Dynamic Pricing
“What price maximizes margin for this product?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
What price maximizes margin for this product?
Retailers reprice millions of SKUs manually or with basic rule engines (competitor price minus 5%), leaving $8-15B in annual margin on the table industry-wide (McKinsey). Price elasticity varies dramatically by product, customer segment, time of day, inventory level, and competitive context. A $0.50 price increase on a low-elasticity item yields pure margin, while the same increase on a high-elasticity item drives customers to competitors. Most pricing tools treat each SKU independently, missing the cross-product effects: raising the price of brand-name cereal by $0.30 shifts 12% of demand to the store brand.
Quick answer
Effective dynamic pricing requires understanding price elasticity in context: not just how price changes affect one SKU, but how they ripple across substitutes, complements, and customer segments. Raising brand-name cereal by $0.30 shifts 12% of demand to the store brand. A relational model captures these cross-product substitution effects because it connects products, competitor prices, customer segments, and inventory levels into a single graph. On SAP SALT benchmarks, relational approaches hit 91% accuracy vs 75% for XGBoost on demand prediction tasks that underpin pricing decisions.
Approaches compared
4 ways to solve this problem
1. Rule-based competitive matching
Set prices as a fixed offset from competitor prices (e.g., competitor minus 5%). The most common approach in grocery and mass retail.
Best for
Commodity categories where price is the primary purchase driver and competitor prices are easily scraped.
Watch out for
Ignores elasticity differences across products and customer segments. A $0.50 cut on inelastic items is pure margin destruction. No awareness of cross-product effects.
2. Price elasticity models (regression-based)
Estimate demand curves from historical price-quantity data using log-linear regression or similar. Calculate optimal price per SKU.
Best for
Retailers with clean price-variation history (A/B tests or natural price changes) across a moderate number of SKUs.
Watch out for
Each SKU is modeled independently, so cross-product substitution is invisible. Raising the price of Coca-Cola shifts demand to Pepsi, but the model does not account for the offset revenue.
3. XGBoost demand models with price features
Train gradient-boosted models to predict demand as a function of price, promotion status, and competitor price. Optimize price using predicted demand curves.
Best for
Analytics teams with feature engineering capacity who can build and maintain price-demand models across categories.
Watch out for
Feature engineering for cross-product effects is extremely manual. You need to hand-code every substitution pair. SAP SALT shows 75% accuracy on demand prediction, leaving significant margin on the table.
4. KumoRFM (relational foundation model)
Connects products, price history, competitor data, customer segments, inventory, and promotions into a relational graph. Learns cross-product elasticity and substitution patterns automatically.
Best for
Retailers managing 10,000+ SKUs where cross-product substitution and customer segment-level elasticity drive the margin opportunity.
Watch out for
Requires price variation in historical data. If you have always priced at a single point, there is no demand curve to learn from.
Key metric: SAP SALT demand prediction: relational 91% vs XGBoost 75%. Cross-product substitution effects account for 15-30% of category margin optimization.
Why relational data changes the answer
Single-SKU pricing models optimize each product in a vacuum. They tell you the optimal price for tortilla chips is $4.29 based on its own demand curve. But they cannot see that raising chips from $3.99 to $4.29 shifts 8% of demand to the store-brand chips at $2.99, or that the customer segment visiting at 6pm is 40% less price-sensitive than the morning crowd. These cross-product and cross-segment interactions live in the joins between products, transactions, customer_segments, and competitor_prices tables.
A relational model captures the full substitution matrix automatically. It learns that the $0.30 price increase on brand-name chips nets $142 in margin from the remaining 440 units, but also shifts 35 units to the store brand, generating an additional $31.50 in store-brand margin. The net margin optimization across the category is $173.50, not $142. Multiply this across 40,000 SKUs and you see why relational pricing models add 3-8% gross margin where single-SKU models add only 1-3%.
Pricing each SKU independently is like a chess player who only thinks about the piece they are moving. They push a pawn forward without noticing it exposes their bishop. A relational pricing model sees the whole board: raising the price of brand-name chips is the pawn move, the exposed bishop is the store-brand substitute that absorbs demand, and the winning strategy accounts for both moves together.
How KumoRFM solves this
Relational intelligence built for retail and e-commerce data
Kumo connects products, competitor prices, transaction history, customer segments, inventory levels, and promotional calendars into a relational graph. The model learns the demand curve for each product in context: SKU-4310 at $3.99 in Store S-14 will sell 480 units this week, but at $4.29 it will sell 440 units with a net margin gain of $14,400. The graph captures cross-product substitution effects, so the model also predicts that the $0.30 increase will shift 35 units to the competing SKU-4311, netting $11,200 in combined margin gain.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
PRODUCTS
| sku_id | name | category | current_price | unit_cost | competitor_price |
|---|---|---|---|---|---|
| SKU-4310 | Casa Crunch Chips 12oz | Snacks | $3.99 | $1.40 | $4.19 |
| SKU-4311 | Store Brand Chips 12oz | Snacks | $2.99 | $0.90 | N/A |
| SKU-4520 | Premium Coffee Beans 1lb | Beverages | $12.99 | $6.50 | $13.49 |
PRICE_HISTORY
| sku_id | store_id | price | units_sold | date |
|---|---|---|---|---|
| SKU-4310 | S-14 | $3.99 | 68 | 2025-09-22 |
| SKU-4310 | S-14 | $3.49 | 95 | 2025-09-15 |
| SKU-4310 | S-14 | $4.29 | 58 | 2025-09-08 |
INVENTORY
| sku_id | store_id | on_hand | days_of_supply | next_delivery |
|---|---|---|---|---|
| SKU-4310 | S-14 | 520 | 7.6 | 2025-09-29 |
| SKU-4311 | S-14 | 340 | 12.1 | 2025-10-02 |
| SKU-4520 | S-14 | 85 | 4.2 | 2025-09-27 |
CUSTOMER_SEGMENTS
| segment | price_sensitivity | avg_basket | share_of_traffic |
|---|---|---|---|
| Value Seekers | High | $38 | 35% |
| Convenience | Low | $62 | 25% |
| Premium | Very Low | $95 | 15% |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT SUM(PRICE_HISTORY.UNITS_SOLD, 0, 7, days) FOR EACH PRODUCTS.SKU_ID, STORES.STORE_ID ASSUMING PRODUCTS.CURRENT_PRICE = 4.29
Prediction output
Every entity gets a score, updated continuously
| SKU_ID | STORE_ID | OPTIMAL_PRICE | PRED_UNITS | PRED_MARGIN | VS_CURRENT |
|---|---|---|---|---|---|
| SKU-4310 | S-14 | $4.29 | 440 | $1,271.60 | +$142.00 |
| SKU-4311 | S-14 | $2.99 | 375 | $408.75 | +$31.50 |
| SKU-4520 | S-14 | $13.49 | 78 | $545.22 | +$38.50 |
Understand why
Every prediction includes feature attributions — no black boxes
SKU-4310 (Casa Crunch Chips) at Store S-14
Predicted: Optimal price $4.29, 440 units, +$142 margin vs current
Top contributing features
Price elasticity for this SKU
-0.35 (low)
28% attribution
Competitor price gap
$0.10 below comp.
24% attribution
Inventory level (sufficient stock)
7.6 days supply
19% attribution
Cross-product substitution rate
8% to store brand
17% attribution
Customer segment mix (low sensitivity)
40% Conv+Prem
12% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about dynamic pricing
How does AI dynamic pricing handle competitor price changes?
The simplest approach is rule-based matching (competitor price minus X%). Relational models go further by learning how your customers actually respond to competitor price gaps. If your competitor drops coffee beans by $1 but your customers are low price-sensitivity (convenience shoppers), the model may recommend holding price and absorbing the minor volume loss. The decision depends on the full context: customer segment, inventory level, and margin contribution.
Does dynamic pricing require real-time data?
Not necessarily. Grocery and mass retail typically reprice weekly or during promotional cycles. E-commerce can reprice hourly or daily. The frequency depends on how fast your competitive landscape shifts. What matters more than real-time data is connected data: linking price changes to demand outcomes across products, stores, and customer segments so the model learns the right elasticity curves.
How do you avoid a race to the bottom with AI pricing?
By optimizing margin, not just volume. Rule-based competitive matching (always beat the competitor by 5%) leads to margin erosion. Relational pricing models optimize for total category margin, which sometimes means holding a higher price on inelastic items even when competitors drop theirs. The model surfaces which products can absorb a price premium and which are truly price-competitive, preventing blanket discounting.
Bottom line: Increase gross margin by 3-8% across the assortment without sacrificing volume, adding $15-40M in annual profit for a $5B revenue retailer.
Related use cases
Explore more retail & e-commerce use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
