Master Outline & Study Guide

Tableau — Olist E-commerce Operations & Customer Experience Analysis

Purpose of this document: A complete, self-contained reference for studying this portfolio project. Use it to prepare for interviews — to explain what was built, how it works, why each decision was made, and to answer probing technical questions. When studying with Claude, this is the file Claude reads to teach and quiz you.

One sentence to anchor everything: This project is a five-dashboard Tableau analysis of the Olist Brazilian e-commerce marketplace (99,441 orders) — built on an 8-table relationship model — examining order fulfillment, revenue, customer reviews, and marketplace dynamics, and synthesizing all of it into an interactive executive summary with LOD expressions, parameters, and forecasting.

Table of Contents

  1. The 30-Second Pitch
  2. Why This Project Exists (Context)
  3. The Tech Stack
  4. The Dataset — Olist Brazilian E-commerce
  5. The Data Model — 8 Tables via Relationships
  6. Calculated Fields & Advanced Tableau Techniques
  7. Dashboard 1 — Order Fulfillment Performance
  8. Dashboard 2 — Revenue & Sales Performance
  9. Dashboard 3 — Customer Experience & Review Quality
  10. Dashboard 4 — Marketplace Ecosystem (Products & Sellers)
  11. Dashboard 5 — Executive Summary
  12. Key Findings (Consolidated)
  13. Tableau Concepts to Know Cold
  14. Limitations & Honest Caveats
  15. Interview Q&A
  16. How to Walk Through This Project Live
  17. Glossary

1. The 30-Second Pitch

This project is a comprehensive Tableau analysis of the Olist Brazilian e-commerce marketplace — 99,441 orders spanning September 2016 to August 2018. It’s built as five interconnected dashboards that examine the marketplace from four perspectives — operational performance, financial health, customer satisfaction, and ecosystem dynamics — plus an executive summary that ties them together.

The whole analysis runs on an 8-table data model connected through Tableau’s relationship model (not joins), with calculated fields for delivery time and late- delivery flags, LOD (Level of Detail) expressions for month-over-month comparisons, parameters for interactive filtering, and Tableau’s native forecasting for a 3-month revenue projection.

The headline story: Olist’s core operations are strong — a 97% completion rate, delivery times that improved from 50+ days to 12.5, $15.8M revenue, 21% YoY growth — but there are clear pressure points: a 6.8% late-delivery rate above the 5% target, a strong link between delivery delays and falling review scores, and heavy revenue concentration in São Paulo. Every dashboard ends in business recommendations.

One-line version: “I built a five-dashboard Tableau analysis of 99,000 Olist e-commerce orders — on an 8-table relationship model with LOD expressions, parameters, and forecasting — covering fulfillment, revenue, customer experience, and marketplace health, all synthesized into an interactive executive summary.”

2. Why This Project Exists (Context)

The business problem. E-commerce marketplaces live or die on reliable fulfillment and customer satisfaction — late deliveries erode reviews, bad reviews erode repeat purchases, and the whole flywheel slows. Operations, finance, and customer-experience teams need visibility into delivery times, revenue trends, review-score drivers, and seller health to make decisions. This project builds exactly that visibility.

The four lenses. The project deliberately examines the marketplace from four interlocking perspectives, one dashboard each, plus a fifth that synthesizes them: operational performance (fulfillment), financial health (revenue), customer satisfaction (reviews), and ecosystem dynamics (products & sellers). The stated goal is to move beyond surface metrics and uncover the relationships between them — where delivery performance drives satisfaction, where geographic concentration is both a strength and a risk.

Why it’s a strong portfolio project. It’s the portfolio’s flagship Tableau project, and it demonstrates the tool’s full depth — not just charts, but a proper relationship data model, calculated fields, LOD expressions, parameter-driven interactivity, Top N sets, native forecasting, and dashboard actions. It also demonstrates analytical synthesis: five dashboards that each answer one question but compound when read together, ending in an executive monitoring surface. It’s published live on Tableau Public.

3. The Tech Stack

   
Tool Tableau Desktop (workbook published to Tableau Public)
Deliverable tableau_olist_ops_cx_v1_raw_load.twbx — a packaged Tableau workbook
Dataset Olist Brazilian E-commerce Public Dataset (Kaggle)
Techniques Relationship data model · calculated fields · LOD expressions · parameters · Top N sets · native forecasting · dashboard actions

The mental model. Tableau is doing three jobs here: it’s the data modeling layer (connecting 8 tables via relationships), the calculation layer (calculated fields and LOD expressions that derive metrics not in the raw data), and the visualization/delivery layer (5 dashboards). A .twbx is a “packaged workbook” — it bundles the workbook and its data into one file, so it’s fully portable.

One connection detail worth knowing: the data uses a live connection (rather than a static extract), which the project notes was chosen for real-time filtering and exploration during development.

4. The Dataset — Olist Brazilian E-commerce

What it is. The Olist Brazilian E-commerce Public Dataset from Kaggle — real (anonymized) order data from Olist, a Brazilian marketplace that connects small sellers to large e-commerce platforms.

Attribute Value
Total orders 99,441
Time range September 2016 – August 2018
Granularity Order-level, with customer, seller, product, payment, and review detail
Total revenue $15.8M
Average Order Value (AOV) $159.33
Average review score 4.09 / 5
YoY revenue growth ~21% (2017 → 2018)

The 8 tables:

  • Orders — the central fact table; one row per order, with the timestamps that drive delivery analysis (purchase, delivered-to-customer, estimated delivery) and order_status.
  • Order Items — one row per item in an order; carries price and freight.
  • Customers, Sellers — the two sides of the marketplace.
  • Products + Product Category Translation — product detail; the translation table maps Portuguese category names to English.
  • Order Payments — payment method and value (credit card, boleto, etc.).
  • Order Reviews — the 1–5 star review scores.

A Brazil-specific detail to know: boleto is a Brazilian bank-slip payment method — ~18% of revenue here — it matters because it serves underbanked customers who don’t have credit cards. The other ~80% of revenue is credit card (Brazilians lean heavily on installment payments). Knowing what boleto is signals you understood the market context.

5. The Data Model — 8 Tables via Relationships

A deliberate, defensible technical choice — know why it was made.

The 8 tables are connected using Tableau’s relationship model, not traditional joins. The relationships:

  • olist_orders_dataset is the central fact table (order_id primary key).
  • Customers link via customer_id; Order Items link via order_id (many-to-one with orders); Products link via product_id; the Category Translation links via product_category_name; Sellers via seller_id; Payments and Reviews via order_id.

Relationships vs. joins — the interview point. A traditional join physically merges tables into one flat result at one fixed granularity — which causes double-counting when you mix tables of different grain (e.g., joining order-level data to item-level data multiplies the order rows). Tableau’s relationship model is “smarter”: it keeps the tables separate and lets Tableau choose the right join and granularity per visualization at query time. So a chart counting orders aggregates at order grain, and a chart summing item revenue aggregates at item grain — from the same model, with no double-counting. That flexibility “across different levels of analysis” is exactly why relationships were chosen over joins.

Data preparation: calculated fields were built for delivery-time analysis, the late-delivery flag was built by comparing actual to estimated delivery date, and delivery/completion metrics were filtered to "delivered" orders.

6. Calculated Fields & Advanced Tableau Techniques

The metrics that drive the dashboards don’t exist in the raw data — they’re built with calculated fields. Know these.

6.1 Core calculated fields

  • Delivery Days = DATEDIFF('day', [Order Purchase Timestamp], [Order Delivered Customer Date]) — the number of days from order to delivery.
  • Late Delivery Flag = IF [Order Delivered Customer Date] > [Order Estimated Delivery Date] THEN 1 ELSE 0 END — 1 if the order arrived after its estimate.
  • Is Delivered = IF [Order Status] = "delivered" THEN 1 ELSE 0 END.
  • Order Completion Rate = SUM([Is Delivered]) / COUNT([Order Id]).
  • Late Delivery Rate = SUM([Late Delivery Flag]) / COUNT([Order Id]) (for delivered orders).

6.2 LOD expressions — the advanced technique

LOD (Level of Detail) expressions let a calculation run at a different granularity than the visualization it sits in. The project uses 8 FIXED LOD calculations to compute prior-month values — Previous Month Revenue, Previous Month Orders, Previous Month Avg Review, Previous Month Delivery Time — each pinned to a fixed level so it can be compared against the current month regardless of how the view is filtered.

Why LOD matters — the interview point. A normal aggregate in Tableau is computed at the granularity of the chart. But a month-over-month comparison needs two values at once — this month and last month — at a fixed level. A FIXED LOD expression computes a value at a granularity you specify, independent of the view, which is exactly what makes the prior-month and MoM % change calculations possible. LOD expressions are the single most “advanced Tableau” thing in the project.

6.3 The rest of the advanced toolkit

  • 17 additional calculated fields — the MoM % change percentages, display-text helpers with conditional ▲/▼ arrows, trend indicators (Improving/Declining/Stable), and color-coding helpers for conditional formatting on KPI cards.
  • Parameters — a Date Range parameter (5 preset time options driving global date filtering) and a Selected Metric parameter (4 options driving a dynamic “Key Insight” box). Parameters are user-facing controls that change what the dashboard shows.
  • Native forecasting — Tableau’s built-in forecasting engine produces a 3-month revenue forecast with automatic seasonal adjustment and a 95% confidence interval, plus a linear trend line.
  • Top N sets and dashboard actions for interactive drill-down.

7. Dashboard 1 — Order Fulfillment Performance

Business question: How is the marketplace performing operationally — order volume, delivery speed, late deliveries, order status?

The visuals: 4 KPI cards (Total Orders, Avg Delivery Time, Order Completion Rate, Late Delivery Rate), a monthly order-volume line chart, an average-delivery-time trend, a late-delivery-rate trend, a delivery-time distribution histogram, and an order-status breakdown.

Key findings:

  • 97% order completion rate — reliable fulfillment.
  • Dramatic growth — monthly order volume rose from ~400 (late 2016) to a peak of ~7,500 (early 2018).
  • Delivery speed improved enormously — average delivery time fell from 50+ days in early operations to stabilizing around 12.5 days by 2017–2018.
  • 6.8% late-delivery rate overall, with periodic spikes of 10–20%.
  • Early-operations stress — Aug–Nov 2016 showed 100% late delivery (the launch/ scaling phase); operations stabilized to a ~10% baseline by early 2017.

The teachable point. This dashboard establishes the operational baseline — and surfaces the project’s central tension: operations are strong overall (97% completion, big speed improvement) but the 6.8% late rate exceeds the 5% target, and the long tail of 45+ day deliveries is the thread the customer-experience analysis (Dashboard 3) later picks up.

8. Dashboard 2 — Revenue & Sales Performance

Business question: How is revenue growing, and where does it come from — by category, payment method, and geography?

The visuals: KPI cards (Total Revenue, AOV, Items Sold, Growth Rate), a monthly revenue trend, revenue by product category, revenue by payment method, and revenue by state.

Key findings:

  • 21.01% YoY revenue growth (2017 → 2018) — healthy expansion. Monthly revenue grew from near-zero (late 2016) to exceeding $1.2M by early 2018.
  • Diversified category mix — top 10 categories are relatively balanced; Health & Beauty leads but is only ~9% of revenue (no single category dominates).
  • Credit card dominates payment — ~80% of revenue via credit card (Brazilian installment-payment culture); ~18% via boleto (serving underbanked customers).
  • Heavy geographic concentration — São Paulo alone = 37% of total revenue; the Southeast region (SP, RJ, MG) drives 60%+ of sales.
  • Healthy AOV of $159.33 — customers make meaningful, not trivial, purchases.

The teachable point. Revenue is growing well and is category-diversified — but it is geographically concentrated. São Paulo at 37% is both Olist’s strength and its biggest strategic risk; the recommendation is geographic expansion into underserved northern and central-western states.

9. Dashboard 3 — Customer Experience & Review Quality

Business question: How satisfied are customers, and what drives their review scores?

The visuals: KPI cards (Total Reviews, Avg Review Score, 5-Star Rate, 1-Star Rate), a review-score distribution, a review-score trend, review score by delivery status, a review score vs. delivery-days chart, and top/bottom-10 categories by review score plus a category-performance scatterplot.

Key findings:

  • Average review score 4.086, with a strong 58.25% 5-star rate — high overall satisfaction.
  • But a meaningful 11.61% 1-star rate — a sizable poor-experience segment.
  • Delivery speed drives satisfaction — the project found a strong negative correlation (−0.65) between delivery time and review score: faster delivery → higher ratings; long delivery → lower ratings. On-time orders average significantly higher satisfaction than late ones.
  • Category gaps are real — bottom categories trail the average by ~0.2–0.5 points.
  • Sentiment stabilized around 4.0–4.3 after early-2016 volatility, though the recent trend is gently declining.

The teachable point — the most actionable finding in the whole project. This dashboard is what connects Dashboard 1’s delivery problem to a business consequence: the −0.65 delivery-time/review-score correlation means late deliveries directly cost customer satisfaction, and with an 11.6% 1-star rate, fixing logistics is the single highest- leverage lever. (Be ready to note it’s a correlation, not proven causation — §14.)

10. Dashboard 4 — Marketplace Ecosystem (Products & Sellers)

Business question: How healthy is the marketplace itself — product mix, seller distribution, concentration, and category-seller dynamics?

The visuals (three sections — products, sellers, integrated): product performance, seller distribution and growth, a seller quadrant (revenue vs. review score), and a category-state heatmap for white-space analysis.

Key findings:

  • Healthy, democratized marketplace structure — 2,970 sellers, and the top 10 sellers account for just 12.93% of revenue (low concentration — competitive, not an oligopoly).
  • Quality and scale coexist — the seller quadrant shows most sellers maintain 4.0+ review scores regardless of revenue level — platform quality standards are working.
  • Category diversification — the top 5 categories are 39.25% of revenue; the other 60%+ spreads across 66 other categories.
  • E-commerce-optimized product mix — the platform favors lightweight, affordable products (good shipping economics).
  • Single-item shopping behavior — only 1.142 items per order — customers buy targeted, not baskets → a cross-sell opportunity.
  • Steady seller onboarding (100–200 new sellers/month), and clear white-space opportunities in underserved category-state combinations.

The teachable point. This dashboard checks the structural health of the marketplace — and the verdict is positive (low concentration, broad seller base, quality maintained at scale). The category-state heatmap turns that into an action: target seller recruitment at specific empty category-state cells.

11. Dashboard 5 — Executive Summary

Business question: Can leadership see the whole marketplace’s health on one screen and drill down when needed?

This is the synthesis dashboard — and the most technically advanced. It consolidates Dashboards 1–4 into a single monitoring surface:

  • KPI cards with MoM % change indicators (powered by the 8 LOD expressions) — conditional ▲/▼ arrows and Improving/Declining/Stable trend labels.
  • A revenue trend with Tableau’s native 3-month forecast (95% confidence band) and a linear trend line.
  • Geographic drill-downs and a revenue density map.
  • Parameter-driven interactivity — the Date Range parameter refilters everything; the Selected Metric parameter drives a dynamic “Key Insight” box.

Key findings surfaced on this page:

  • Delivery alert — the 6.8% late rate exceeds the 5% target (tied to the −0.65 review correlation).
  • Orders/revenue divergence — orders grew +3.5% MoM while revenue fell −5.2% MoM → AOV compression (customers ordering more but spending less per order).
  • The 3-month forecast predicts recovery with an upward trend line — suggesting the −5.2% dip is seasonal, not a systemic downturn.
  • Despite that, declining review-score trend needs proactive intervention.

The teachable point. Dashboard 5 demonstrates the synthesis skill — five separate analyses distilled into one executive surface — and it’s where the advanced Tableau (LOD-driven MoM cards, parameters, forecasting) all comes together. It’s designed for a weekly executive review cadence, drilling into Dashboards 1–4 when a KPI breaches its threshold.

12. Key Findings (Consolidated)

Memorize the headline numbers.

Theme Finding
Scale 99,441 orders · $15.8M revenue · AOV $159.33 · 21% YoY growth
Operations 97% completion rate; delivery time improved 50+ days → 12.5 days
The problem 6.8% late-delivery rate vs. a 5% target — structural, not seasonal
The link −0.65 correlation between delivery time and review score
Satisfaction Avg review 4.09; 58.25% 5-star, but 11.61% 1-star
Geography São Paulo = 37% of revenue; Southeast = 60%+
Payments ~80% credit card · ~18% boleto
Marketplace 2,970 sellers; top 10 = only 12.93% of revenue (healthy, democratized)
Behavior 1.142 items/order — single-item shopping; cross-sell opportunity
Recent signal Orders +3.5% MoM but revenue −5.2% MoM → AOV compression

The five-part story: (1) Olist’s core operations are genuinely strong and scaled well; (2) but the 6.8% late-delivery rate is the one clear operational failure; (3) and it matters because late deliveries are strongly correlated with lower reviews; (4) revenue is healthy but geographically over-concentrated in São Paulo; (5) the marketplace structure is healthy and competitive (low seller concentration), with clear white-space for expansion. Logistics is the highest-leverage fix.

13. Tableau Concepts to Know Cold

Calculated field — a new field you define with a formula (e.g., Delivery Days, Late Delivery Flag); the metrics not present in the raw data.

Relationships vs. joins — a join physically merges tables at one fixed granularity (and risks double-counting across grains); a relationship keeps tables separate and lets Tableau pick the correct join and grain per visualization. This project uses relationships for exactly that flexibility.

Fact table / dimension — the central transactional table (Orders) vs. the descriptive tables linked to it (Customers, Products, Sellers).

Granularity / grain — the level of detail one row represents (order-level vs. item-level); mismatched grain is what causes double-counting in joins.

LOD (Level of Detail) expression — a calculation that runs at a granularity different from the view. FIXED computes at a level you specify, ignoring the view’s filters; used here for all the prior-month values. (INCLUDE and EXCLUDE are the other two LOD keywords.)

DATEDIFF — a Tableau function returning the number of date units between two dates; used for Delivery Days.

Aggregate functions — SUM, COUNT, COUNTD (count distinct), AVG.

Parameter — a user-facing control (dropdown, slider) that feeds a value into calculations; here, the Date Range and Selected Metric parameters.

Set (Top N set) — a dynamic subset of dimension members (e.g., the top 10 categories).

Forecasting — Tableau’s built-in projection of future values with seasonal adjustment and confidence intervals; here a 3-month revenue forecast.

Trend line — a fitted line (e.g., linear) overlaid on data to show direction.

Dashboard action — interactivity that links sheets (filter, highlight, drill on click).

KPI card — a visual showing one headline number, often with a MoM indicator.

MoM (Month over Month) — comparing a metric to the prior month; the % change.

Live connection vs. extract — a live connection queries the source in real time; an extract is a static snapshot. This project used a live connection for exploration.

.twbx — a packaged Tableau workbook bundling the workbook and its data into one portable file.

Tableau Public — Tableau’s free hosting platform for sharing interactive dashboards.

14. Limitations & Honest Caveats

Volunteer these — the project itself lists them.

  1. Truncated boundary periods. August 2018 shows a sharp order-volume drop consistent with incomplete data collection, and Aug–Nov 2016 has too few orders for reliable trends. Insights from those boundary months are directional, not definitive.
  2. No logistics provenance data. The dataset has no carrier identity, warehouse locations, or external context (holidays, promotions, weather) — so delivery-performance variation can’t be attributed to specific operational causes.
  3. Geographic granularity ceiling. Locations are city/state level only — no postal-code coordinates — so delivery distance can’t be measured, even though it’s a likely confound in delivery-time patterns.
  4. The review–delivery link is correlational. The −0.65 relationship is strong, but the data can’t isolate delivery experience from product quality or seller communication. It’s a strong correlation, not proven causation — be precise about this.
  5. Single marketplace, single country, single era. All findings are specific to Olist’s Brazilian marketplace in 2016–2018 — the payment mix (boleto), logistics realities, and consumer behavior are particular to that market and time.
  6. The recent MoM dip is one data point. The −5.2% revenue MoM is interpreted as seasonal (and the forecast supports recovery), but it’s a single month near a data-truncated boundary.

15. Interview Q&A

Practice these out loud.

Q1. Give me the overview of this project. “It’s a five-dashboard Tableau analysis of the Olist Brazilian e-commerce marketplace — about 99,000 orders from 2016 to 2018. The five dashboards cover order fulfillment, revenue and sales, customer experience and reviews, the marketplace ecosystem of products and sellers, and an executive summary that ties it all together. It’s built on an eight-table data model connected through Tableau relationships, with calculated fields, LOD expressions, parameters, and native forecasting. The big theme is that Olist’s operations are strong but late deliveries are hurting customer satisfaction.”

Q2. Why did you use Tableau relationships instead of joins? “Because the eight tables sit at different granularities — orders are order-level, order items are item-level, payments and reviews are their own grains. If I did traditional joins, mixing those grains would double-count — joining order-level data to item-level data multiplies the order rows. Tableau’s relationship model keeps the tables separate and lets Tableau choose the right join and the right granularity for each individual visualization at query time. So a chart counting orders aggregates at order level, and a chart summing item revenue aggregates at item level, from the same model, with no double-counting.”

Q3. What’s an LOD expression and where did you use one? “An LOD — Level of Detail — expression lets a calculation run at a different granularity than the chart it’s in. A normal Tableau aggregate is computed at the view’s level of detail. But for a month-over-month KPI card, I need this month’s value and last month’s value at the same time, at a fixed level. So I used FIXED LOD expressions — I built eight of them, for previous-month revenue, orders, average review, and delivery time. FIXED computes the value at a granularity I specify, independent of the view’s filters, which is what makes the MoM percentage-change calculations possible.”

Q4. What were your key calculated fields? “The core ones drove the delivery analysis. Delivery Days is a DATEDIFF between the purchase timestamp and the delivered date. Late Delivery Flag is an IF — one if the actual delivery date is later than the estimated delivery date, otherwise zero. Is Delivered checks the order status. And then Order Completion Rate and Late Delivery Rate are ratios built on those. None of those metrics exist in the raw data — they’re all derived in Tableau.”

Q5. What’s the single most important finding? “The link between delivery performance and customer satisfaction. The fulfillment dashboard showed a 6.8% late-delivery rate, above the 5% target. The customer-experience dashboard then showed a strong negative correlation, about −0.65, between delivery time and review score — late orders get meaningfully lower ratings. With an 11.6% one-star rate, that makes logistics the highest-leverage fix in the whole analysis: improving delivery directly protects reviews, and reviews drive repeat purchases. I’d add that it’s a strong correlation, not proven causation — the data can’t fully separate delivery from product quality.”

Q6. You found revenue down 5% but orders up — what does that mean? “That’s average order value compression — customers are placing more orders but spending less per order. Orders grew about 3.5% month over month while revenue fell about 5.2%. It’s a warning sign worth investigating — it could be discounting, a product-mix shift, or seasonality. My executive dashboard’s three-month forecast actually predicted recovery with an upward trend, which suggests the dip is seasonal rather than systemic — but I’d still recommend looking into cross-selling and bundle strategies to stabilize AOV.”

Q7. How did you handle the geographic concentration finding? “São Paulo alone is 37% of revenue and the Southeast region is over 60%. That’s both Olist’s strength and its biggest strategic risk — heavy dependence on one region. My recommendation was geographic expansion: invest in marketing and seller recruitment in underserved northern and central-western states, using the category-state heatmap from the marketplace dashboard to find specific white-space — category-state combinations with no seller presence.”

Q8. What does the executive summary dashboard add over the other four? “Synthesis. The other four dashboards each answer one question in depth. The executive summary distills all of them into one monitoring surface — KPI cards with month-over-month arrows powered by the LOD expressions, a revenue forecast, a geographic map, and parameter- driven filters. It’s designed for a weekly executive review: you scan the KPIs, and if one breaches its threshold, you drill into the detailed dashboard behind it. It’s also where the advanced Tableau — the LOD-driven cards, the parameters, the native forecasting — all comes together.”

Q9. What’s boleto and why does it matter to this analysis? “Boleto is a Brazilian bank-slip payment method — about 18% of Olist’s revenue. It matters because it serves underbanked customers who don’t have credit cards. The other 80% is credit card, reflecting Brazil’s installment-payment culture. It’s a good example of why market context matters — a US e-commerce analysis wouldn’t have a boleto segment at all, and recognizing it shapes the recommendation to expand payment accessibility to capture more of the underbanked market.”

Q10. What are the limitations of this analysis? “A few I’d flag. The boundary periods are truncated — August 2018 has incomplete data and late 2016 is too sparse for reliable trends. There’s no logistics provenance — no carrier identity, no warehouse locations — so I can’t attribute delivery problems to specific causes. Locations are only city and state level, so I can’t measure delivery distance, which is probably a confound. And the delivery-to-review relationship is correlational, not causal — it’s strong, but the data can’t isolate delivery from product quality or seller communication. All findings are also specific to one marketplace, one country, one era.”

Q11. How would you extend this project? “If I had carrier and warehouse data, I’d attribute the late deliveries to specific logistics causes rather than just observing the rate. With postal-code coordinates I’d add delivery distance as a variable, which is likely driving a lot of the delivery-time variation. And I’d love to test the delivery-to-review link more rigorously — ideally isolating delivery experience from product quality — to move from correlation toward causal attribution before committing major logistics investment.”

16. How to Walk Through This Project Live

If asked to screen-share the Tableau workbook:

  1. Open on the Executive Summary (Dashboard 5) — lead with the synthesis: KPI cards, the forecast, the map, the parameters. Then say “let me show you the four analyses behind it.”
  2. State the structure — “five dashboards, four lenses — operations, revenue, customer experience, marketplace — plus the executive synthesis.”
  3. Show the data model — the 8 tables, and explain why relationships, not joins (the granularity / double-counting point). This is a strong technical talking point.
  4. Show the calculated fields — Delivery Days, Late Delivery Flag — and the LOD expressions for the MoM cards. Spend time on LOD; it’s the most advanced technique.
  5. Walk Dashboard 1 → 3 as a story — Dashboard 1 finds the 6.8% late rate; Dashboard 3 shows the −0.65 delivery/review correlation. That’s the project’s core narrative.
  6. Show Dashboard 4 — the healthy marketplace structure and the category-state heatmap.
  7. Return to Dashboard 5 — show the forecast and the parameter interactivity, and frame it as a weekly executive monitoring surface.
  8. Close on the recommendation — logistics is the highest-leverage fix; and volunteer the correlational caveat.

Pacing tip: spend the most time on (a) the relationships-vs-joins decision, (b) the LOD expressions, and (c) the Dashboard 1 → Dashboard 3 delivery-to-reviews story. Those are the differentiated technical skill, the advanced Tableau, and the smartest analytical insight, respectively.

17. Glossary

  • Olist — a Brazilian e-commerce marketplace connecting small sellers to large platforms; the dataset’s source.
  • .twbx — a packaged Tableau workbook bundling the workbook and its data.
  • Tableau Public — Tableau’s free platform for publishing interactive dashboards.
  • Relationship (Tableau) — a flexible link between tables that lets Tableau pick the join and grain per visualization (vs. a fixed join).
  • Join — physically merging tables at one fixed granularity.
  • Granularity / grain — the level of detail one row represents.
  • Fact table — the central transactional table (Orders here).
  • Calculated field — a user-defined field built with a formula.
  • DATEDIFF — Tableau function for the number of units between two dates.
  • LOD (Level of Detail) expression — a calculation at a granularity different from the view; FIXED, INCLUDE, EXCLUDE.
  • FIXED LOD — computes a value at a specified level, independent of view filters.
  • COUNTD — count distinct.
  • Parameter — a user-facing control feeding a value into calculations.
  • Set / Top N set — a dynamic subset of dimension members.
  • Forecasting (Tableau) — built-in future-value projection with confidence intervals.
  • Trend line — a fitted line showing data direction.
  • Dashboard action — click-driven interactivity linking sheets.
  • KPI card — a single-number visual, often with a MoM indicator.
  • MoM (Month over Month) — comparison of a metric to the prior month.
  • AOV (Average Order Value) — total revenue ÷ number of orders ($159.33 here).
  • Completion rate — share of orders with status “delivered”.
  • Late delivery rate — share of orders delivered after their estimated date.
  • Review score — the 1–5 star customer rating.
  • boleto — a Brazilian bank-slip payment method serving underbanked customers.
  • Seller concentration — the revenue share held by the top sellers (low = healthy).
  • White-space analysis — finding unserved category-state combinations for expansion.
  • Live connection vs. extract — real-time source query vs. a static data snapshot.

This study guide documents the project as built. The authoritative references are the Tableau workbook tableau_olist_ops_cx_v1_raw_load.twbx, the portfolio page index.md, and the live Tableau Public dashboard. When this guide and the workbook disagree, the workbook wins.