How Much Did That Order Actually Make? Retail Finance Turns to AI
Customers buy online, pick up in store, return to a different location—and one sale fragments into several messy ledger entries. Databricks built Genie as finance's AI sidekick: first, figure out what that order really earned; then, keep an eye on trapped cash and decide when discounts are actually warranted.
Based on Databricks' official blog and Unilever's public case study. All figures, user counts, and industry forecasts follow the source material; the numbers in the profit waterfall are illustrative, designed to clarify the concepts rather than represent any specific store's ledger.
The Problem: The Easier You Make It to Buy, the Harder It Is to Count
Forget the product name for a second. This article is about one thing first: the path a purchase takes is so tangled now that the money quietly leaks away along the way. By the time the report is done, the floor has changed again.
Here's the most common example:
You order a jacket in the mobile app and choose "ship from a nearby store, delivered tonight." Three days later it doesn't fit, so you return it to a different store. To you, that's just "bought it, returned it." To the company's books, it splits into at least three headaches:
Tap Through It: How One Item's Booked Profit Erodes, Layer by Layer
Click through the four steps. The numbers are illustrative, showing how "planned a lot → actually kept less" happens.
The books say this jacket was supposed to contribute $40. At this point, nothing has shipped and nobody has returned it—it's the "ideal" ledger.
In reality, it ships from a store, where picking and inventory holding cost more. Only $29 is left. That gap is the "invisible cost of fulfillment."
The customer returns it, the refund goes out, handling costs kick in, and profit drops to $19. If the return lands on a different channel's books, the e-commerce report may still show "selling well."
What's actually in your pocket is $19. If finance only sees the "$40 plan" or "e-commerce is flying," they'll misread the situation. Genie's first job is to answer: after fulfillment and returns, how much did each channel and each product category actually make?
The Shift: Get an AI "Finance Buddy" Instead of Just Another Dashboard
Databricks' answer is called Genie: a data intelligence AI assistant positioned as a colleague who can take plain-language questions from finance and deliver conclusions you can act on—not just a dashboard that refreshes pretty charts.
Traditional BI Dashboard
Rigidly shows "what the data says." Metric definitions are often static. People have to assemble the business meaning themselves, and the charts you prepped for the meeting fall apart the moment the discussion veers.
Genie Finance Buddy
Understands plain language and business context (the ontology). Answers are traceable, governed by permissions, and budget-conscious about AI costs. It prepares repricing, stock-transfer, and price-hold options—but a person makes the call.
Why It Keeps Saying "Understand the Business Logic"
Finance was never afraid of massaging numbers. What it fears is this: the numbers are right, but the story is wrong. When definitions, channels, and cost drivers go stale, the answer can be calculated with perfect precision—and the decision still lands in a ditch. The main article calls this the ontology: capturing "what this number actually means inside your company" and keeping it current as the business changes. Databricks CEO Ali Ghodsi is quoted, in effect: most enterprise AI is confidently wrong guessing; the root cause is a context problem, not an intelligence problem.
The main article is also explicit about security: every number traces to a source, permissions are enforced, and the AI's own costs are governed—only then is an answer safe enough to act on. Genie prepares reprice / transfer stock / hold a price,but the human stays in charge.
The Three Questions Genie Helps Finance Track
These aren't three unrelated KPIs. They're a profit loop where each step feeds the next: first see the true profit clearly, then free up trapped cash, then hold the full price while demand is still there.
Q1: After fulfillment and returns, what did we actually earn?
By channel and by product, after shipping, store distribution, and return markdowns—where does real profit land? If you can't see this clearly, the later moves on transfer and price-hold could hit the wrong products.
- This is the endpoint of the profit waterfall above
- Planned markup ≠ what we actually keep
- Promotion engines can force unplanned discounts
Q2: Which inventory is trapping cash?
Excess, aging, and misplaced stock. The report says "in stock," but the money is stuck. You need to free that cash before the goods age and force markdowns—through transfers, stopping replenishment, or clearance, with supply chain and merchandising making the call.
- The true-profit heat map will flag where contribution is weak but inventory keeps piling up
- Move cash early, and you take fewer forced markdowns
Q3: Where does full-price selling start to slide?
Don't wait until you can't move the goods to discount. When demand is still strong, use AI to spot the slide early and avoid discounting whenever possible. Genie can prepare hold / reprice options, but the decision belongs to merchandising and operations.
- By the time the markdown order lands, the erosion has often already started
- As agents accelerate pricing decisions, finance needs to see "forming risks"
Loop: Three separate fights, one profit engine
See true profit → find trapped cash → hold the full price that's still supportable → feed the next round of profit. The main article puts it this way: it turns three separate fights into one reinforcing mechanism. Genie learns continuously across all three questions; the human stays in the loop.
Digging into Q2: Three Ways Cash Looks "In Stock"
Excess: The inventory looks fine on the books, but turns slowly. Cash is buried in units that won't sell, leaving true demand short at the SKUs that actually need it.
Aging: Time is eating the full price. If you don't move now, you'll slide into forced markdowns—Q3 fails before you get there.
Misplaced: The goods are sitting in the wrong warehouse or store, while another location is missing out on sales. The move is a transfer, not a reorder. Genie explicitly prepares transfer stock.
The Result: Unilever Compresses "Days" into "Minutes"
The main article keeps the results section thin; the customer page has the verifiable scale and impact. This is the proof that the "finance buddy" isn't just a slogan.
Unilever's finance team manages a massive volume of data spanning SAP, customer transactions, and reporting. They spent years building a centralized finance data lake, connecting 30+ SAP sources and legacy systems. The foundation was there, but getting answers out still meant manual prep and spreadsheet stitching. You'd walk into a meeting with a finished analysis, the discussion would veer, and it was back to another round of extracting data.
“With Genie, we've changed how finance works. What used to take days now takes minutes, and we can explore questions much more freely.”
Cyro Souza · Unilever · Finance asks in plain language, gets analysis in minutes“Validating a single hypothesis could require significant analytical effort… If the initial assumption proved incorrect, teams often needed to restart… With Genie, you can test multiple hypotheses in minutes.”
Hypothesis validation: from nearly starting over on a wrong assumption to testing several market hypotheses in the same session“Configuration is the heart of Genie. You need clear definitions and structure, but once that is in place, people can just ask questions and move forward.”
The prerequisite: governed data + a Genie space configured for finance (metrics, synonyms, business context). You don't just drop in a large language model and expect it to talk.On value realization, the case study maps faster, freer customer-credit-risk and market-hypothesis analysis to an expected "multi-million-euro" annual cost avoidance. Credit risk also moves from periodic reviews to something much closer to continuous monitoring. The data lake solved "do we have one version of the numbers?" Genie solved "can we keep asking follow-up questions in the middle of a live discussion?"
Boundaries: Genie Keeps Watch, It Doesn't Decide Prices
- It does not directly make pricing, promotion, or markdown decisions. Those belong to merchandising, marketing, supply chain, and operations. Genie provides an accurate, governed view that surfaces risk early and points the responsible team at the action.
- Finance's mission hasn't changed—speed and complexity have. More margin-related decisions are being made by agents; the tool has to keep up and carry the governance.
- Ontology + governance = answers you can stake a decision on. The meaning is right, and the source, permissions, and AI costs are all controlled.
- Configuration is the heart. Unilever: without solid definitions and structure, conversational analysis falls apart.
How Much Did That Order Make? AI Helps Retail Finance Close the Books
Retail finance teams use Databricks' Genie to track true profit, trapped cash, and full-price sales—explained on one page with visuals.
↓ One page, with one animated diagram
A jacket is ordered online, shipped from the nearest store, and returned three days later to a different location. That's three hurdles for the books: store fulfillment costs more than a central warehouse, the return may land on another channel's ledger, and the price may have been adjusted by a system. Money leaks quietly at each step, and you only notice in the monthly report.
Databricks calls this tracking tool Genie—an AI assistant you ask in plain language, built to act like a finance colleague. The same gross-margin number can be perfectly accurate but still not correct until it's wired back to channel, demand, and cost to serve. That's the gap Genie closes.
Lays out "what the data says." Metric definitions go stale, people assemble the business meaning themselves, and a single follow-up question breaks the prepped charts.
Understands plain language and which number attaches to which channel and cost. Answers trace to their source, stay inside permission controls, and pricing or stock moves are only prepared as options—a person decides.
Genie helps finance track a self-reinforcing profit loop: first see what this order actually earned, then free up the cash trapped in inventory, then hold the full price while demand is still there.
The Databricks blog keeps the results brief; most numbers come from Unilever's public case study. Unilever's finance team had already connected 30+ SAP sources into a data lake, but pulling answers still meant manual spreadsheet work, and any follow-up question meant starting over. With Genie, the live discussion can just keep going.
Genie keeps watch and prepares repricing or transfer options—but a human makes the call.
✔ Flag inventory that's trapping cash
✔ Spot full-price erosion early
Returned to another store closer to work.
- × Store shipping costs more than the warehouse
- × The return landed on another store's books
- × A system quietly repriced it once
≈€3M expected annual savings
Members can copy or download the full Markdown, and let their AI subscribe and ask.
Found this useful? Leave your email — every new explainer, the moment it's out. Unsubscribe anytime.