Everyone Asks If You're Using AI. Who Asks You If It's Working?

The question I hear most often from pharma access and market access leaders right now is some version of: "Are you using AI?" It sounds simple. It is anything but.

Anyone who’s made a career in pharma access recognizes a familiar pressure when something arrives in new packaging. AI adoption looks different from past technology cycles, but the underlying dynamic is the same: executives and boards want tangible progress, vendors want to showcase their ability to deliver brand outcomes, and access teams are singularly accountable for patient adoption.

Our recent market survey of brand and commercialization leaders, from global top 20 to small specialty companies, tells an interesting story. More than 80% of patient and market access professionals said it's appealing when a vendor is positioned as tech- or AI-forward. But when those same respondents were asked what they expect from a tech-forward partner, 61% said smarter, more adaptive execution. Not AI specifically, just results.

Your real opportunity isn't to adopt AI. It's to define what "better" looks like for your brand's specific problems, and then decide where AI belongs in that picture. The same respondents cited their top three markers of partner success: reduced patient abandonment, increased new prescriptions, and high execution accuracy. Technology alone delivers none of those. Execution strategy does. AI can support it, in the right places, for the right reasons.

I sat down with ConnectiveRx Chief Product Officer Cindy Baksh and Chief Technology Architect Steve Randall to work through what that looks like in practice and define the questions access leaders should raise before the next vendor meeting, executive update, or budget cycle.

Dowd: Pharma leaders are under pressure to incorporate AI into their programs and prove measurable impact. How should leaders define an AI strategy that starts with the business problem, not the tool?

Baksh: For pharma manufacturers, AI should be treated as a strategic capability that strengthens the patient support ecosystem, not a standalone technology initiative. The friction comes when the hype of AI runs ahead of its practical application. The partners getting it right start by defining the business problem, then build a prescriptive AI strategy around the specific places where AI can measurably improve performance. That means anchoring every use case to the outcome it is meant to move, whether that is faster time to therapy, reduced abandonment, stronger accuracy, or better pull-through.

The next generation of patient services partners will differentiate themselves not by the AI technologies they deploy, but by the measurable outcomes they deliver.

Dowd: Where do AI initiatives most commonly break down? Inside a pharma organization, or inside a partner relationship?

Baksh: There are many opinions here, but PwC put it directly in their 2026 AI Business Predictions: technology should deliver only about 20% of an AI initiative's value. The other 80% comes from redesigning how work gets done, so you have both your people and your AI working where they can uniquely make the most impact.¹ Across the pharma companies I talk to, global manufacturers and specialty companies alike, most are running that ratio in reverse. The energy goes into tooling and agents. But without the organizational retooling to change roles, workflows, decision rights, and handoffs around what AI produces, the metrics leaders care about most – faster speed to therapy, lower abandonment, greater accuracy, stronger pull-through – never fully materialize. AI only creates value when it functions alongside humans in a redesigned operating model; otherwise, the tool may be deployed, but the impact gets lost.

Dowd: That tracks with what I hear from access leaders. The sharper question isn't "what's your AI roadmap?" It's "are we structured to act on what AI surfaces?" Steve, how does that show up from the technology side?

Randall: Absolutely. That's the conversation worth having internally before any partner comes through the door. If your internal workflows aren't designed to absorb what AI produces, the tool becomes shelfware, regardless of how good it is.

A credible vendor should be able to explain where they are not using AI and why. Some problems genuinely need it. Others need direct data sources, well-defined business rules, and answers that can't vary. Knowing the difference, and being willing to say so, is what separates a considered approach from a technology showcase.

Healthcare also comes with a different set of expectations. Manufacturers increasingly want to know exactly where AI is being used, how it's governed and whether their own compliance and legal requirements support those use cases. Those aren't obstacles, they're essential guardrails.

Dowd: In patient access specifically, where does accepting a probabilistic answer carry the most risk?

Randall: Benefits verification is the clearest example. A probabilistic answer, one that's "usually" right, is a completely different thing than a correct answer. When a patient's access to therapy is on the other end of that distinction, "usually right" isn't good enough. The question worth asking any partner: when the answer is wrong, what does that cost the brand?

Dowd: And that cost question tends to land differently once a brand has felt it. Cindy, how does that play out in eBV, specifically?

Cindy: Most players in the hub and patient support space use predictive analytics for eBV, meaning AI scans thousands of claims to estimate what a patient's benefit result probably is. We take a different approach. Every eBV we process comes from direct payer-source data. The AI layer we're building on top of that adds a confidence assessment, something like a FICO score for the result. If a drug almost always requires prior authorization and a payer comes back saying no PA is needed, that's a flag. The AI is there to strengthen accuracy, not substitute for sound underlying process.

Dowd: Where do you see AI creating meaningful impact when it's matched to a specific problem?

Randall: In our solution called Gatekeeper, AI brings relevant information together for 310 million people and more than 100 patient identity elements in less than a second to accelerate decision-making to screen patients who aren't in your target population. But there's always a clear path for human review, and that's by design. In patient access, you don't remove the expert from the process. Effectiveness means helping them reach a better decision point faster.

Baksh: And you see that same thinking in ShieldRx, our copay misuse prevention solution, AI identifies patterns that warrant closer attention, so potential impropriety is caught before payment occurs. We’re applying that same logic to payment anomaly detection in buy-and-bill programs, flagging unusual patterns earlier to add another layer of confidence that every payment is accurate. Different use cases, same logic: apply AI in targeted areas where it genuinely improves accuracy, speed, or decisioning. That’s where you’re going to see measurable wins in your own brand metrics.

The goal was never to use AI everywhere. It was to get patients to therapy faster, execute more accurately, and support the people doing this work in making smarter decisions. Access leaders who anchor every AI conversation to those outcomes will make better decisions for their programs.

Reference

1. PwC. 2026 AI business predictions. PwC. Published December 2, 2025. Updated July 22, 2026. Accessed July 31, 2026. https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-predictions.html