Memory

What Torok knows about your business — the lens every analysis looks through. Add things below, or just say "remember this" in any chat.

Ground rules, the profile and targets ride in every chat. Considerations are pulled in only when a question calls for them. Turning an entry off keeps it here but stops Torok using it.

Acme Outdoors sells premium camping gear DTC (tents, sleep systems, accessories). Contribution margin ~62%, AOV ~$180. Peak seasons: May–Aug and the holiday window. Wholesale exists but ads are judged on DTC revenue only.

SL Updated Jul 8 · in every chat, for everyone

Definitions the whole workspace shares. These ride along in every chat, for everyone — so when anyone asks about performance, the org gets the same numbers through the same lens.

Revenue means UTM-attributed Shopify revenue. Platform-reported revenue is reference only — every ROAS Torok quotes uses the UTM number.
Every chatdefinitionSL Jul 6 · from chat: Why don't Meta and Shopify match?
Inventory questions use Shopify "available", never "on hand" — on-hand includes committed units and overstates what we can sell.
Every chatshopifyDK Jul 10
Midwest runs on a lower bar: IL / WI / MN convert better on tents, so they may run down to 3.0x where the blended target is 3.5x.
Every chatgeo: midwestSL Jul 14 · from chat: Which states convert best?

Like ground rules, targets apply in every chat for everyone.

Blended target: 3.5x ROAS at 62% margin. Treat this as the bar unless a narrower target below applies.
blendedSL Jul 8
Brand search should clear 8x; flag anything under 6x — it usually means a feed or tracking problem, not demand.
google_adsSearch | BrandSL Jul 8
Meta prospecting is allowed to run at 2.5x while we scale new creative — don't recommend cuts on the blended bar alone.
meta_adsprospectingDK Jun 30

Context Torok applies when the question calls for it. Each one is either shared with everyone or just yours.

Weekend CPCs on Search | Generic run ~18% cheaper than weekdays — worth a dayparting look before any budget cut.
google_adsProposed by Torok from yesterday's chat
Bing brand clicks are mostly navigational — its ROAS overstates incrementality. Never recommend scaling on Bing ROAS alone.
bing_adsDK Jul 2 · from report: UTM Audit — July
Retargeting | 30d supports Q4 gift buyers — don't propose pausing it Oct–Dec even when in-month ROAS dips.
meta_adsseasonalSL Jun 28 · from chat
Show week-over-week deltas as percentages, not absolutes, and lead with the table.
Just meformattingSL Jul 15

The workspace's shared context folder — files and sheets that are canonical for every analysis. Everyone's chats reference the same versions, so nobody gets different numbers.

margin_by_category.xlsx
38 KB · uploaded Jul 1 by Dana · the margin table every ROAS threshold is computed from
Canonical
SKU planning — 2026 ⧉
Google Sheet · re-synced nightly · launch dates + inventory buys the seasonal analyses lean on
Canonical
promo_calendar.md
6 KB · updated Jul 18 by Sam · sale windows, so demand spikes aren't misread as performance
Canonical
Markdown, CSV, spreadsheets, or a linked Google Sheet. Anything here is available to every chat and every Analyze run in the workspace.

Ground rules, the business profile, targets, and context files are the workspace's shared lens — every chat and every Analyze run works from the same ones, so no two people get different numbers. Considerations are pulled in when relevant. Torok cites the memory or context it applied whenever it changes a number or a recommendation, and each entry traces back to the chat or report it came from. Keep it curated: a bloated memory makes every answer slower.