Query Architecture Review (Onboarding)

Query Architecture Review (Onboarding)

1. 01

   Inventory the client's brand names, product lines, executive names, and common misspellings before writing a single query

   Pull the list from the client's own style guide and site search logs rather than guessing; misspellings and legacy product names generate a large share of genuine mentions.

2. 02

   Separate monitoring streams by intent: brand health, competitor share of voice, category topics, and crisis triggers

   A single catch-all query set makes sentiment averages meaningless because a product complaint and a competitor comparison get scored the same way.

3. 03

   Write Boolean queries with explicit exclusions for homonyms and unrelated industries

   A brand named after a common word will pull noise from unrelated verticals; test each query against a 7-day historical window and count false positives before sign-off.

4. 04

   Confirm language and regional coverage against the client's actual markets

   Platforms differ on which languages and geographies they index well, so document the gaps in writing instead of discovering them during a launch week.

5. 05

   Set the sampling and volume ceiling per stream and record it in the account brief

   When a stream exceeds the plan's mention cap, the platform silently drops data; note the cap so analysts know when a quiet week is a coverage artifact.

6. 06

   Define the escalation threshold that moves a mention from dashboard to human review

   Agree on the trigger with the client in advance, for example a verified journalist or an account above a follower count, so nobody is paging anyone at 2am over a bot post.

7. 07

   Run a two-week parallel test against the client's previous monitoring method and reconcile the delta

   Differences in mention counts between old and new setups are the fastest way to lose trust in week one; explain the gap before the first client-facing report.