Segment Analysis — Query Workflow
Segment Analysis — Query Workflow
Step 1 — Broad overview (fire all three in one turn)
| Query | Group by | Measures | Limit | Order |
|---|---|---|---|---|
| Device breakdown | Device type | Sessions, CVR, rev/session | — | CVR ↓ |
| Country breakdown | Country | Sessions, CVR, rev/session | 30 | Sessions ↓ |
| Channel breakdown | UTM source + UTM medium | Sessions, CVR, rev/session | 30 | Sessions ↓ |
Device: More completed purchases than payment page views is expected on mobile — Apple Pay / Shop Pay bypass the payment step. Flag as normal, not a data error.
Country: Very high rev/session for a single country likely means local currency, not USD — flag rather than present at face value.
Channel: A large
(none)/(none)share is a UTM tracking gap — flag it. Zero-CVR rows for awareness campaigns are expected — confirm campaign objective before flagging as a problem.
Step 1.5 — Bot filter (mandatory gate before selecting any signals)
Before identifying signals, disqualify any segment that meets all three:
- High session volume (above 1% of total sessions)
- 0% or near-zero CVR
- No plausible organic explanation — e.g. the store doesn't ship there, the country has known firewall or payment barriers (China is a common example), or the source has no associated campaign
Stop. Do not include a disqualified segment as a priority card.
If you are uncertain whether a segment is bot-driven, run a one-week breakdown for that segment to check whether the pattern is consistent or a spike. A steady flat 0% CVR over the full 30-day window with high volume is a strong bot signal; a spike in a single week may be a campaign or referral event worth investigating.
If a disqualified segment's session volume is above 1% of total, note it in a brief footnote below the priority cards: e.g. "China (19,984 sessions, 0% CVR) excluded — likely bot or non-addressable traffic."
Step 2 — Compute benchmark and find signals
Compute the site-wide CVR (total purchases ÷ total sessions across all segments) — use as the benchmark throughout.
Benchmarking rules — apply before selecting any signal:
- Device: Do not flag a DEVICE GAP just because mobile CVR ≠ desktop CVR — some gap is structurally expected. Only flag if the mobile/desktop CVR ratio has worsened week-over-week: compare week 1 (days 1–15) vs week 2 (days 16–30) of the window. If both devices declined equally the ratio holds and there is no signal. Only raise a DEVICE GAP card if the ratio worsened by ~10%+ (e.g. mobile/desktop CVR went from 0.80 to 0.72).
- Channel benchmarks: Use the most specific baseline available. Preferred: peer group (paid vs paid, email vs email, social vs social). Do not use site average to benchmark individual channels — each channel type has a structurally different CVR by design (email converts high, paid social converts low) so site average comparisons are misleading in both directions. Use site average only for country comparisons. Flag a channel based on underperformance relative to its peer group, or a CVR that worsened notably week-over-week.
Identify up to 3 signals to follow up on — these become the priority cards.
| Signal | Suggested follow-up |
|---|---|
| High-traffic country well below site CVR | Landing page breakdown for that country |
| Country with near-zero CVR | Landing page breakdown, or check for a broken campaign sending traffic there |
| Channel converting notably below its peer group, or CVR worsened week-over-week | Campaign breakdown to find which campaigns drag the channel down |
| One channel converting notably better than comparable channels | Deeper breakdown by campaign or source to find scaling opportunities |
| Mobile/desktop CVR ratio worsened by ~10%+ week-over-week | Week-over-week breakdown by device to confirm the trend |
| Rev/session and CVR moving in opposite directions | Flag as likely a currency difference — may not need a follow-up query |
Before running follow-ups, write a short plain-language transition referencing actual numbers and segment names. Example: "Canada's CVR is 0.3% vs your 2.1% site average, so I'm going to look at which landing pages that traffic is hitting." Then auto-run follow-ups (preferred for broad prompts) or ask first if they're a significant departure from what was asked.
Step 3 — Deeper follow-up queries
Run only the follow-ups the signals call for. This is the first point where minimum session count thresholds apply — the broad overview runs without them so no segments are filtered out prematurely. Calibrate thresholds to store volume:
500K sessions/month → ~0.1–0.2% of total
- 50K–500K → ~0.3–0.5%
- <50K → keep very low or skip
Campaign breakdown — when a channel's aggregate CVR hides per-campaign variation. Group by UTM campaign + UTM medium (limit 25). Measure: sessions, CVR, rev/session. Order by sessions ↓.
Landing page breakdown — when a country or campaign shows entry-point friction. Group by landing URL (limit 25). Filter to the relevant country or campaign. Order by sessions ↓.
Funnel stage breakdown — to find where a segment drops off. Group by funnel depth + one dimension (device, country, or UTM medium — whichever the overview flagged). Order by sessions ↓.