If I ran commercial at a pharma company today, I would not start with a tool. I would start by counting how many days pass between a brand decision being made and the first rep actually saying it to a doctor. In most companies that number is somewhere between six weeks and never, and nobody owns it. Everything below is built around shortening that one number.
This is not a list of products to buy. It is the order I would do things in, and the order matters more than the tools.
Step 1: Spend the first two weeks counting days, not building anything
Before any budget moves, I would pick the last three brand decisions we made and trace each one forward.
Sign off date. Date the content went into review. Date it cleared. Date it hit the cycle meeting. Date the first call happened where a rep actually delivered that message. Then the gap between the first and the last.
I would do the same backwards for one competitor move the field spotted, and see how long it took to reach the brand team and whether anything changed as a result.
Three traces forward, one trace backward. That is the whole diagnostic, and it takes two weeks because you are chasing people for dates, not because the work is hard.
You will find the delay is not spread evenly. It is concentrated in two or three queues, and they are almost never where leadership assumes.
Step 2: Start with the review queue, because it is the most expensive thing you own
In mid and large pharma, review cycles commonly run 50 to 60 days per piece of content against an industry benchmark closer to 21 days. Reviewers routinely describe promotional pieces coming back ten or more times before they clear.
That is the single biggest queue in a commercial organisation, and it is where I would start. Not because it is glamorous, but because it is measurable, everyone already hates it, and nobody has to change their job description for it to get better.
What I would actually do: put an AI pre check in front of submission, not inside the review itself.
Before anything goes to medical, legal, or regulatory, it gets checked against the approved reference pack. Every claim mapped to its source. Missing qualifiers flagged. Language that overstates efficacy flagged. Prior review comments on similar pieces surfaced so the same correction is not made for the eleventh time.
The reviewers keep every bit of their authority. They are not being replaced or sped up. They are simply receiving cleaner submissions, which is the actual cause of the ten rounds. Teams that restructure review this way report cutting cycle time meaningfully while keeping the same compliance standard.
I would be explicit about this internally on day one, because the fastest way to kill an AI project in pharma is to let medical and regulatory believe you are trying to automate their judgment. You are not. You are fixing what lands on their desk.
Step 3: Replace the rep dashboard with a brief
Reps do not read dashboards. They look at them when a manager asks about them.
Across sales roles generally, reps spend only around 30 percent of their time actually selling. In pharma the shape is similar, and a large share of the rest goes to preparation and admin that a system already has the data for.
So instead of another dashboard, every rep gets a short brief before each call. Not a report. Five or six lines:
- What this doctor said last visit, in their words
- What has changed since, including any prescribing shift
- What was promised and whether it happened
- The one thing worth asking this time
- Anything from the field about this account in the last month
That is it. It has to be readable in the car before walking in.
The reason this works is not that it saves time. It is that it makes the previous call matter. Right now most call histories are written for compliance and never read again by a human being.
Step 4: Kill the call report and capture it by voice
Roughly 71 percent of field sales reps spend five or more hours a week on manual CRM entry. That time buys the company data that is mostly bad, because it is typed at 7pm from memory.
I would remove the typed call report entirely.
Instead the rep talks for sixty seconds after leaving. What happened, what the doctor pushed back on, what they need. The system turns that into the structured fields the CRM wants, flags anything that needs a compliance eye, and routes anything urgent to the right person the same day.
Two things improve at once. The rep gets hours back. The company gets better data, because it was captured while the memory was fresh instead of reconstructed on a Friday night.
I would be honest that this needs a serious conversation with compliance before it runs, and that recording anything involving an HCP has rules that differ by market. That conversation is part of the project, not an obstacle to it.
Step 5: Stop the message being diluted on the way down
This is the one that actually costs the most and gets measured the least.
A brand decision is made at country level. It goes to a cycle meeting. Field managers interpret it for their regions. Reps interpret it again for their territories. By the time it reaches a doctor it has been through four translations, and nobody checked whether it survived.
I would not try to control this by sending more slides. More slides is what caused it.
Instead, each layer gets a working assistant that holds the original intent and the boundaries, and helps that person translate it for their own context without losing the core. A regional manager asking how to position this for their accounts gets an answer grounded in the actual brand direction and the actual approved claims, not their memory of a meeting three weeks ago.
The point is not to make everyone say the same sentence. A rep in a different market should adapt it. The point is that what they adapt is the real strategy rather than a fourth hand summary of it.
Step 6: Give the brand team the field's intelligence every week
Right now field intelligence dies in three places: a WhatsApp group, a manager's head, and the free text box at the bottom of a call report that nobody queries.
Meanwhile the brand team is planning next cycle on data that is two months old.
The voice capture in step 4 already collects this. So the only new thing is a weekly digest to the brand team: what competitors did, where access problems appeared, which objections are rising, which messages are landing and which are being ignored.
One page. Weekly. Same format every time so it becomes readable at a glance.
This is the cheapest step on the list and it is usually the one that changes behaviour fastest, because brand managers have never had this and immediately want it.
Step 7: Measure two numbers and ignore everything else
I would report exactly two things upward.
Days from brand decision to field execution. The number from step 1. This is the number the whole programme exists to move.
Share of calls where the intended message was actually delivered. Not sent. Delivered, and captured as such.
I would refuse to report licences issued, logins, messages generated, or time saved per rep. Those numbers always go up and they have never once proved anything to a general manager.
If those two numbers move, the programme worked. If they do not, it did not, regardless of how good the demos looked.
What I would not touch in year one
Being straight about this matters more in pharma than in most industries.
Nothing patient facing. Not adverse event handling, not patient support messaging, not anything that could be read as medical advice. The risk is asymmetric and the upside is not worth it in year one.
Not forecasting. It is the most requested AI project in commercial and the worst place to start. The data is messy, the attribution is contested, and a wrong forecast is very visible.
Not replacing medical or regulatory judgment. See step 2. Support the queue, never the decision.
Not the biggest brand. Start on a stable, mid priority brand where being wrong once is survivable. The flagship is where you go second, with evidence.
The honest summary
Most commercial AI programmes in pharma fail because they start with a tool that makes one job faster, in an organisation where the delay was never inside the jobs. It was in the queues between them.
Review queue. Cascade. Field intelligence coming back up. Those three are where the weeks go.
Fix those in that order, measure the two numbers, and stay out of anything patient facing until you have something proven to point at.
Frequently asked questions
Where should a pharma commercial team start with AI? Start with the content review queue. Review cycles commonly run 50 to 60 days per piece at mid and large pharma against a benchmark closer to 21 days, so it is the largest single queue in the organisation. It is also measurable, widely disliked, and improvable without changing anyone's job description.
Can AI replace the MLR review process in pharma? No, and attempting it is the fastest way to lose medical and regulatory support. The workable approach is a pre submission check that maps every claim to its source and flags missing qualifiers before a piece reaches reviewers, so reviewers receive cleaner submissions while keeping full authority over the decision.
How much time do pharma reps lose to administration? Sales reps across industries spend roughly 30 percent of their time actually selling, and about 71 percent of field reps spend five or more hours a week on manual CRM entry alone. Capturing call notes by voice immediately after a visit recovers much of that and produces better data than typing from memory later.
Why does brand strategy get diluted before it reaches doctors? Because it passes through several translations, from country to cycle meeting to regional manager to rep, and nobody checks whether the original intent survived. Adding more slides makes it worse. Giving each layer access to the original strategy and approved claims lets them adapt the message without losing it.
What should a commercial director measure to prove AI worked? Two numbers: days from brand decision to first field execution, and the share of calls where the intended message was actually delivered. Licences issued, logins, and time saved per rep always rise and have never proved anything to a general manager.
