What Does the 74% Experimentation Number Actually Mean?
The American Supply Association's January 2026 survey of PHCP-PVF distributors found 74% of respondents actively experimenting with AI in some form. On the surface, this sounds like an industry on the cusp of transformation. But the same survey found only 19% have implemented AI across strategic business segments - meaning AI is embedded in a core process, measured, and being expanded.
The 55-point gap between those two numbers is the most important data point in the survey. It describes an industry where AI experimentation has become table stakes but AI implementation remains rare. Most distributors have tried something. Very few have deployed something that demonstrably changes how they operate at scale.
Understanding why that gap exists - and what the 19% did differently - is more valuable than any specific AI tool recommendation.
Why Do PHCP Distributors Get Stuck in Pilot Purgatory?
Pilot purgatory is the state where a promising AI experiment never converts to a production deployment. It is endemic to distribution because of three structural factors unique to the industry.
First, ERP dependency. Most PHCP distributors run their business on an ERP system (Epicor, Eclipse, Infor, or a legacy custom system) that was not designed for AI integration. Connecting an AI application to live transaction data requires API work, IT resources, and vendor cooperation that many mid-market distributors do not have readily available. Pilots run on data exports. Production requires real-time data access. That gap kills many deployments.
Second, data quality underestimation. Leaders planning an AI deployment frequently assume their data is cleaner than it is. When implementation begins and the AI starts ingesting real data, the errors surface: duplicate customer records, inconsistent product codes across branches, historical transactions tagged to wrong accounts. Cleaning this data is not glamorous work, and it takes longer than planned, stalling the deployment timeline.
Third, wrong ROI metric. Many distributors measure AI pilots by adoption rate - how many reps are using the tool - rather than by business outcome. A tool with 40% adoption that cuts quote time in half generates more value than a tool with 80% adoption that saves 10 minutes per week. When the measurement is wrong, decisions about whether to scale are made on incomplete information.
What the 19% Did Differently
Distributors who have moved from experimentation to strategic implementation share a consistent pattern. They selected a single use case where the manual process was painful enough that people were motivated to change behavior. They cleaned the data that specific use case depended on before deploying AI on top of it. And they measured one business outcome - not adoption, not user satisfaction, but a number that appeared on a P&L or operational dashboard.
The use cases that most commonly cleared these three bars in PHCP distribution: quote automation (high frequency, measurable time savings, data exists in ERP), demand forecasting for seasonal products (direct inventory cost impact, transaction history is typically clean), and customer health scoring for churn detection (measurable in revenue recovered from reactivated accounts).
The PHCP-Specific Data Readiness Assessment
Before evaluating AI vendors or tools, a PHCP distributor should conduct an honest data readiness assessment across five dimensions:
- Product master completeness: What percentage of your active SKUs have complete specification data (manufacturer, model, UOM, dimensions, certifications)? If below 70%, quoting and substitution AI will underperform.
- Customer data consolidation: Do you have a single record per customer that aggregates all purchase history, contacts, and pricing tiers? Or is that data fragmented across branches, reps' spreadsheets, and legacy system fields?
- Transaction history depth: Do you have at least 24 months of clean transaction data (order date, customer, SKU, quantity, price, branch)? This is the minimum for meaningful demand forecasting and churn detection.
- ERP integration capability: Can your ERP expose data via API or scheduled export to a third-party system? What is your IT team's capacity to support integration work?
- Change management readiness: Do you have a champion in the business - not IT, but a sales or operations leader - who will drive adoption of a new tool within their team?
Score each dimension honestly. The dimensions where you score lowest are the implementation risks that need mitigation before you select a vendor.
The Build vs. Buy Decision in 2026
Two years ago, most PHCP distributors evaluating AI faced a stark choice: buy an off-the-shelf tool and accept its limitations, or build a custom solution and accept its cost and timeline. That choice has narrowed considerably. Several ERP vendors now offer AI modules (Eclipse AI, Epicor's CoPilot features), and a new generation of distribution-specific AI vendors have emerged with pre-trained models on distribution data.
The build-vs-buy calculus for 2026: if your use case is standard and your ERP vendor offers a native AI feature for it, start there. The integration cost is lower, the time-to-deploy is faster, and the initial training data is already structured correctly. Custom builds are warranted only when your use case is genuinely differentiated - a proprietary pricing model, a unique substitution logic, or a customer segmentation approach that is core to your competitive strategy.
The 74% experimenting have mostly tried generic tools. The 19% implementing have mostly committed to a specific integration with their production data. That specificity is the difference.
All Tools