FC Stow Performance and Throughput Dynamics

I argue that the current system (at Amazon PER4) of performance measurement (KPI/Units Per Hour) for Stowers can be improved, both in measurement and in optics. Stowers are subject to weekly performance check-ins. For example, if an Associate is under performing (say at 85%) for the previous week, then a 1:1 discussion will take place in the Aisles. This is an opportunity for the associate to explain why (or not), but also to explain the performance expectations according to their level. In this chat, it may come to surface that there was an error or explanation that explains the ‘performance’, so in fact the measurement is actually measuring something else, and not the actual units stowed.

However, how this percentage is calculated is not explained, and the factors around the associates tasks may be innaccurate due to human error. A FC runs in an interlinked process. If no items are available to stow, then rates will decline. Changing tasks is done by a PA, who might make a mistake, or misrepresent the time due to complexity. An associate is not compelled to update their PA (Process Assistant) due to the intensity of work, and lack of availability. Keeping a personal track of tasks, problems, items is difficult and also adds time.

I isolate truth from the hand barcode scanners in a Units Per Session KPI to determine the rate, which allows the conversation to focus on ‘the gaps’. The gaps being the operations.

These ideas and numbers are available to stakeholders. It is just not a driver in daily decision making to certain floor staff. This is probably due to the nature of accuracy and repeatability as you transition and improve a FC on a weekly basis… with it’s own bureaucracy. So consider this a non-novel approach, but an emphasis and encouragement of focus, with a UX/UI that is more engaging and informative.

I also bring forward (in the UI) fatigue from consecutive working days. Depending on the stower’s tenure, they will become mentally and physically fatigued. Effort made to alternate tasks or cross train departments does not negate the impact of fatigue, but is welcome. This is physical and largely mental health related. I need to do extensive research in this area to provide predictions, so it is constrained to physical aspects.

---

The Bits

The simulation starts at 8am (1 hour after stowers have been stowing). You can scrub through time with the timeline and watch the days fill up. There is a multiplier, so you can click it to 1x and watch the day as it happens.

The stow target is set to 105% for 22.1k units. There are 15 stowers.

---

Engineered Labor Standards (ELS)

My concept builds upon Engineered Labor Standards, or ELS, which are a way to ask a simple question: **how much work should this task reasonably take? **Not every unit is equal. A small, easy item is not the same as a heavy item, a long walk, or a bin that takes extra handling. ELS helps account for those differences, so the dashboard is not only counting how many items someone stowed. It is also trying to show how difficult the work was, and whether two people with different unit counts may have done a similar amount of real work.

What the Numbers Mean

The dashboard uses the same four performance lenses as the per-session stow model. Each metric divides output by a different slice of time, so each one answers a different question. None of them is “the truth” on its own. Read together, they separate speed while working, freight difficulty, how much of the shift was actually spent stowing, and the traditional units per paid hour view.

1. UPH (macro / shift UPH)

The number of items/units stowed for every hour on the clock during the shift. Walking, waiting, breaks, and congestion all count in the denominator, so this metric combines individual pace with everything else that happened that day.

2. Session rate (scan-to-scan)

The rate of stowing only during active stow time, measured from the first scan to the last scan in a continuous work stretch. Time outside that active window, such as long gaps before the next scan, is excluded from the denominator. This is closer to “how fast work occurred while stowing was actually happening.”

For a single continuous session, active session hours are the elapsed time from first scan to last scan, converted to hours.

3. Weighted session rate

This is similar to session rate, but harder or heavier items count for more than easier small items. Each item class receives a weight from engineered standards. The weighted “effective units” are then divided by active session hours, so cherry-picking only small items does not automatically dominate the chart.

Here, $n_s,\, n_m,\, n_h$ are the counts of small, medium, and heavy units stowed, and $W_s,\, W_m,\, W_h$ are the matching effort multipliers.

4. Flow efficiency

Of the time a worker was expected to be available for production, defined as paid shift time minus authorised breaks, what fraction was spent inside active stow sessions? A high session rate with low flow efficiency often indicates fast bursts with substantial idle or off-task time. A moderate session rate with high flow efficiency may indicate steady work on more difficult freight.

The second line expresses the same ratio as a percentage, using a number between 0 and 100 instead of 0 and 1.

Worked Example: Associate A vs. Associate B

Both associates work within the same shift envelope: 9.5 paid hours and 0.5 hours of authorised paid breaks. This leaves 9 hours as expected production time for flow efficiency. Associate A focuses on small, easier units; Associate B stays on heavy freight.

	Associate A	Associate B
Units stowed	2,850 (all small)	1,140 (all heavy)
Active session hours	6	8
Effort weight per unit	1.0	3.0

1. UPH (units ÷ full paid shift, 9.5 hours)

2. Session rate (units ÷ active session hours only)

3. Weighted session rate (effective units ÷ active session hours; B’s items count triple)

4. Flow efficiency (active session hours ÷ 9 expected hours)

Raw UPH makes A look like the stronger stower by a wide margin. Session rate still favours A because every unit is counted the same. Weighted session rate nearly closes the gap, with B within about 10% of A. Flow efficiency shows that B spent most of the expected window actually stowing, while A left roughly a third of that window outside active sessions. In this reading, A appears to be a fast sprinter with gaps, while B appears to be a steady pacer on harder work.

Problem Solver at session end

In about 15% of sessions, a stower finishes with a Problem Solver interaction: an item was not scannable, not assigned to the cart, needed an expiry date, or could not be stowed on the floor (HRV, hazmat, and similar). The session timeline shows an amber hatched zone at the tail — similar to a break overlay, but with PS item markers inside it.

The stower typically waits 20–60 seconds (dashed “Waiting” strip) before Problem Solver touches the item. Then either:

• Handoff — the item leaves with Problem Solver and the session ends, or
• Resolve and stow — Problem Solver fixes the issue and the stower scans one or two final units (amber-ringed bars) before the session ends.

Toggle Problem solve on the session detail view to show or hide the zone.

How to Read This Dashboard

The synthetic shift in the live dashboard is calibrated to finish above plan — about 106% of the 22,137 day target at end of shift — reflecting a regular-pace stow day rather than an under-delivered one.

If someone’s UPH looks low but weighted session rate and flow efficiency look strong, they may be working difficult freight, dealing with congestion, or receiving a fairer evaluation than raw units per paid hour would provide. If session rate is high but flow efficiency is low, they may be stowing quickly while active but not remaining in task for much of the shift. The timelines and side-by-side metrics should be used to determine which explanation best fits the observed pattern.