If you have been reading about dynamic pricing, you have probably noticed that the term covers a surprisingly broad range of approaches. At one end, you have simple rules like "match the cheapest competitor." At the other, you have machine learning models that ingest years of transaction data and adjust prices in real time based on demand elasticity, weather patterns and inventory velocity.
Our complete guide to dynamic pricing covers the full spectrum. This post focuses on the practical question most mid-market retailers ask first: should we start with rules-based pricing, or jump straight to AI?
The short answer is that rules-based pricing is the right starting point for the vast majority of retailers with fewer than 2,000 SKUs. Here is the long answer.
What rules-based pricing actually looks like
A rules-based pricing engine lets you define conditions and actions in plain language. The structure typically follows an if-then pattern:
- If a competitor's price for Product X drops below your current price, then match it minus $2, provided the new price stays above your margin floor of 18%.
- If you are the cheapest across all tracked competitors for a category, then raise the price by 3% until you are no longer cheapest or you hit a ceiling of $149.
- If a product has been out of stock at two or more competitors for 48 hours, then increase the price by 5%.
These rules are transparent. You can read them, explain them to a colleague and predict exactly what will happen when conditions change. When something goes wrong, you can trace the logic step by step.
In PryceScan's pricing engine, rules are configured through a visual interface. You select the products or categories, choose a strategy type, set your bounds and activate the rule. There is no code to write and no data science team required.
| Product | Current | SmartPrice | Change | Status |
|---|---|---|---|---|
| Dyson V15 | $899 | $849 | -5.6% | Pending |
| Samsung TV 65" | $1,299 | $1,249 | -3.8% | Approved |
| Apple AirPods Pro | $379 | $399 | +5.3% | Pushed |
| Sony WH-1000XM5 | $499 | $479 | -4.0% | Pending |
Strengths of rules-based pricing
Transparency. Every price change has a clear, auditable trail. You can tell your category manager exactly why a product moved from $89 to $84 yesterday afternoon. This matters enormously in regulated industries and in any organisation where pricing decisions need sign-off.
Predictability. Because the logic is deterministic, you can simulate what would happen under different market conditions before activating a rule. If Competitor A drops by 10%, you know your prices will respond in a specific, bounded way.
Speed to value. A retailer can set up their first pricing rule in under an hour and see results the same day. There is no training period, no model calibration and no minimum data requirement.
Low cost of ownership. Rules-based systems do not require data engineers, ML infrastructure or GPU compute. The ongoing cost is the subscription to the pricing tool itself, plus the time your pricing analyst spends reviewing and refining rules.
Auditability. For retailers operating in markets with consumer protection regulations, having a clear record of why every price changed is not optional. Rules-based systems produce this by default. Every change links back to a specific rule, a specific competitor data point and a specific threshold.
Limitations of rules-based pricing
Manual rule creation. Someone needs to define, test and maintain every rule. For a catalogue of 500 products across 5 categories, this is manageable. For 10,000 SKUs across 50 categories with seasonal variations, the number of rules grows quickly.
No learning. Rules do not improve over time. If a particular pricing strategy consistently underperforms, the system will not notice. A human needs to spot the pattern, diagnose the cause and update the rule.
Limited optimisation. Rules can react to competitor prices, but they cannot optimise for complex objectives like maximising revenue across an entire category while maintaining a specific margin profile per product tier. That kind of multi-variable optimisation is where AI starts to shine.
What AI-powered pricing actually looks like
AI pricing systems use machine learning models - typically regression models, demand forecasting models or reinforcement learning agents - to predict the optimal price for a given product at a given moment.
The inputs go far beyond competitor prices. A well-trained model might consider:
- Historical sales volume at different price points
- Day of week and time of year
- Inventory levels and replenishment lead times
- Marketing campaign schedules
- Category-level demand trends
- Customer segment behaviour
- External factors like weather or economic indicators
The model outputs a recommended price (or a price range) that maximises a defined objective - typically revenue, profit or a blended metric.
Strengths of AI pricing
Optimisation at scale. For a retailer with 5,000 or more SKUs, individually crafting rules for every product is impractical. An AI model can evaluate all products simultaneously and find pricing patterns that no human would spot.
Demand sensitivity. AI models can estimate price elasticity - how much demand changes when the price moves by a given percentage. This means they can identify products where a small price increase will barely affect volume (and therefore boost profit) versus products where price sensitivity is high and competitive pricing is essential.
Continuous improvement. Unlike static rules, ML models retrain on new data. If consumer behaviour shifts - as it did dramatically during 2020 - the model adjusts its recommendations over time.
Cross-product effects. Advanced models can account for substitution effects (raising the price of Product A shifts demand to Product B) and complementary effects (lowering the price of a printer increases cartridge sales). Rules-based systems rarely handle these interdependencies.
Limitations of AI pricing
The black box problem. When an AI model recommends changing a product from $79 to $93, it is often difficult to explain exactly why. The model's reasoning is distributed across thousands of weighted parameters. This creates real problems for stakeholders who need to approve or justify pricing decisions.
Data requirements. Useful AI pricing models need substantial historical data. As a general rule, you need at least 12 to 18 months of transaction data per product, with enough volume to capture seasonal patterns. Products with fewer than 50 transactions per month rarely have enough signal for reliable demand modelling.
Infrastructure cost. Running ML models requires compute infrastructure - either cloud-based or on-premise. Training, serving and monitoring models adds operational complexity. Most mid-market retailers do not have the engineering capacity to manage this in-house.
Implementation timeline. Where a rules-based system can be live in days, an AI pricing implementation typically takes 3 to 6 months. That includes data integration, model training, validation, A/B testing and gradual rollout.
Overfitting risk. Models trained on limited data can "learn" patterns that are actually noise. A model might conclude that prices should be higher on Wednesdays because of a data coincidence, then consistently overprice midweek products.
Cost. Enterprise AI pricing platforms typically start at $50,000 to $100,000 per year. That is before implementation fees, data engineering time and ongoing model management. For a retailer doing $5 million in annual revenue, the ROI calculation is challenging.
A practical comparison
| Factor | Rules-Based | AI-Powered |
|---|---|---|
| Setup time | Days | 3-6 months |
| Minimum data needed | Current competitor prices | 12-18 months of transaction history |
| Transparency | Full - every decision is traceable | Limited - model reasoning is opaque |
| Cost (annual) | $3,000 - $15,000 | $50,000 - $200,000+ |
| Best for catalogue size | Up to 2,000 SKUs | 2,000+ SKUs |
| Maintenance | Pricing analyst | Data science team |
| Demand optimisation | No | Yes |
| Cross-product effects | No | Yes (advanced models) |
| Regulatory auditability | Built-in | Requires additional tooling |
When AI pricing makes sense
AI pricing is worth the investment when three conditions are met:
1. Large catalogue with sufficient data. If you manage more than 2,000 SKUs and have at least 18 months of transaction data with reasonable volume per product, you have the foundation for meaningful model training.
2. Complex competitive dynamics. If your products compete across multiple marketplaces with frequent price changes from dozens of competitors, the number of pricing decisions per day exceeds what rules can practically handle.
3. Revenue at scale. The cost of AI pricing needs to be justified by the revenue it influences. For most retailers, this means annual revenue above $20 million. Below that threshold, the incremental improvement over well-configured rules rarely covers the cost.
If you do not meet all three conditions today, start with rules-based pricing. You will capture 80% of the value of automated pricing at a fraction of the cost and complexity.
The hybrid approach
The smartest retailers we work with use a hybrid approach. They start with rules-based pricing across their full catalogue, then layer in AI for specific high-value categories where the data supports it.
For example, a consumer electronics retailer might use rules-based pricing for accessories and cables (commodity products where "beat the cheapest by 5%" is a perfectly good strategy) while using AI-driven demand optimisation for premium products where price elasticity varies significantly by season and promotional calendar.
This approach keeps costs manageable, maintains transparency where it matters most and focuses AI investment where the return is highest.
Getting started with rules-based pricing
If you are ready to move beyond manual pricing, a rules-based approach is the practical first step. Our step-by-step tutorial on setting up your first pricing rule walks you through the entire process, from selecting target products to testing and activating your rule.
Once your rules are running, you will want to think about governance. Not every price change should be pushed automatically. Our guide to price approval workflows explains how to set thresholds for auto-push versus manual review, so your team stays in control without creating bottlenecks.
The bottom line
Rules-based pricing is not a compromise. It is a deliberate, practical choice that gives mid-market retailers the automation they need without the complexity they do not. The transparency, auditability and speed-to-value of rules make it the right starting point for any business that has not yet automated its pricing.
AI pricing is genuinely powerful for the right use case. But if you are managing fewer than 2,000 SKUs and your annual revenue is under $20 million, the ROI simply is not there yet. Start with rules. Get your pricing operations running smoothly. Build the data foundation that will make AI worthwhile when you are ready to scale.
PryceScan's pricing engine is designed for exactly this path - rules-based automation today, with the data infrastructure to support AI-powered optimisation when your business grows into it.