The Market Changes Weekly. The Company Decides Once a Year.
One way to find where an AI decision system will pay off is to compare two frequencies: how often a decision’s inputs change, and how often the decision itself is made. The wider the gap, the stronger the case.
When the basis for a decision changes every week but the decision is made only once a year, that gap is often where a decision system helps most. When I look for places to apply AI, I start by looking for gaps like this, before asking where people complain the most or what’s easiest to automate.
I’ve spent six years in pharma commercial, and it has the widest gap I’ve seen. I’ll use it as the example to show why the two frequencies drift so far apart, and what that costs.
The prescription drug business
In most businesses, the person who chooses, the person who uses and the person who pays are the same, or at least sit at the same table. Prescription drugs are different. Doctors write the prescriptions, patients take the drugs, public insurance pays most of the bill, and drug companies are limited to compliant scientific promotion. They can’t even advertise prescription drugs to patients.
The decision a drug company wants to influence is the doctor’s prescription. But the person making it isn’t spending their own money, the one paying doesn’t make it, and the company’s own tools are tightly boxed in by regulation. So the whole commercial operation, from the sales force and medical conferences to market access and distribution, comes down to one thing: putting a limited budget where it’s most likely to influence prescriptions indirectly.
Where the money should go is the central decision in this business.
The curve bends
Intuitively, it looks like a straight line: more reps and more visits mean more sales, and cutting spending by 20% cuts sales by 20%.
In fact the line bends. Put spending on one doctor, say the number of visits, on the horizontal axis, and the prescriptions that doctor writes on the vertical. The first few visits make a clear difference: the doctor learns about the drug, looks at the evidence and starts trying it. After that, each extra visit adds fewer prescriptions, until the curve almost flattens out.
So how much volume is lost by cutting 20% depends on which 20% gets cut. Cut where the curve is already flat and sales barely move. Put the other way, part of today’s spending is never quite worth it: loyal customers visited too often, and low-potential doctors who stay on the list year after year. I once built a coverage model for a retail pharmacy team. For the high-value stores, the staffing actually deployed was more than forty percentage points off what the model said it should be. The team wasn’t short of people. They just weren’t where they could do the most good.
Finding that spending and moving it to where the curve is still steep is the reason sales force effectiveness (SFE) exists as a function.
The curve can change overnight
The trouble is that the curve isn’t fixed. In China, a single policy can redraw it completely.
Win a VBP tender (China’s volume-based procurement) and the price drops sharply in exchange for guaranteed purchase volumes at public hospitals. Hospital volume is effectively locked in by policy, and extra visits bring in very few extra prescriptions. Lose the tender and it’s the opposite: the hospital market is mostly gone, and the business has to turn to retail pharmacies, online channels and private hospitals. Get onto the national reimbursement drug list through price negotiation and it’s a different game again. Being on the list doesn’t get a drug into hospitals. It still has to get through each hospital’s drug committee one by one, and at that stage the sales force is worth more than ever.
The same drug in three policy situations calls for completely different strategies. However sensible last year’s allocation of resources was, it can be entirely wrong the day a policy result is announced. Beyond policy, competitors launch, hospitals change directors, doctors move to other cities. Across a national market, the basis for allocation shifts almost every week.
Yet allocation happens once a year
Allocation is SFE’s job: which customers to cover, how often to see each one, how many people are needed, how to draw territories, how to set quotas, how to pay bonuses. It runs on an annual cycle. Once a year there’s a planning season lasting a few months, when last year’s data is used to build next year’s plan, and after that the plan barely changes. I once took apart a pharma company’s annual plan; the full calculation used 12 Excel tools. When one run takes that much effort, nobody wants to run it again mid-year.
Nobody is to blame for this. When judgment is expensive, this work can only be afforded once a year. Why that makes it a compromise is the subject of a separate post, Annual Planning Is a Compromise.
Then again, some of the slowness is deliberate. If quotas changed every week, reps couldn’t get on with their work, and they’d stop trusting the company. So the decision system I have in mind doesn’t change everything every week. Someone looks every week, can make a change when one is needed, and can explain why when nothing changes.
Beyond pharma
The same approach should work outside pharma. What I do is list the decisions a business makes again and again, such as pricing, production scheduling, customer allocation and budget adjustments, and write two numbers next to each: how often its inputs change, and how often the decision is made.
The row where those two numbers are furthest apart is usually where I start. If the data that row needs is already in the company’s systems, the first project starts to take shape.
