Measuring and Improving On-Time Delivery: Definition, Metric, Levers

Why is “almost always on time” not enough?

On-time delivery is one of the qualities customers remember longest. A late delivery shifts the customer’s own planning, and whoever has been disappointed twice builds in their own buffer next time or asks another provider. Yet on-time delivery is rarely measured in small companies. People have an impression, and it is usually better than reality.

That is not due to whitewashing but to memory. Punctual orders go unnoticed; late ones come with reasons that sound like exceptions: the supplier, the illness, the special request. This leads to the guiding question of this guide: How does a small company measure its on-time delivery so that the figure reveals causes, and how does it improve it permanently?

To answer it, we first set a clear definition. Then we introduce a thinking model, Little’s law, which explains why full order books shift dates, review findings from production research, and weigh whether to promise short or safe dates. At the end there are a measurement process, a list of causes, and the levers.

A date that is not measured is an intention, not a promise.

What exactly is on-time delivery? A definition that holds

On-time delivery is the share of orders or deliveries that reach the customer complete on the promised date. In logistics, the same figure is often called OTIF, short for on time in full. Three specifications make the definition robust.

1. Measure against the promise. The yardstick is the date the company confirmed, not the date the customer wanted. Whoever mixes the two measures their sales conversations instead of their processes.
2. On time means within a defined window. For deliveries this is often the promised day; for installations perhaps the promised week. The window is set once and not negotiated case by case.
3. Complete means everything that was ordered. A partial delivery on the date is not on time, even if it feels that way.

Besides the rate, a second figure is needed: lateness, the actual date minus the promised date, in days. A negative value means early. Only lateness shows whether a missed date was missed by one day or by three weeks. Whoever also measures on-time delivery for their suppliers uses the same definition, as the guide to evaluating suppliers describes.

The thinking model: Little’s law, average and spread

In 1961, the mathematician John Little proved a relationship that holds in every queue, whether at the supermarket, in the workshop, or in the pile of quotes (Little 1961). Little’s law states: average throughput time equals the average stock of open orders divided by throughput. If a workshop has an average of 30 orders on hand and finishes 10 a week, an order takes three weeks on average. If 15 more orders are added without throughput rising, it takes four and a half weeks.

Something uncomfortable follows. Whoever starts more orders at once to be faster becomes slower, as long as capacity stays the same. Every additional open order lengthens the waiting time of all others. The article on capacity planning shows the other side of the same logic: at 80 percent utilization, a new task waits on average four times as long as its processing takes. At 95 percent, nineteen times as long.

For on-time delivery, a second distinction comes in. Production research breaks delivery performance into two parts (Thürer, Stevenson, Land, and Fredendall 2019): the average, that is, how late orders are on average, and the spread, that is, how differently late they are. An image from target shooting makes this tangible. If all hits are tightly grouped but left of center, the spread is right and the average is not. If they are scattered around the center, the average is right and the spread is not. Each error needs a different correction.

What research shows about controlling delivery dates

Thürer, Stevenson, Land, and Fredendall (2019) studied how three control tasks interact that determine delivery dates in companies with many different orders. Due date setting determines which date is promised to the customer. Order release decides when an accepted order actually goes into production. Capacity adjustment changes the available working time, for example through overtime or reallocation. This was studied in a simulation of a job shop.

The result: each of the three tasks mainly affects one particular objective. Due date setting and capacity adjustment complement each other because they improve different parts of delivery performance. Limited order release mainly reduces the amount of open work in the shop and can improve throughput when capacity cannot be adjusted. How much a company should rely on which task depends, according to the authors, on which objective matters most to it.

The study is a simulation, not an observation of real companies. Its value lies in separating the levers that are usually mixed up in daily work. A company that only works overtime treats the average and leaves the spread in place. A company that only promises more generously shifts the average without changing anything in its processes. That due date setting and order release can be combined even without software is shown by Thürer and Stevenson (2016) with a simple card-based system that controls both tasks.

Where the disruptions come from is a second question. Heim, Peng, and Jayanthi (2014) summarize earlier plant-level studies as finding that mainly factors of dynamic complexity impaired delivery performance, such as fluctuating demand and short-notice changes. The mere variety of products and parts was less often found to be significant there. The authors test these factors with more detailed product-level data. For small companies, the direction is worth considering: short-notice changes often harm delivery dates more than a broad range of products.

Promise short or promise safe? Two perspectives

The first view says: short delivery times sell. Whoever promises three weeks while competitors say five wins the order. Better to promise tight and then do everything to make it. When speed really matters more than price is discussed in the article on delivery time as a sales argument.

The second view says: reliability sells longer. A customer promised five weeks and supplied after five weeks plans with the company. A customer who hears three weeks and is supplied after four and a half plans around it. Add the planning fallacy described by Kahneman and Tversky (1979): people regularly underestimate when a task will be finished because they ignore interruptions and competing tasks.

I consider the second view the more robust one, with one condition. The promise should come from your own data, not from hope or fear. Whoever knows the actual throughput times of recent months can name a date they will keep in nine out of ten cases. And whoever wants to promise shorter dates must first reduce throughput time, not the promise.

How to make on-time delivery measurable in four weeks

1. Set the definition

   Write down what counts as the promised date, how large the window is, and what complete means. These three sentences apply to all orders.

2. Record two dates per order

   For each order, the promised and the actual date are recorded, plus a checkmark for complete. This takes half a minute per order.

3. Analyze lateness as a distribution

   Once a week, look at the rate and the distribution of lateness: how many were on time, how many one day late, how many more than a week late?

4. Tag causes

   Each late order gets a cause code from a short list. After four weeks, the list shows where delays cluster.

5. Add it to the dashboard

   On-time delivery and average lateness become part of the one-page KPI dashboard, with a threshold at which action is taken.

Which levers improve on-time delivery?

Five levers follow from the measurement and the list of causes. Which should come first is shown by the distinction between average and spread.

• Promise from data: dates are promised according to the nine-out-of-ten rule and extended automatically when utilization is high. Affects the average.
• Limit release: new orders go into work only when the number of open orders is below a limit. According to Little’s law, throughput time falls. Affects the spread.
• Work in order of due date: the sequence follows the promised date, not the customer who asks most loudly. Affects the spread.
• Adjust capacity at the bottleneck: additional hours pay off only where work piles up. How to find this point is described in the article on Goldratt’s theory of constraints. Affects the average.
• Remove external causes: supplier dates are measured too, and customer changes after the order confirmation get a new date. Affects both.

One last lever concerns not the date but how a missed one is handled. Whoever says early that it will be later and names a new, reliable date loses less trust than whoever stays silent until delivery day. How to deliver such news is described in the guide to delivering bad news. In on-time delivery, a newly agreed date still counts as missed; otherwise the figure loses its meaning.

Conclusion: how does a company measure and improve on-time delivery?

We began by asking
a small company measures its on-time delivery so that the figure reveals causes, and how it improves it permanently. The answer begins with a clear definition: measured against the promise, with a fixed window, only complete deliveries count. Lateness as a distribution comes on top, because only it separates a problem of the average from a problem of the spread.

As Little’s law and research on due date control show, the levers differ. Realistic promises and capacity at the bottleneck help against a poor average; limited release and sequencing by due date help against a wide spread. The sober observation is: on-time delivery does not come from effort at the end of an order but from the decision at the beginning about how much to take on at once.

The next step

1. Write down the definition of on-time delivery for your company in three sentences: promised date, window, complete.
2. From Monday, record the promised and actual date for every order.
3. Count how many orders are open at once today and divide that number by the orders finished per week.
4. Compare the result with the delivery times you currently promise.
5. After four weeks, evaluate the first cause codes and decide whether you have a problem of the average or of the spread.