is better when a PM and an analyst can argue in the same room. We will not invent a partner for you.
From people who sat this Lab
Instrumentation day assumed more SQL than our PM pair had. We caught up; the first two labs were steep. Session stitching finally made the success rate argument stop.