How Prorize Applies AI to Revenue Management: The Roles of Predictive, Prescriptive and Generative AI in Supporting Smarter Pricing Decisions
Ask a large language model what to charge for a 10×10 climate-controlled unit, and you can get an answer in seconds. It may be confident, articulate, and entirely plausible.
That is exactly the problem.
Plausible is not optimal. A price that sounds right can still leave revenue on the table, slow move-ins, drive avoidable move-outs, or violate the rules your business runs on. A polished explanation can make a weak decision look credible until its consequences appear in the rent roll.
Nearly every pricing system now claims “AI”. The label alone tells an operator almost nothing about how a price is produced. What matters is which kind of AI is used, where it is applied, and what it is trusted to decide.
At Prorize, we use AI to strengthen the science of revenue management. Any system entrusted with pricing must connect customer behavior to a measurable revenue objective, evaluate the economic tradeoffs, and enforce the operator’s business rules every time.
This paper explains how Prorize applies AI and the boundaries that make it useful and trustworthy.
Key Takeaways
- Pricing needs an economic foundation. Demand forecasting, price-sensitivity estimation, and optimization remain essential to maximizing expected revenue. No language model removes that need.
- An LLM’s output alone is not a reliable pricing engine. Calculation errors, inconsistent rule enforcement, variable outputs, and uncalibrated confidence can create unjustified price changes and costly mistakes. Pricing requires validated models and enforceable controls.
- Different AI capabilities serve different purposes. Predictive AI estimates outcomes, prescriptive AI selects actions, and generative AI helps people interpret information and improve the process.
- Automation needs accountability. Reliable data, enforceable constraints, human oversight, and performance measurement determine whether automation deserves trust.
The Price is a Decision about Customer Behavior
Generative AI is remarkably good at language. But a rent is not a sentence. It is a decision about how customers will respond, with real money attached.
The central question in revenue management is how a different price will change customer behavior and, ultimately, revenue. A higher rate may earn more per rental while reducing move-ins. A promotion may fill units faster while lowering effective rent. An existing customer rate increase may produce additional revenue while increasing move-out risk.
In self-storage, evaluating those tradeoffs means forecasting move-ins, move-outs, price sensitivity, length of stay, reservation cancellations, lead conversions, and how customers choose among unit types. Promotions, competitive pricing, unit availability, and operating constraints also shape the decision.
Pricing is not a single-dimensional decision problem tied to a single measure like occupancy or competitive price. Occupancy or a competitor’s rate can be useful information. Neither, by itself, determines the price that maximizes expected revenue. The same occupancy level can call for very different actions when demand, price sensitivity, and expected move-outs differ.
A recommendation must therefore connect a forecast of customer behavior to an explicit optimization objective. That connection is where revenue management creates much of its value.
The Engineering Behind Every Recommendation
At Prorize, we view pricing as a decision science problem with an explicit economic objective and operating constraints. Solving it takes predictive modeling and optimization, supported by a software platform that connects data, pricing science, business rules, and day-to-day workflows.
Every layer involves deliberate design choices: How raw data is processed and validated, which variables are derived at what level, and whether behavior is modeled at the unit-type, property, or market level, and how those variables combine into an accurate forecast.
Uncertainty also matters. Forecasts may need to represent a range of outcomes, not a single number, and optimization should evaluate pricing choices across plausible demand scenarios while respecting business constraints. Workflows must turn recommendations into timely action. A sophisticated interface cannot compensate for a weak demand model, an inappropriate objective, or inconsistent constraints.
The objective is not to guess, copy competition, or generate a number that appears reasonable. It is to determine the price that best supports the operator’s revenue-management objective given the available information and business constraints. And because market conditions shift, that answer keeps moving: a pricing rule that works today may fail tomorrow.
How Prorize Applies AI
Prorize’s Self-Storage Revenue Optimizer, SSRO™, combines three complementary layers: predictive AI, prescriptive AI, and generative AI. Each serves a distinct role.
Layer 1: Predictive AI estimates what is likely to happen
SSRO™ uses statistical and machine-learning models to forecast demand and move-outs, estimate customer responses to price change and customer lifetime value. These models utilize many data points and identify meaningful patterns across stores, unit types, and markets.
The purpose is to estimate likely outcomes before choosing a price. Because market conditions change, forecasts must be continuously evaluated and recalibrated against observed behavior.
Layer 2: Prescriptive AI determines the optimal action
Forecasting alone does not tell an operator what to charge. SSRO™ uses optimization to translate forecasts into rent recommendations that support the revenue objective within operating constraints and operator-defined rules.
A rising demand does not automatically justify a higher rent. The right response could be a small increase, a larger increase, no change, or a decrease. The decision depends on many factors including available inventory, price sensitivity, forecasted net move-ins, and the expected revenue impact of each alternative.
Existing Customer Rate Increases (ECRI) also illustrate the distinction. The largest increase is not necessarily the most profitable. SSRO™ evaluates customer type and tenure and expected move-out sensitivity to an increase to determine the adjustment that best supports the operator’s revenue objective.
Layer 3: Generative AI explains, assists, simplifies, and improves the pricing processes
This is where large language models earn their place: making revenue management easier to understand and use, and making the underlying pricing process faster and more effective.
For users, this includes the SSRO™ Assistant, AI-generated alert narratives, and tools that help explain recommendations and reports. Questions like “Why did SSRO recommend this price?” or “What is this report telling me?” are communication and interpretation problems, which is exactly what LLMs do well.
Behind the scenes, Generative AI strengthens the pricing process itself: spotting unusual patterns, surfacing potential data or recommendation issues, accelerating analysis, supporting model development and testing, and automating routine work. That gives our pricing scientists and revenue managers more time to improve models and evaluate outcomes.
Our philosophy is simple: use predictive and prescriptive AI to make the pricing decision, and use Generative AI to explain, support, accelerate, and continuously improve the pricing process.
Why We Don’t Let an LLM Choose the Price
A general-purpose large language model can discuss pricing and suggest a number. It can also be connected to databases, calculators, forecasting models, and optimization tools. In that architecture, it may coordinate useful analysis, but the reliability of the price still depends on the underlying models, tools, and controls.
This is why Prorize does not use an LLM’s free-form response as its pricing decision engine. Without a validated decision process, generated prices can vary with prompt wording or irrelevant context rather than any real change in demand.
Before letting an LLM set rents, operators should understand what is missing:
- Demand models behind the price need calibration. The system must estimate how price affects customer behavior for the relevant store, unit type, or market. Naming the right factors in a prompt does not establish those relationships, and general-purpose training across unrelated industries is no substitute for models calibrated to self-storage customer behavior and economics.
- Models are optimized for an explicit economic objective. The system must evaluate alternatives against a defined goal, such as maximizing expected revenue, within measurable constraints. A generated number alone does not demonstrate that price optimization occurred.
- Numerical answers can be wrong. LLMs can perform many quantitative tasks, especially with tools, but generated answers can still contain calculation or reasoning errors. Numerical plausibility does not establish that a price is feasible or optimal. In pricing, errors translate quickly into lost revenue or occupancy.
- Business rules are not guaranteed. Rate bounds, parity rules, and unit-type relationships need to be enforced by software and checked before action. A free-form LLM response may interpret or apply those rules differently from one interaction to another.
- An explanation is not an audit trail. An LLM-generated rationale may not faithfully represent how an answer was produced [1]. A narrative explanation does not establish which inputs or alternatives actually drove the recommendation. Pricing decisions need defined inputs, model versions, objectives, constraints, optimization outputs, and overrides that can be independently reviewed. Reproducibility must be designed into the surrounding system.
- Confidence and context need validation. A model’s confident tone is not a calibrated measure of uncertainty. It may overemphasize irrelevant details or apply context inconsistently. Forecast accuracy and recommendation quality must be assessed against observed outcomes.
- Context can mislead. LLMs can use large amounts of context effectively. Sometimes, they can connect unrelated information, overemphasize irrelevant details, or apply prior context inconsistently. They do not provide a reliable measure of confidence for each recommendation. This makes human judgment and appropriate controls important when LLMs are used in commercial workflows.
- Security and control matter. Generative AI can introduce risks such as prompt injection and sensitive-information disclosure [2]. Systems need control over what a model can access, change, or execute, especially when actions affect customer rents. Those risks become more consequential when a model is allowed to directly make or execute commercial decisions.
Domain expertise matters in pricing. Prorize’s models are purpose-built for self-storage and are calibrated around its customer behaviors, operating constraints, and economics of this industry. By contrast, general-purpose reasoning does not provide that specialization on its own and is exposed to information across many unrelated sectors and is not inherently specialized in the dynamics of self-storage revenue management.
The lesson is not to avoid LLMs. It is to use them where their strengths create value and surround them with the right controls. Prorize uses generative AI to help users understand information, summarize results, improve workflows, and interact more efficiently with the pricing process. But the price itself is determined by purpose-built statistical, forecasting, machine-learning, and optimization models operating within defined business rules.
An LLM can help explain and improve the pricing process. It should not replace the quantitative decision engine behind the price.
The Guardrails around AI
Start with reliable data
Prorize checks on pricing inputs for format, validity, and integrity. Business rules help identify missing rents, unusual changes, outliers, and inconsistencies in effective rents after discounts. We apply data minimization, using the information needed for the revenue-management task rather than every available customer attribute.
For new customer pricing, SSRO™ operates primarily at the property and unit level and does not require personally identifying the renter. Client data remains within client-specific model boundaries rather than being pooled to train models across clients or industries.
Automate within approved limits
Operators define thresholds within which SSRO™ recommendations can be accepted automatically; anything outside those boundaries goes to review. Users can accept, modify, or override any recommendation when local knowledge or business judgment warrants.
This supports workflows ranging from individual review to automation within approved limits. Revenue managers contribute context, resolve exceptions, and retain accountability for how the system is used. This creates a spectrum from recommendation to human approval, to bounded automation rather than an all-or-nothing choice between manual pricing and autonomous AI.
Evaluate results and improve the models
Prorize evaluates model performance using metrics such as move-in and move-out forecast accuracy, and monitors business outcomes including occupancy, revenue performance, and competitive price positioning. Comparing predictions and recommendations with actual results helps identify where recalibration or model improvement is needed. Those outcomes must be interpreted alongside market conditions and operator actions.
The Next Stage of AI in Revenue Management
Prorize’s roadmap includes enhancements in promotion optimization, marketing spend support, budgeting and forecasting, tightly controlled LLM-assisted rent adjustments, and a revenue manager AI agent that automates routine workflows so operators can focus on exceptions and higher-level pricing strategies.
Every expansion of automation must preserve the discipline of revenue management. LLM-assisted adjustments should run through validated quantitative models, enforceable business rules, and appropriate approval controls.
Operators should expect more from AI than a convincing answer. They should be able to examine the behavioral assumptions behind a recommendation, the objective it serves, the constraints it respects, and the evidence used to evaluate its performance.
The design principle stays the same: add intelligent automation around the revenue-management engine without compromising its rigor.
At Prorize, AI strengthens the science of revenue management. Its value must be demonstrated in pricing decisions and outcomes that can be measured and defended.
References
[1] Anthropic. Reasoning models don’t always say what they think. 2025.
