explōrātiō looks at your sales history and identifies what repeats in the same basket — pairs and combinations that become a POS combo, a digital bundle, or an argument in your next supplier negotiation. You receive a report with the recommendation ready to use.
You send your transaction history¹ in the format you already have — POS, ERP, or e-commerce platform exports. We return a report with basket combinations that repeat, ranked by link strength, with a clear indication of where the evidence is strong and where it is still too early to conclude.
¹ assuming bases without classification errors; otherwise we will assess whether you need Data Clean
The study arrives ready to use — you receive the list of repeating combinations, ranked by link strength, with the recommendation behind each one. It works for the team that builds shelf layout, for whoever designs digital bundles, and for the next conversation with the manufacturer.
Your recommendation engine’s “customers who bought X also bought Y” is surfacing rare pairs and pushing them to everyone. We return a prioritized list, with link-strength measures, ready for your team to integrate — no more blind bundle tests.
When the question is what belongs in the same basket — by store, region, or banner — and the decision needs to stop relying on category gut feel. The study shows patterns that repeat everywhere and patterns that hold only in one region, without blending everything into an average that hides the difference.
When the question is what to keep, consolidate, or cut in the mix. The study separates the product that drives the sale from what sits beside it (complement) from the product that competes for the same customer with another in your own portfolio (substitution) — so you do not cut an item thinking it is fat when it is actually holding the basket together.
Understand which products your customers buy in the same cart — with or without repeat purchase.
Five pieces in the report. Each one exists to answer a concrete question that will come up in the meeting with sales, trade, or leadership — not to pad slides.
List of basket combinations that repeat, ranked by how much each pair actually reinforces the other (not just how often they appear together, which can be a popular-product effect). Answers “does buying A increase the chance of buying B?” with a number — not analyst gut feel.
Visual map of the product network. At a glance you see which items are basket “centers,” which are satellites, and which groups form. A fast way to decide which products can be offered in a promotion.
Same base, two readings: products that appear together in the basket (if one grows, the other tends to come along) and, when your data has customer identification, products that alternate across different baskets from the same customer (one eats the other’s sale). Avoids the classic mistake of cutting a SKU that looked like fat but was holding the basket together.
When the scope calls for more rigor — large catalog, very dense base, supplier who will audit — we add extra statistical indicators (Zhang, Jaccard, Kulczynski, certainty factor) that filter spurious association. Each is defined in the glossary; in practice, it is more confidence to defend the recommendation.
Every rule in the report comes in CSV, with the parameters and cuts we used. Your BI, trade, or CRM team can redo the reading, plug into the recommendation system, or keep testing — without relying on what was said in the meeting.
Six steps, from receiving the base to the readout meeting. Everything is agreed upfront — thresholds, cuts, timeline.
From briefing to final delivery
The thresholds that filter which rules enter the report are agreed in the briefing and recorded in scope. When your base includes repeat-customer identification, we add alternation analysis across baskets to separate complement from substitution. If the base needs treatment before the study, we add Data Clean timeline and say so plainly.
Each scenario describes a concrete decision — shelf combo, digital bundle, manufacturer conversation, mix rationalization, cross-banner readout — and how the report arrives to support it. Illustrative scenarios to show how explōrātiō can help.
Illustrative scenarios · do not represent explōrātiō clients
Three study sizes, one-time delivery. Price varies by how many products and categories enter the analysis. Before you commit, we look at a sample of your base in the initial conversation — if your case fits a smaller package, we say so.
R$ 4.557
R$ 11.947
R$ 18.947
SKUs and categories count toward the package limit — anything beyond becomes an add-on or a new study. For agencies that want to deliver the study under their brand, and for data teams that want only the technical cut to integrate, we have specific formats — just ask. Transaction volume well above typical may extend the timeline, not necessarily the package.
A study is about scope, and scope sometimes shifts — a new category enters, another banner appears, you want an extra cluster. Instead of hiding the rule, we keep the table in plain sight: what triggers an add-on, what becomes a new study, and what each case costs on top of the package you signed.
+35%
Change up to 25% of signed scopeEx.: Focal scope (R$ 4.557) → R$ 1.595.
+50%
Change above 25% of scopeEx.: Focal scope (R$ 4.557) → R$ 2.279.
25%
Limit by added SKUsEx.: Focal scope — 80 SKUs. +20 (25%) → targeted. 21st extra SKU (≥30%) → broad.
25%
Limit by added banners or categoriesEx.: Full study — 3 categories. +1 category (25%) targeted; 2nd extra (50%) → broad. Applies to extra banner/channel.
30d
One round after formal deliveryOne round after formal delivery (Full study and Multi-banner portfolio). Focal scope: initial briefing only.
New
Outside add-on — separate quoteAnother banner or quarter, granularity (SKU→EAN), substitution map without customer ID, new category cut, or joint study with elasticity — quoted separately.
Criteria A and B measure % of scope changed · higher index prevails · targeted add-on +35% or broad +50% on package · all within signed scope
CSV or export from your ERP/POS with at least sale ID and product (SKU or standardized description). Optional: banner/store and value. We accept ; separator and common Brazil encoding — we align in the briefing.
No. It is a delivered study with fixed timeline and scope. You receive static PDF and CSV, not dashboard access or login. A new round (another banner, another quarter) is contracted separately.
Confidence answers “when A appears, what is the chance B appears too?” Lift answers “is this association stronger than what would happen just because B is popular?” We use both because a pair can have high confidence and low lift when B is too frequent. Full definitions in glossary terms 05 and 06 below.
Complement (halo) shows when two items tend to appear together in the same basket — that is the default readout of the study. Substitution requires the base to have repeat-customer identification to observe alternation across different purchases from the same customer. This is explicit in scope before we run. See glossary terms 07 and 08.
Apriori requires minimum volume per pair — in the briefing we define support and scope (one banner, one category). If the base is too thin, we say so before closing scope, not after delivery.
Transfer via secure channel; use only to run your study. Nothing is trained for other clients or reused. After delivery, we agree retention and deletion per your policy.
We count distinct SKUs that enter the basket matrix after cleaning, or categories when scope is by family (e.g., entire oral care). Focal scope: up to 80 SKUs or 1 category. Full study: up to 250 SKUs or 3 categories. Multi-banner portfolio: up to 600 SKUs or 5 categories. If you exceed the signed package, we agree add-on or new study before we run.
After scope and data are closed: Focal scope in 10 to 15 business days; Full study in 20 to 25; Multi-banner portfolio in 30 to 40. A very messy base or scope that grows midstream may extend the calendar — we align in the briefing.
Technical terms that appear in the report, CSV, and readout meeting. It is here because people who operate data (BI, trade, category) will want to open and audit — if you only receive the recommendation, you do not need to read it.
Study of which products appear together in the same transaction (cart, receipt, order). The goal is not to forecast total sales — it is to find repeating associations for mix, kit, shelf, cross-sell, and supplier negotiation.
Classic association-rule algorithm: starts from the most frequent items and generates combinations that meet a minimum support. In the study we use implementation via mlxtend, with parameters declared in scope and in the CSV workpaper.
Each row is a sale ID; each column is a product (SKU or standardized item). The value is 1 if the item was in that cart, 0 otherwise. It is the Apriori input — we do not use quantity or value in the basic rule unless explicitly agreed in scope.
Frequency with which the item set (or pair A,B) appears in transactions: P(A ∩ B) in the scoped universe. A high threshold reduces noise and keeps only robust patterns; a low threshold brings more rules, including weak ones. We define the minimum in the briefing — not after delivery.
Conditional probability P(B | A): among carts that have A, in how many does B also appear. Answers “if the customer took A, what is the chance of taking B?” Useful for operations and assortment; alone it can overstate very frequent items — that is why we cross with lift.
Ratio between observed confidence and what would be expected if A and B were independent: lift = P(B|A) / P(B). Lift > 1 indicates positive association (B “rises” when A is in the cart); lift ≈ 1, weak pattern; lift < 1, negative association. Trade and category often prioritize lift at the supplier table.
When product A pulls B into the same basket — they appear together more often than chance. It is the classic Apriori signal (lift > 1, minimum support). Supports combo recommendation, shelf adjacency, and digital bundle (study scenario 01).
When A and B rarely appear in the same basket, but the same customer alternates between them across different purchases — substitution signal, not complement. Requires repeat-customer ID. In assortment and pricing literature, cross-effects are often read across distinct categories: between brands, between pack sizes, between channels, between price tiers (good/better/best), and between regions or demand clusters.
Directional implication: presence of A associated with presence of B in the same cart, with metrics calculated for that pair. It is not causality — it is co-occurrence in history. The report delivers a rule ranking, not only a symmetric pair list.
Indicators beyond lift and confidence to compare rules when scope calls for more statistical rigor. Jaccard measures set overlap; Kulczynski balances A→B and B→A asymmetry; Zhang and certainty factor help filter spurious associations in dense bases. Included in Full study or Multi-banner portfolio packages, as agreed.
Network visualization: nodes = products, edges = rules above agreed thresholds (thickness or color can reflect lift or confidence). For committee, supplier, and internal alignment — included in the PDF, without relying on tool login.
When the base includes store or channel identifier, we run Apriori per cut or compare patterns across banners. Answers whether the kit holds nationally or only in one region — different from blending all stores into one ranking without warning.
Alternative algorithm to Apriori for mining association rules — usually scales better on very large catalogs. The study uses Apriori by default (via mlxtend); if transaction and SKU volume require it, we agree FP-Growth in scope, with the same metrics and deliverables package.
Auditable export with rule table, metrics, support/confidence/lift thresholds, and applied cuts. Lets the BI or trade team redo the reasoning and plug into CRM or the recommendation engine — without relying on what was said in the meeting.
Few transactions, catalog too wide with unrealistic support, or product too rare to form a stable pair. In those cases the report states the caveat and suggests a smaller cut, longer window, or Data Clean before a new round — we do not deliver a pretty rule without a base.
Basket study delivered within signed scope. Ready for the next category meeting, for trade to defend at the supplier table, and for CRM to integrate — no “almost ready,” no endless revision round, no half delivery.
The initial conversation is to look at a sample of your base and say plainly: there is a basket study here that solves the problem, or it is still too early. If yes, we sketch scope, cut, and timeline on the spot. If not, you still leave with the diagnosis.