brands
ChatGPTA selection's brands, each with the standard measures for the part of it the selection holds, ranked by any of them: who trades in a place, a mall or an airport included, is the brands of that place's selection, and one brand is the selection of that brand alone (brand_quip,country_us). Each brand lists the names it trades under with their share of its spend, and its audience and cluster (audience); include_neighbors adds the brands its customers also buy from. audiences, clusters and first_seen_after narrow the list, and the answer counts the matching brands per audience and per category: that is how to discover brands, above all online ones, which have no stores to find them by.
categories
ChatGPTWhat a selection is made of: the standard measures per category, so a place's mix of trades, a mall's included, ranked by any of them.
cross_shop
ChatGPTWhere a merchant's own buyers also spend over a window: for each other merchant, the share of those buyers who shopped there (penetration) and what they spent per buyer. Scoped to the merchant's own place type by default (all restaurants, all gyms) -- the competitive set; each row names the other merchant's category. region (with optional locality) restricts the buyer base to households HOMED there. Use it when comparing a regional merchant against national ones: without it, a national chain's penetration is diluted by buyers who live nowhere near the regional merchant.
cross_shopping
ChatGPTWhere a cohort's customers spend, against a comparison group (vs): one row per category, or per brand of the targets you name. Each row gives both groups' reach, spend, transactions and share of wallet, with indices (100 = vs). Rows rank by sort: reach by default, where else the customers buy, the common question; wallet_share for where their money goes; wallet_share_index for what sets them apart from vs, with min_reach to keep a handful of customers from leading. Read an index with its reach: 142 on a reach of 0.0014 is a few customers, not an insight. Nothing is left out as obvious: the cohort's own brands and categories are rows too, one row per brand, for the caller to drop. Comparison groups: all US ({living_in: "country_us"}), residents of the cohort's place (for a cohort limited to a place), a category's buyers ({selection: "category_...", min_spend: 200}), or another cohort. Several calls build a table of several groups.
customer_homes
ChatGPTWhere a cohort's customers live, by region, metro area and city at once (one read of their homes), each place with its share of the customers and an index against a second cohort's share living there (default every member of the panel). Customers without a trusted home are not placed.
distribution
ChatGPTHow a cohort's customers spread over four measures, all from one read of the cohort: spend at its selection in the year ($100 bands to $3,000 and over), orders there in the year (bands of one, 20 and over, and a 0 band for the customers who bought nothing there in the year, as lapsed customers have not), the share of their category wallet it takes over the two years (10-point bands; its quadrants are rfm_segments' wallet mix), and spend everywhere over the two years ($10,000 bands to $300,000 and over, the measure the cohort's wallet filter bands). Every band gives its share of the customers and of their spend at the cohort's selection, so one call also says what share of the spend the customers buying once or twice a year bring; each measure's quintiles cut it into five equal groups, the edges for wallet or min_spend. With at, spend, orders and wallet share count at those targets instead (a brand's buyers by their vet visits a year); with vs, a second cohort on the same bands.
districts
ChatGPTAudience districts: clusters of stores named after the trait their shoppers lean to most against the shoppers of the country's typical store, and after the neighbourhood, street or town that holds most of their spend; a district can reach well beyond that place (named_after_share, extent_km). A district describes who shops there (the panel members buying at its stores over the prior year and the year), not who lives there: how many, who they are (the leading traits), what they spend, the busiest streets, and the brands its shoppers single out. Most shoppers first. A district is a place: its id selects its stores, and the online spend of the customers living inside it, in any selection.
fetch
ChatGPTFull profile for one merchant: headline metrics (revenue, transactions, customers, stores, growth, basket, visit cadence, geography) plus its RFM customer segments, as a readable document. The segment table is the rfm_table grain (the merchant's store selection over a fixed 24-month window), not rfm_segments' whole-merchant grain -- their totals differ by design. Takes an id from search. For filtered or comparative analysis use search_merchants and rfm_table directly.
find_brands
ChatGPTResolves words to brand ids, each with its label, what tells it apart and its clusters, never a figure. Matching ignores case and punctuation ("dicks" finds DICK'S), matches fragments, and finds a brand by any name it trades under ("walmart.com" finds Walmart). Given a cluster, category or group id instead of words, it lists the brands in it, most customers first: a cluster's brands, or every brand trading in a category or group.
find_categories
ChatGPTResolves words to category ids, each with its label and group. Matching ignores case, matches words from their start, and follows retired category names to the categories that took their spend ("grocery" finds Supermarkets, which took the retired Grocery store's). Given a group id instead of words, it lists the group's categories, most customers first. A category holds the whole spend of the stores and websites filed under it, not what was bought: a big-box chain's groceries are in Big-box & general merchandise, not Supermarkets (Walmart, Target). To count such a chain in a trade, add its brand (category_supermarkets,brand_walmart), which adds all of its spend.
find_places
ChatGPTResolves words to place ids of every kind: countries, regions, metro areas, cities, audience districts, and venues (enclosed malls, airports, named retail areas). Matching ignores case and punctuation and matches fragments; a trailing state or country says where the words before it lie ("Chicago, IL" is Chicago in Illinois). Labels repeat (London is a region, a metro and a city; Westfield is a town in several states and a mall in London): kind and within tell them apart.
guide
ChatGPTHow Iceflower's ids, selections, cohorts and figures work: the rules every other tool assumes. Call it once at the start of a conversation, before the other tools, unless your client already showed you Iceflower's instructions.
local_buyer_profile
ChatGPTThe households homed in a city (locality plus region) or a whole state: how many they are, their panel-spend distribution, and the behavioural dimensions and spend categories where they run FURTHEST above the panel as a whole (index 100 = panel average). Rows are the most over-indexed entries, not every category -- an everyday category that indexes near 100 (groceries) is usually absent because it is unremarkable, and the summary note states the exact selection rule. Grounds a question like "who are the buyers around here" in the actual local base, rather than the customers of any one merchant.
mall_compare
ChatGPTPeer centres for a given shopping centre -- optionally within one state or one revenue tier -- each with its revenue, customers and biggest tenants by revenue. Peers share the subject's venue type (a shopping centre compares against shopping centres, an airport against airports, an open-air retail area against retail areas) and tiers are computed within that type; summary.subject reports the subject's own tier and size. Use it to see what comparable centres host that this one does not.
mall_lookup
ChatGPTOne shopping centre's revenue, customers and store count, plus its tenant list with each tenant's revenue and location, biggest tenants first. Venues are mapped outlines: enclosed malls, airports, and named open-air retail areas (retail parks, pedestrian shopping quarters). A loose name resolves to the closest venue and the answer says which one it used. Figures cover the stores inside the venue's mapped outline, so scope varies with how the outline is drawn. When summary.tenants.truncated is true, page through the rest with offset. Tenant rows carry the same store_id values as store_locations, so the two answers join on it.
merchant_summary
ChatGPTOne row per named merchant from the merchant directory: category, mapped physical store count (0 for online merchants), panel transactions, revenue, customers and average transaction. The cheapest way to size a brand before asking anything else about it. Names that do not resolve are listed in summary.not_found and the rest still answer.
merchants_in_area
ChatGPTMerchants with physical stores in a city (locality plus region), across a whole state, or across a metro area (a US Census MSA or a UK Travel To Work Area), ranked by store revenue: store count, transactions and revenue. An optional category filter narrows it. The way to find a brand's local competitors. Give either region (optionally with locality) or metro.
monthly_series
ChatGPTA cohort's monthly spend and transactions per panel member, by default the 24 months of the prior year and the year, at its own selection or at the targets in at. from reaches back to January 2024; months before the prior year count only the cohort's people, who bought there in the two years, so they are not all of a selection's spend then (a cohort of residents is everyone living there). One line per call: a call per brand, store or group compares them, and each line divided by its first month compares their growth whatever their size. A cohort of residents (living_in alone) has no selection of its own; without at, its line is their spend everywhere.
personas
ChatGPTThe persona run stored for a selection, if one was generated in the app: per segment group, the cluster of customers that segment forms in a district or city, the traits that set it apart, and the persona written for it. Runs are made for brands or categories, everywhere or in countries, regions or metro areas, and for stores, with both channels; a selection naming one channel reads the run when the other holds none of its spend. Cities, districts and venues have no runs of their own. This tool only reads; a run built on an earlier data month comes back stale.
places
ChatGPTWhere a selection trades: the standard measures per region, metro area or venue, by where its stores are; online spend counts where its customers live, so places also says where a brand sells online (a venue holds only its stores). level is the grain. A venue's tier ranks it within its kind and country (Trophy, then A to D), so its peers are the venues of its own kind and tier; who trades inside one is brands with the venue's id as the selection.
retention
ChatGPTHow a cohort's new customers come back, month by month since their first purchase at its selection. A new customer's first purchase there falls in the signup months (from, to) after at least 12 months in the panel without one, so a customer who buys there once or twice a year is not taken for a new one; a customer back after a longer gap counts as new. Month 0 is the first purchase's month; each month counts the new customers still followed then (in the panel that month or later), so one who left the panel is not a lost customer, and the later months rest on the earlier signups alone (observed_share). Per month: the share who bought at the selection that month, the share who had bought there again since month 0, and their spend there per customer. A yearly fee or renewal shows as a rise at month 12. by_quarter adds each calendar quarter's signups as a curve of its own with its reach: whether newer customers stay better, and whether fewer arrive. The cohort's filters apply but segments and min_spend, which are measured over the year and so would pick new customers by how they turned out.
retention_curve
ChatGPTMonthly-active retention for a merchant's sign-up cohort: the share of households that made a first purchase together and are still active, by months since. The cohort window adapts to the panel's coverage.
revenue_flow
ChatGPTWhere a cohort's customers moved their spend: for the same customers, spend at the cohort's selection and at up to 15 competing brands in the prior year against the year, the two years the cohort spans, including spend that stopped or started (none). A competitor's spend counts only in the selection's own categories, the ones its spend falls in: against a pet store, a supercenter's spend does not count, its pet food included, as a store's spend all falls in its own category. Spend moving between a brand's own stores, websites and apps stays with the brand. bridge is the answer in short: the selection's change and what made it, net per competitor.
revenue_timeseries
ChatGPTMonthly revenue, transactions and customers per merchant over a window. Optionally scoped to the stores in one city (locality plus region) or a whole state (region alone); with neither it is national and covers all channels.
rfm_segments
ChatGPTThe standard RFM segment table for one whole merchant: per segment, customers and revenue with their shares, lifetime value, average order value, change against the year-earlier window and the share-of-wallet mix. For a region, a category or a map viewport rather than a whole brand, use rfm_table.
rfm_table
ChatGPTCustomer segmentation (RFM: recency, frequency, monetary) for a merchant, a region, or any combination of the two. Returns the ten standard segments -- Champions, Loyal, Potential Loyalists, Recent, Promising, Need Attention, About to Sleep, At Risk, Can't Lose, Hibernating -- with each segment's customer count and share, revenue and share, lifetime value, average order value, visit-count and days-since-last-visit ranges, and its share-of-wallet mix. Share-of-wallet splits each segment's customers four ways by how much they spend and how much of their spend at the merchant's place type (restaurants, gyms, pet stores ...) the merchant captures: Low Value (spends little, small share), Barnacles (spends little, large share), High Potential (spends a lot, small share -- the growth opportunity), True Loyalists (spends a lot, large share). At least one filter is required. Filters combine: merchants AND categories narrow the brand side, countries, regions AND metros narrow the geography, and a bbox narrows to a map viewport. Without a merchant filter, only physical stores are covered (a category or region on its own never includes online merchants). Use rfm_segments instead for ONE whole merchant: it reports year-over-year movement per segment and lets you move the window, neither of which this tool can do. This one is for any selection smaller or broader than a brand -- a region, a category, a viewport. Their totals for the same merchant differ because they are built from different grains; that is expected and not a contradiction to report.
search
ChatGPTFind merchants on the Iceflower consumer panel by name. Returns matching merchant identifiers to pass to fetch for the full profile, or to search_merchants and rfm_table for specific metrics. Use search_merchants instead when you want metrics, filtering or ranking -- this tool only resolves names.
search_merchants
ChatGPTFind merchants on the Iceflower consumer panel and read their headline metrics: revenue, transactions, customers, store count, revenue growth year over year, average basket, typical days between visits, and where their stores are. Ranked by transaction volume, highest first. Every filter is optional -- with no filters this returns the biggest merchants on the panel. A brand that sells in stores AND online is two merchants here, one per channel ("Walmart" and "walmart.com"), each with its own metrics; linked_channels names the other side. Add them together for a brand total, and expect the web app's merchant cards to show that combined figure under one name. Filter with channel to see only one side. Metrics cover a rolling multi-year window of US consumer card transactions.
spend_order_histograms
ChatGPTHow much, and how often, households buy at a merchant over a trailing window: buyer households binned by annual spend ($100 bins) and by annual order count, with the mean and median of each.
store_locations
ChatGPTIndividual stores for the named merchants -- address, latitude/longitude, transactions and revenue -- ranked by revenue, optionally narrowed to one city, state or metro area (a US Census MSA or a UK Travel To Work Area). store_id is an opaque per-store identifier; treat it as a whole, never parse it. mall_lookup tenant rows carry the same values, so the two answers join on it.
stores
ChatGPTA selection's physical stores with address, coordinates and the standard measures, ranked by any of them. Online spend has none: places gives a brand's online figures where its customers live, customer_homes where those customers are. A store's id is opaque after its kind; pass it back in a selection.
summary
ChatGPTA selection's figures in total: spend and transactions per panel member, reach (the share of the panel's members who bought), the change against the year before, spend per customer, stores, the customers' segment families and the trait they lean to most. brands, stores, categories and places give the same measures row by row.
trait_profile
ChatGPTWhat sets a cohort's customers apart: the twenty customer traits as indices against the panel, and with vs also against a second cohort: a brand's buyers against its category's buyers, one group of stores against another. Per segment, rfm_segments with include_traits.
cross_shop
Claudefetch
Claudelocal_buyer_profile
Claudemall_compare
Claudemall_lookup
Claudemerchant_summary
Claudemerchants_in_area
Clauderetention_curve
Clauderevenue_timeseries
Clauderfm_segments
Clauderfm_table
Claudesearch
Claudesearch_merchants
Claudespend_order_histograms
Claudestore_locations
Claude