Data Science Philadelphia (DataPhilly)

We’re kicking off a new DataPhilly webinar series focused on the real-world challenges of applying analytics, ML, and AI in business—and our first session tackles a problem many teams know too well: great models that never get used. This is session 1 out of 3.
Most analytics teams assume adoption fails because models aren’t explainable or sophisticated enough. In reality, resistance often comes from how models shift control, accountability, and decision authority inside organizations. This session challenges analytics and ML professionals to rethink what it means to build “good” models—not just technically correct ones, but models designed for trust, acceptance, and real decision-making.

What you’ll learn:

• Why validated, accurate models are often ignored in pricing, forecasting, and commercial decisions
• How model outputs, confidence intervals, and “optimal” recommendations can build—or erode—trust
• What repeated requests for tweaks really signal about decision risk
• Why designing for adoption may be as important as designing for accuracy

If you’ve ever asked yourself, “If a model is accurate but never acted on, is it really a good model?”—this session is for you.

Abstract: Why Good Models Fail in Business Decisions
Most analytics teams believe their biggest challenge is building better models. In practice, the harder problem is getting good models used. Pricing, forecasting, and commercial decisions are full of technically sound analyses that were validated, approved, and then quietly ignored.
This session challenges analytics and ML professionals to rethink why adoption fails. Drawing on real-world experience, it argues that resistance to machine learning is rarely about lack of explainability or technical sophistication. Instead, it reflects how models redistribute control, accountability, and decision authority inside organizations.
Rather than asking how to simplify models for business users, the talk asks a more uncomfortable question: what responsibilities do analytics teams have in designing for acceptance, not just correctness? The session explores how model outputs, confidence intervals, and “optimal” recommendations can either build trust or undermine it, and why repeated requests for tweaks are often signals of unresolved decision risk.
Provocation: If a model is accurate but never acted on, is it really a good model?