The Bullwhip Effect Hits Clinical Supply Chains

By Eshaan Jain, Senior Consultant

A site coordinator logs two extra patients into an interactive response technology (IRT) system this week. That is a small number. By the time a depot reorders against it, that number can turn into pallets of drug product shipped somewhere it was never needed, while another depot runs short of the same product for reasons unrelated to real patient demand.

I saw this same pattern for years at Amazon, where I worked as a senior technical PM on a last-mile supply chain logistics team and helped build a machine learning system that extracted clause data directly from contract text across a $40 billion+ annual portfolio at 95% accuracy. Small variances at the customer end of that network routinely turned into large swings in orders, staffing, and inventory several tiers upstream. Supply chain researchers gave this pattern a name decades ago: the bullwhip effect.

The same distortion runs through clinical trial supply. A small change in enrollment or dosing at a site becomes an oversized order at a depot, or a stockout unrelated to actual patient demand, by the time the signal passes through IRT, depot inventory systems, and central planning – three tiers of avoidable variance, all traceable to how the signal moves rather than to what patients actually need.

What The Bullwhip Effect Actually Is

Hau Lee, V. Padmanabhan, and Seungjin Whang described the pattern formally in a 1997 Sloan Management Review paper: the variance in end customer demand grows at each tier as the order signal moves upstream from retailer to distributor to manufacturer.¹ They traced it to four causes that do not require anyone to make a mistake. Each tier updates its demand forecast independently, rather than using the real signal from the end customer. Orders are batched into periodic runs rather than being placed continuously. Promotional pricing pulls purchases forward and creates artificial spikes. Rationing behavior during a shortage causes buyers to overorder in anticipation of receiving less than they requested.¹

Every one of those four causes has a direct clinical trial analog. A site reports enrollment on a schedule, batched by monitoring visits or data entry cycles. A depot reorders when a threshold is tripped, which converts a smooth trickle of actual dosing into a step function of large orders. Central planning, several tiers removed from the patient, sees the depot's order pattern rather than the patient's dosing pattern and forecasts against that increasingly distorted signal.

Where The Signal Gets Amplified

Investigational medicinal product (IMP) waste in clinical trials runs high in part because of this kind of forecasting distortion. McKinsey has put the average clinical trial medication waste at as high as 50%.² That figure reflects supply chain planning against the wrong signal at each handoff, tier by tier, well before a dose nears its expiration date.

IRT systems worsen the situation in a specific mechanical way when they run on fixed reorder thresholds. A depot set to reorder whenever inventory drops below a fixed level places identical-sized orders, regardless of whether the dip reflects a real increase in enrollment or a one-week reporting delay from two sites. Suvoda's review of clinical IRT deployments cites N-SIDE, a supply forecasting vendor, as showing that realistic IMP waste from manual spreadsheet-based forecasting is closer to 70% once true drug utilization is accounted for, compared with the 30% figure trial teams tend to assume going in.³ The same review found that IRT systems configured with flexible reorder parameters, instead of fixed thresholds, cut supply costs by 10% to 20%.³

How Last-Mile Networks Dampen The Same Distortion

Dampening this kind of distortion within the last-mile network involved three practices, each aimed at having every tier work off a single demand signal rather than a chain of derived delayed signals.

The first was a single point of demand truth. Instead of each tier in the network keeping its own version of demand, one for the warehouse, one for the carrier, and one for the regional operations team, the goal was a single system that every tier read from and that was updated in near real time. Reconciling separate versions after the fact was where distortion crept in, because each reconciliation step introduced lag and rounding errors.

The second was smoothing the batch size. When orders got placed in large periodic batches instead of smaller continuous ones, the batching itself created spikes unrelated to underlying demand. Moving toward smaller, more frequent order cycles removed a source of noise before it could compound upstream.

The third was structured data at the source. The clause extraction system I helped build read contract terms directly from the original agreement text with 95% accuracy, covering the $40 billion+ portfolio, instead of relying on a summary or a manually rekeyed version that a downstream team maintained on its own. Every additional hand that touches a number before it reaches central planning is a place where that number can drift from what actually happened.

The Same Principle In CPQ And CLM Data

I apply the same principle now leading Salesforce CPQ (configure, price, quote) and CLM (contract lifecycle management) transformations. Contract and quote data used to live in separate systems that updated on different schedules, and downstream teams kept their own local copies to work around the lag. Unifying CPQ and CLM data into a single governed system of record, rather than letting each downstream system reconcile its copy on its own timeline, cut contract creation time by 50%. The reduction came from removing a chain of derived signals between systems that previously updated independently of each other.

Clinical supply organizations run the same architecture problem across IRT, depot inventory, and central planning systems that were built and configured at different times, often by different vendors, each with its own update cadence. Each of those cadence gaps is a place where a small enrollment or dosing change can be restated, rounded, or delayed before the next tier acts on it.

Three Practices That Carry Over Directly

First, treat IRT as the single point of demand truth for depot and central planning decisions. If depot planning or central forecasting works from a spreadsheet extract, a periodic report, or a locally maintained copy of enrollment data rather than directly from the IRT system, that extra copy is a place where the signal can drift and be delayed. Route depot and central planning decisions through the same live system that every site enters data into.

Second, smooth reorder batch sizes at the depot level. A fixed threshold that triggers one large periodic reorder converts small week-to-week enrollment noise into large order swings. Reordering in smaller, more frequent increments tied to actual dispensing, rather than a fixed inventory floor, prevents normal enrollment variance from resulting in an oversized order or an unnecessary stockout.

Third, separate the signal from the correction. When a site underreports for a week and catches up the next, or a depot pads an order to cover a prior shipment's shortfall, central planning needs to see that correction labeled as such. An unlabeled correction is one of the most common ways a single one-time event turns into a recurring forecast assumption several tiers upstream.

Where To Focus The Next Review

A useful test for any clinical supply team: pick the last unplanned order or stockout a depot flagged and trace it backward through the IRT log to the specific enrollment or dosing entry that triggered it. If that entry does not reflect a real current change in patient demand, the distortion occurred during the handoff between site, depot, and central planning, and the next process review should focus on that handoff.

References:

1. Lee, H.L., Padmanabhan, V., and Whang, S., "The Bullwhip Effect in Supply Chains," Sloan Management Review, Spring 1997. Full text: Georgia Tech ISYE course archive. Overview: Understanding the Bullwhip Effect in Supply Chains, MIT Sloan Management Review
2. McKinsey & Company, cited on average clinical trial medication waste of approximately 50%: Digitalized Drug Forecasting Minimizes Waste In Clinical Trial Supply Chain, Clinical Leader
3. N-SIDE realistic IMP waste and IRT cost reduction figures, cited in: How To Cut Drug Supply Costs With Seamless Forecasting And IRT Systems, Suvoda

About The Author:

Eshaan Jain is a senior enterprise product leader with almost 15 years of experience at the intersection of AI/ML, Quote-to-Cash (CPQ), and Contract Lifecycle Management (CLM). Currently serving as a Salesforce Product Owner / Senior Product Manager at a management consulting firm.

Eshaan has also spent nearly five years at Amazon (2020–2025) as a Salesforce Technical Product Manager. He was responsible for the Global Supply Chain Transportation Procurement (GSCTP) platform, where he managed a $40B annual procurement spend and led the digital transformation of Amazon's entire last-mile transportation contract lifecycle. 

Eshaan's career also includes foundational leadership roles at PricewaterhouseCoopers (PwC) and Accenture, specializing in Salesforce assurance and IT controls.

In addition to his corporate achievements, Eshaan is an active researcher in the field of AI and Machine Learning. He has published three peer-reviewed papers with IEEE and Elsevier. He holds a Master of Science in Computer Science from the University of Southern California (USC) and a B.Tech in Computer Science from GGSIPU in Delhi, India.