Glossary

A complete library of terms, metrics, and recommendations for data & analytics

Decision Latency

« Back to Glossary Index

Decision latency is the time between a signal appearing in the business and an action being taken on it. It covers four stages: detecting the signal, understanding the problem, evaluating alternatives and executing. A report can be accurate and still be useless if it arrives after the window to act has closed.

This guide explains how decision latency works with a retail example, how to measure it, and how it relates to decision coverage and closed-loop value. It draws on the conversation between Anirban Nandi of Albertsons Companies and Shashank Garg in Why Retail’s Problem Was Never the Data.

What Is Decision Latency?

Decision latency is the elapsed time from the moment a signal appears in the business to the moment someone acts on it. A signal can be a stockout, a sales decline, a pricing change or a customer behavior shift. The action can be a replenishment order, a markdown, a campaign change or a transfer of inventory.

It differs from data latency, which is only how old the data is when it reaches a report. Decision latency includes data latency but also the time taken to understand the problem, weigh the options and get the decision carried out.

How Decision Latency Works in Retail

Anirban Nandi, who leads AI and Data at Albertsons Companies, uses fresh produce as the example. A retailer has excess strawberries in several stores, with limited shelf life remaining. A traditional reporting process might flag the problem the next morning. By then, the options to transfer inventory, adjust replenishment or run a targeted markdown have narrowed significantly. The data was accurate and the report was correct, but the opportunity expired.

In the episode he puts it this way: the value of data is not just in knowing the right answer, it is in knowing while there is still time to act on it.

The Four Stages of Decision Latency

Detect the signal. How long does it take for the business to see that something changed?

Understand the problem. How long does it take to work out what the change means and why it happened?

Evaluate alternatives. How long does it take to compare the options, with their costs and trade-offs?

Execute. How long does it take for the decision to be approved and carried out?

Nandi notes that the time is lost in different places: sometimes in data latency, sometimes in analysis, and often in coordination between teams. The first job is to find out which stage is slowest.

How to Measure Decision Latency

Shashank Garg’s advice is to start with one decision and measure the time from signal to action. His example: if a stockout signal takes four days to become a replenishment order, four days is the number, not how fresh the dashboard is. Four days may be acceptable for replenishment. The point is that the speed of the intelligence should match the economics of the decision, because not everything needs to stream and not everything can wait for a weekly report.

A practical way to begin: pick one recurring decision that costs margin every week, record when the signal first appeared and when the action was taken, split the gap into the four stages, and find the longest one. Then give the decision an owner and one governed definition of the metric it relies on.

Decision Latency, Decision Coverage and Closed-Loop Value

Garg argues that most organizations measure inputs rather than outcomes: a modernized platform, a data quality score of 99.8 percent, adoption at 80 percent. All of that can be true, and none of it shows that a single decision got better. He offers three outcome measures instead.

Decision latency is the time from signal to action.

Decision coverage is the share of recurring weekly decisions that a decision intelligence platform actually supports, including external data rather than internal transactions alone.

Closed-loop value asks whether the action delivered the lift the model predicted, and whether the model learned from the gap.

Decision Latency vs. Decision Velocity

The two terms are close. Decision velocity usually describes how fast decisions are made across an organization. Decision latency names the delay on one decision, which makes it easier to measure because it has a clear start, the signal, and a clear end, the action.

How to Reduce Decision Latency

Agree one definition of each metric. When finance, merchandising and digital report different sales numbers, time goes into debating the number instead of acting on it. A governed semantic layer defines each metric once, with its logic, lineage and owner.

Give each decision an owner. Garg points to incentives: merchandising is paid on margin, supply chain on service level and stores on labor cost, so agreeing takes longer. Clear decision ownership shortens the coordination stage.

Automate carefully. Nandi’s guidance on AI agents is to start with recommendations, validate their quality, add human approvals, and only then automate well-understood, lower-risk decisions. Automating a poorly understood process creates confusion at machine speed.

Frequently Asked Questions

Decision latency is the elapsed time between a signal appearing in the business and an action being taken on it. It covers four stages: detecting the signal, understanding the problem, evaluating alternatives and executing.

Data latency is how old the data is when it reaches a report. Decision latency includes that delay but also the time spent understanding the problem, weighing options and carrying out the decision.

It depends on the decision. Shashank Garg’s replenishment example suggests that days can be acceptable for some decisions, while a perishable item such as fresh produce may need action within hours. Set the target from what a delay costs.

Anirban Nandi points to data delays, slow analysis and coordination between teams. Garg adds incentives: when each team is measured on a different number, agreeing on what to do takes longer.

Pick one recurring decision, record when the signal first appeared and when the action was taken, and split the gap into detecting, understanding, evaluating and executing. The longest stage shows where to focus.

Decision coverage is the share of recurring weekly decisions that a business supports with its data platform, semantic layer or decision intelligence system, including external data and not only internal transactions. It is assessed through a periodic audit.

Closed-loop value asks whether an action taken on a recommendation delivered the lift the model predicted, and whether the model learned from any gap. It moves measurement from whether an insight was delivered to whether the outcome improved.

It can, but Nandi warns that automating a poorly understood business process creates confusion at machine speed. Agents need trusted business context, explicit decision authority and continuous verification of outcomes first.

We'd Love to Hear from You!

Open Popup