How advanced data governance, precision, and accountability are shaping the next generation of AI-driven pricing intelligence.
Suzanne, your role at the helm of PricingAI at Pricefx positions you uniquely at the intersection of AI and commerce—how has your career path evolved to lead you here, and what perspectives has it shaped around pricing innovation?
Growing up in Los Alamos, New Mexico, I was surrounded by science and engineering role models from an early age. I studied Applied Math and Biostatistics and started my career at Procter & Gamble designing pharmacokinetic studies for clinical trials. About 3 years in, I had the opportunity to help build a Trade Promotion Optimization system, which would help brand managers optimize their marketing budget. Underlying this system were models of consumer demand that considered a variety of factors — price and promotional elasticity, seasonality, trends, and product “cannibalization” among substitutable products. This led me to join DemandTec, a cloud-based B2C price optimization provider, serving global retailers like Walmart, Target, Best Buy, and Sainsbury before our ultimate acquisition by IBM.
During this time, what was possible with data and analytics and AI was rapidly evolving, to where we could digest and analyze very granular data in real time. In 2018, I broadened my horizons by joining Facebook (now Meta), and worked in Ads Product for more than 5 years, developing features and support for small businesses all over the world who relied on Meta’s advertising surfaces (Facebook, Instagram, WhatsApp). Then after a brief hiatus, I rejoined the pricing world with Pricefx, first as a Data Science Consultant and then leading PricingAI.
In terms of how my journey has shaped my perspective, I’ve seen the evolution from static to dynamic data and models, and am thrilled that today’s pricing teams can take advantage of a wide range of AI-based models and algorithms to provide insight into their business and customers.
Having worked with both global enterprises like Walmart and small businesses on Meta platforms, I recognize that pricing innovation must adapt to different scales of operation while maintaining its effectiveness. Finally, I’ve seen again and again the importance of being able to translate complex algorithms into tangible business outcomes. Pricing innovation isn’t just about technical sophistication; it’s about creating meaningful business value that users actually adopt and trust.
As businesses lean more heavily on AI to inform pricing decisions, how do you define the threshold for “trustworthy data” in this context?
To be trustworthy, data must earn the confidence of both technical and business teams, and there are several principles that apply regardless of industry.
First, businesses should obviously strive for the most complete and consistent data that they can curate, with as much contextual information as possible on what influences outcomes (e.g. pricing decisions and demand). Beyond price and quantity, this includes information such as promotions/deals, product and customer attributes, seasonal influences, and eventually external data such as competitor information. This core data needs to be harmonized –since it likely comes from multiple systems — and continually validated to detect inconsistencies and outages.
Businesses need to ensure their data is representative of current market reality. Data going back many years can be useful for understanding certain purchasing patterns. For example, in B2C one sees varying shopping patterns in December depending on when Christmas falls during the week. But since many markets are shifting rapidly, it’s generally better to overweight more recent data.
Special treatment for very sparse data is important. Different algorithms will be relevant for high velocity products vs. low velocity, and frequently a hierarchy of algorithms can be built, automatically assessing data sparsity and “falling back” to more appropriate structures as needed.
And one of the exciting elements in leveraging AI is that feedback can be incorporated into the algorithm. For example, AI can be used to group product entities, and a pricing practitioner can review results, identify flaws in the logic either due to missing data or lack of AI context, and have their domain expertise codified in revised logic.
All of this said, I want to emphasize that businesses shouldn’t get overwhelmed by the potential enormity of trustworthy data. Throughout my career I’ve viewed data curation for AI as an evolving journey. My advice is to get started by with what you have and create a data roadmap to both curate additional data sources and refine existing ones.
What are the most common blind spots you’ve observed in enterprise pricing data, and how do they typically impact AI-driven outcomes?
As mentioned, surfacing business data is the first step in beginning to understand potential blind spots. Some common examples:
- Lack of discount and cost understanding: Many enterprises fail to integrate downstream discounts or payments such as rebates, chargebacks, and marketing allowances in their pricing data. This creates an incomplete picture of true net pricing and can bias elasticity estimates. Similarly, margin optimizations against incomplete cost data can lead to suboptimal and misleading results.
- Price implementation fidelity gaps: There can often be a difference between recommended prices and what actually gets implemented. Analysis must be performed to understand user compliance with recommendations, and there are likely learnings in the cases where recommendations were rejected. And the actual prices need to be fed back into AI algorithms, so that they are processing reality.
- Overreliance on price as a driver: While the Price-Volume relationship is typically dominant in B2C, it is often murkier in B2B. Other factors such as relationship tenure and quality, product & service reliability, breadth of products purchased, and geographic proximity may impact purchasing decisions. AI can be leveraged to help discern which factors have historically impacted pricing and shape future pricing decisions.
- Not contemplating external data: Enterprises sometimes become so focused on collecting robust internal data that they neglect important external context. It’s important to identify any temporal influences, anomalies, and shocks that might influence normal pricing decisions and relationships so that AI models understand unusual circumstances. Competitive data is also important to curate, although availability will vary by industry. But even having some competitive context enables setting guardrails that improve AI-based optimization recommendations.
Data quality seems to be both a technical and cultural challenge—how should organizations think about ownership and accountability in maintaining data integrity?
The most successful data initiatives that I’ve seen started with clear “data domain” ownership, with business leaders as “data stewards” articulating their needs for their pricing domain and technical leaders owning the systems and processes that provide high quality data. Having stated shared goals and KPIs helps keep both sides accountable. Something that Meta does incredibly well within cross-functional product teams is setting long term goals, breaking them down into near-term milestones, defining metrics for measuring progress, and setting targets that the whole team can rally around achieving.
Some ideas for bringing this broad framework to life:
- Make quality issues visible. Implement data quality scorecards or dashboards at the start of the project, so that you have a baseline to work from, and normalize regular sharing of the results with leadership.
- Understand and address root causes. Most data issues aren’t about carelessness; they are the result of inadequate tools or incomplete business processes.
- Clarify and streamline the issue resolution process. Define who has decision-making authority for questionable data, and create escalation procedures that balance speed with accuracy.
- Celebrate quality champions. Recognize and reward individuals who consistently contribute and maintain high data quality standards, and share success stories of how great data quality is enabling better pricing decisions.
Internal data is only one part of the puzzle. How should pricing teams approach the integration of external signals like market trends, weather, or competitor pricing into their models?
Once businesses have a plan for getting internal data ready for Pricing AI, there are many interesting external data sources to explore and integrate. That said, before racing to full integration of an external signal, it’s best to start by establishing clear hypotheses about how external factors may be relevant and actionable for your pricing decisions. For example, weather data has intrigued pricing teams for years, but it’s only relevant when one understands how temperature and/or precipitation will trigger (or deter) certain purchases.
As with all model features, it’s good to validate that a new external signal will actually improve model performance before fully operationalizing the feed. I’ve seen examples where new data sources add complexity without improving accuracy. In addition, it is important to consider external signal “stability” – is the source something that will remain available and consistent over time? If not, you’ll need a plan for gracefully deprecating your model’s dependence on it.
With these principles in mind, I recommend a phased approach that tests and incorporates the most relevant data for a business. For example:
Phase 1: Market Context. Competitive pricing intelligence, economic indicators, market growth data, and customer intelligence (for B2B this could be enhanced firmographics)
Phase 2: Advanced Cost Data. Regulatory and tariff information, commodity & input cost indices, and supply chain indicators
Phase 3: Additional Demand Drivers. Weather patterns for weather-sensitive products, localized seasonal and event calendars, social sentiment data
The sequencing and sources of external data will of course vary by industry.
What strategies or tools do you recommend for surfacing data anomalies before they skew AI recommendations?
In my experience, a combination of automated standard AI algorithms and more tailored business rules with ongoing contextual enrichment works best.
A variety of AI algorithms can be used to detect outliers, understand seasonal patterns, and understand multivariate anomalies (e.g. unusual price, volume, and cost patterns). With a combination of business context and exploratory analysis, acceptable thresholds and ratios for a particular business segment and metric can be determined and implemented to enhance and customize outlier detection. When outliers are identified, feedback from experts can be invaluable to tag transactions or time periods with disruptions to the norm.
Operationalizing this process will vary by organization, but ideally, pre-ingestion validation pipelines can quarantine suspicious data for human review before it enters model training. A set of reports and dashboards help analysts both monitor what’s found by algorithms and identify additional patterns that should be added to the repertoire of algorithms.
Many organizations still run on legacy systems that fragment their data—what practical steps can they take to unify and modernize their data pipeline for AI readiness?
Having siloed legacy systems is the norm in most enterprises. The most successful data transformations that I’ve seen prioritize incremental value delivery over “big bang” everything-at-once replacements. In other words, start by identifying some specific high value use cases where you can demonstrate ROI.
In terms of getting started. I’m a fan of starting with a “data inventory” – map out all pricing-relevant data sources and where they reside (ERP, CRM, contract management, external data subscriptions), and document key data entities and relationships across systems. During this inventory, assess data quality and completeness.
Then there are multiple options. One route is to begin formally integrating datasets by implementing APIs that harmonize various sources into one platform; other organizations may start by implementing a “virtual data platform” by creating views that join pricing-related data across systems, to identify inconsistencies and conflicts across systems that require resolution.
Early in the process, it’s important to start building a basic data governance framework. This should cover process ownership, pricing metric definitions (e.g. gross price, net price, pocket margin), data quality norms, and understanding of who will implement and own anomaly detection.
As the AI platform matures, businesses frequently implement feature engineering pipelines — automated transformation processes that convert raw pricing data into AI-usable features. Modern pricing platforms typically provide business-friendly interfaces to enable feature engineering, as well as self-serve reporting and visualization tools so that both IT and business users can explore the data.
Real-time data is often cited as a game changer. What kinds of pricing scenarios benefit most from real-time inputs, and what infrastructure is needed to support that?
While real-time data isn’t mandatory for all pricing scenarios, it is indeed a game changer for certain pricing problems.
Some of the highest value scenarios include:
- Dynamic marketplace environments, such as e-commerce pricing and auction & bid-based environments, where real-time competitor information, cost, and demand signals dramatically improve win rates while protecting margins.
- Supply & Demand “balancing” situations, such as those involving perishable or highly seasonal inventory (e.g. airlines & hotels, last minute services, seasonal attire), and situations where availability may suddenly change due to supply chain disruptions.
- Personalized offer environments, where a customer is in a physical or digital place and exhibits signals that they have a potential need to fulfil / intent to purchase and might be enticed by a discount, but the window for making that offer is short.
- Promotional effectiveness pricing, when time is of the essence, for example, competitive promotion responses, coordinated promotions across physical and digital channels, and performance-based campaign adjustments where real-time response can lead to mid-promotion adjustments.
In terms of the infrastructure to support real-time, this is a very deep topic, but at a high level, multiple components should be considered as part of the design:
- For Data Ingestion and Processing, an event streaming backbone is key to handling high-volume, low-latency data flows. Enabling change data capture (identifying and propagating relevant changes without full data replication), deploying processing capacity close to data sources, and having a strong API management layer are important.
- For the “Intelligence” layer, technology should enable real-time AI infrastructure, rules engines, and in-memory computing
Additionally, real-time systems require investment in governance and control, ensuring both automated guardrails and interfaces that enable Pricing practitioners to monitor and override decisions as appropriate.
As AI tools become more accessible, how can companies ensure that democratization doesn’t come at the cost of precision and data discipline?
Democratization of AI tools and analytical rigor/discipline don’t have to be opposing goals. They can be complementary when designed thoughtfully.
Some guiding principles:
- Build AI pricing platforms with multiple layers. It is entirely feasible to provide AI modules with built-in workflows that guide a non-data scientist through the analytical steps and give them access to basic parameters on the surface. For more sophisticated users, access to additional options can be provided, including the ability to view and edit the source code, or even “bring their own science” to the platform.
- Provide templates and intelligent defaults with override capabilities. Data discipline required for AI can be embedded in default configurations, and model templates can incorporate best practices for a particular industry.
- Keep expert humans in the loop at crucial checkpoints. Especially when building and rolling out AI, it’s valuable to have an advisor (internal, partner, or software provider) who can review results and outcomes for integrity and provide additional best practices.
How do you envision the role of data governance evolving in pricing teams, and what should leaders prioritize in the next 12–18 months to future-proof their AI investments?
Given the rapid adoption of AI over the past few years, many organizations are just getting started on the data governance front. But as governance evolves, some likely trends include:
A shift from reactive to proactive. Traditional data governance focused on fixing problems as they occur. As pricing teams embed AI in their processes, it will be crucial to better anticipate and prevent data quality issues.
An increased comfort with automation. Manual data governance processes can’t scale to meet the data appetite of AI. Organizations will need to design and implement automated workflows that enforce data quality, track the lineage of data, and adhere to any compliance requirements with minimal human intervention.
A governance partnership between IT and Business. Data governance is evolving to be valued as a core business competency, not just a technical function. Pricing teams must help define feature engineering logic, the approach to addressing anomalies, and the roadmap for curating and evaluating additional data sources.
A focus on explainable AI. With modern data infrastructures, pricing AI does not have to be a black box! It doesn’t matter how sophisticated the data processes and algorithms are if the end users of the recommendations don’t trust and adopt them. Ideally, anyone in the pricing organization should be able to drill down into the data sources, understand the workflow associated with data processing, and have a working understanding of the logic applied by the AI.
A quote or advice from the author : Using AI for pricing isn’t just about the sophistication of your algorithms. It’s about the quality of your data, the wisdom of your governance, and the trust of your stakeholders. And the most transformative pricing AI implementations start with specific, high-value use cases and expand through proven success.
Suzanne Valentine
Director of Pricing AI at Pricefx
Suzanne (Suzy) Valentine is Director of Pricing AI at Pricefx. She brings 25+ years of experience in enterprise software and AI-powered merchandising analytics to her role. Prior to joining Pricefx, Valentine led data science teams and initiatives at a variety of organizations, including Meta, IBM, Procter & Gamble, and DemandTec. Pricefx is the global leader in AI-powered pricing software, offering an end-to-end platform solution that delivers the industry’s fastest time-to-value.