← Archimedes01 / 04
Proprietary engine — module 01

Market analysis

Archimedes reads a category the way an analyst would — only across every competitor, every week, without getting tired.
Market analysis — Archimedes module at Atlas Marketing
The module ingests search-demand data, marketplace and retail pricing, social conversation, review text and paid-media coverage for a defined set of 8–20 competitors. Each source is normalised into a weekly panel keyed by brand, product line and geography, so numbers from Semrush, Meta's ad library, GA4 and a client's own sales export can sit in the same table without being compared unfairly.
On top of that panel we run share-of-search estimation, price-ladder reconstruction and topic modelling on review and comment text. Share of search is used as a leading indicator of market share — in most of the categories we work in it leads reported share by roughly one to two quarters, which is the whole reason the module exists.
Everything is versioned. When a number changes, we can show which source moved and when, and every chart in a client report links back to the query that produced it.

Specification

Refresh
Weekly, with a quarterly deep review
Competitor set
8–20 direct and adjacent brands
History required
24 months minimum for reliable seasonality
Setup time
2–3 weeks from data access

Data inputs

  • Search volume & clickstream (Semrush, Google Ads, GSC)
  • Marketplace and retail pricing feeds
  • Meta & Google ad libraries
  • Social conversation and review text
  • Client first-party sales and CRM exports

How it runs

4 stages
  • 01

    Ingest

    Scheduled pulls from search, marketplace, social, ad-library and first-party sources into a raw store, timestamped and immutable.

  • 02

    Normalise

    Currency, geography, category taxonomy and brand entity resolution — including fuzzy matching of misspelled brand mentions.

  • 03

    Model

    Share of search, price elasticity by SKU tier, sentiment and topic clustering on unstructured text.

  • 04

    Review

    An analyst validates outliers before anything reaches a client. No automated number is published unread.

Methods inside the model

4 methods
  • 01

    Share of search

    Brand query volume as a share of category volume, smoothed on a 4-week rolling window and used as a leading proxy for market share.

  • 02

    Price-ladder reconstruction

    Observed prices clustered into tiers to show where a brand actually sits versus where it believes it sits.

  • 03

    Topic modelling

    Embedding-based clustering of review and comment text to surface the purchase drivers and objections a survey would miss.

  • 04

    Entity resolution

    Fuzzy matching so 'Mikel Coffee', 'mikel' and misspellings resolve to one brand across all sources.

What it produces

  • /Weekly category panel (brand × week × geography)
  • /Share-of-search index with 8-quarter history
  • /Price-ladder and positioning map
  • /Driver and objection clusters with sample verbatims
  • /White-space report refreshed quarterly

Honest limits

  • Share of search is a proxy, not a measurement — it is calibrated against reported share wherever a client can provide it.
  • Categories with very low search volume (under ~1,000 monthly queries) are modelled with wider confidence bands, or excluded.
Next modulePredictive modelling →
Talk to us about Archimedes →