Walk into any large pharma company today and you’ll find dashboards everywhere. Sales dashboards. KOL engagement dashboards. Congress activity dashboards. Quality dashboards for the dashboards.
The output of two decades of digital transformation looks impressive. The problem is that almost none of it actually drives a decision. It describes. It rarely decides.
A dashboard answers “what happened?” A decision engine answers “what should we do next?” — and increasingly, takes the first step. That gap is where the real promise of AI in pharma sits, and it’s where most teams are still standing on the wrong side.
What a Dashboard Actually Does
Dashboards are passive by design. They visualize state. They wait for a human to walk past, glance at a number, interpret it through the lens of their experience, decide if it’s good or bad, and then — maybe — do something.
This worked when the volume of state to monitor was small and the people monitoring it had time. Neither is true anymore.
A Medical Affairs team in oncology today is tracking dozens of trials, hundreds of KOLs, real-time congress output across multiple therapeutic areas, regulatory shifts, competitor moves, omnichannel campaign performance, and the operational backbone underneath all of it. The dashboard layer just keeps growing — and the cognitive load to interpret it grows with it.
The honest assessment most leaders won’t say out loud: nobody is actually reading half of these dashboards. They get built, they get demoed, they get budget approval, and they quietly turn into wallpaper.
What a Decision Engine Does Instead
A decision engine inverts the relationship. Instead of waiting to be read, it actively scans the same data, applies a layer of rules and judgment, and surfaces only what needs human attention — with the recommended next step already framed.
In a Medical Affairs context that looks like:
- Anomaly detection that triggers escalation. Not “here’s a chart of KOL engagement over time” but “Dr. K’s interaction pattern dropped 60% this quarter — likely cause: new competitor advisory board. Suggested action: re-engage via the upcoming congress symposium.”
- Insight routing. A new publication drops in your therapeutic area. The system reads it, tags relevance to ongoing strategic priorities, and pings the right MSL with a one-paragraph summary and a draft talking point — instead of leaving it to land in a journal alert nobody opens.
- Trigger-based content drafts. Field insight comes in from the territory. The engine drafts the response card, populates the tracker, surfaces it for review. The MSL approves or edits, not types from scratch.
The pattern is the same in each case: the system does the boring 80% so the human can spend their attention on the judgment-heavy 20%.
Why Pharma Keeps Stopping at Dashboards
Three reasons, in order of how often I hear them.
The first is technical comfort. Dashboards are a known quantity. Every BI vendor sells them. Every IT team can stand one up. Decision engines require integration work, governance frameworks, and the kind of cross-functional sponsorship that’s harder to get.
The second is compliance fear. Letting an automated system suggest an action — let alone take one — feels riskier than just showing a number. It’s not, when designed correctly. The audit trail of a well-built agent is dramatically cleaner than the audit trail of “a human looked at a chart and made a call.” But the perception slows adoption.
The third is the most uncomfortable. A decision engine makes the gaps visible. If your KOL engagement framework is mostly vibes, a system trying to operationalize it will fail loudly. Dashboards let teams keep their tacit knowledge tacit. Decision engines force it into the open. That’s good for the org and uncomfortable for individuals who built careers on being the gatekeeper of “how we do things.”
A Practical Path From One to the Other
You don’t replace the dashboard layer overnight. You build the decision layer alongside it.
- Pick one decision that’s currently bottlenecked on dashboard interpretation. Something where a person regularly stares at a chart, mentally applies a rule, and acts. That rule is your starting prompt.
- Write the rule explicitly. This is the hard step. “I escalate when X drops below Y for two weeks unless Z.” If you can’t write it, you don’t yet have a decision worth automating.
- Build a thin slice. One data source, one rule, one suggested action, one human approver. Don’t try to boil the ocean.
- Log every trigger, every action, every override. This is your governance backbone — and your training data for round two. Making AI Harnesses Audit-Proof in Pharma covers what that logging needs to look like in a regulated context.
- Iterate the rule. Three months in, look at the overrides. The overrides are where your rule is wrong. Fix it. The decision engine gets smarter; the human burden gets smaller.
This is how agentic workflows actually take hold in pharma — one bottleneck at a time, with governance in place, not as a moonshot. And it only works if the substrate underneath is right; if your information still lives in PDFs and PowerPoints, the engine has nothing to chew on. Markdown as the operating layer is what makes the rest of it possible.
The Bigger Picture
Every dashboard in pharma today represents a bet on the assumption that the bottleneck is information visibility. That bet was right for a long time. It isn’t right anymore.
The bottleneck now is decision latency: the time between when the data says something is happening and when a human gets around to acting on it. AI doesn’t fix that by making the dashboard prettier. It fixes it by closing the loop.
The companies that figure this out first won’t have better dashboards. They’ll have fewer of them — and the ones they keep will be the ones that actually decide.
This article was co-authored with Anthropic’s Claude Opus 4.7 model. The ideas, domain expertise, and editorial direction are mine — the AI helped structure, draft, and refine the text.
