Insights

The Marketing Intelligence Latency Metric

A useful signal can reach marketing too late to change the decision it was meant to inform. Measure that delay from the source event to a usable, attributable handoff.

A useful signal can be accurate, well sourced, and worthless by the time it reaches marketing.

A product promise changes on Monday. The launch page ships on Tuesday with the old claim. Sales hears the same objection all week, but the pattern reaches the content team in Friday's digest. A churn interview explains the expectation gap after the next campaign has already gone live.

The issue is not whether the company collected information. It is how long the information took to become usable where a marketing decision was being made. Northstar calls that delay marketing intelligence latency.

Define the Metric Precisely

Marketing intelligence latency is the elapsed time between a commercially relevant event and the moment marketing receives the signal in a usable, attributable form.

Marketing intelligence latency = marketing_ready_at - occurred_at

This is a Northstar operating metric, not an established industry standard. The definition is deliberately narrow. The clock begins at the source event and stops when marketing has the signal, its source, its meaning, and a destination or owner.

That stopping point matters. If the metric runs until a page, campaign, or brief changes, it blends information transport with editorial judgment, review time, production capacity, and release cadence. Track that longer interval separately as signal-to-applied time.

Use Five Timestamps

A single received date tells the team that something was slow. Five timestamps show where the time went.

occurred_at marks the source event: the call ended, the churn interview closed, the support pattern was recorded, or the product decision was approved.

captured_at marks when the source became available to the system. classified_at marks when the system recognized the item as relevant to marketing. marketing_ready_at marks delivery with attribution and enough context to act. applied_at is optional and records when the work changed.

These stamps divide the trip into capture, recognition, routing, and action. The distinction prevents a weekly digest from hiding behind fast transcription or a slow approval cycle from being blamed on the information route.

Latency and Freshness Are Different

Google Analytics defines data freshness as how recently data has been collected, processed, and reported. That is useful language for an analytics product. Company intelligence needs a second distinction because some signals have not completed the trip at all.

For a delivered signal, latency is marketing_ready_at - occurred_at. For an unresolved or unrouted signal, current age is now - occurred_at.

A dashboard should show both. Completed latency explains the route that worked. Backlog age exposes the items still waiting in a queue, a folder, or a weekly meeting.

Measure the Distribution, Not the Average

An average can make the route look healthy while a smaller group of signals waits far too long. Ten easy call transcripts may arrive immediately while two product changes sit without an owner. The average improves even though the risk has not.

Use p50 and p90. The median shows the typical completed trip. P90 shows the point at which nine out of ten completed trips were at least that fast, leaving the slowest tenth beyond it.

Google's Site Reliability Engineering guidance makes the same statistical point for service latency: averages can obscure a long tail, while percentiles show the typical case and slower edge of the distribution. See its guidance on service level indicators and percentiles. Marketing intelligence is a different system, but the measurement principle transfers cleanly.

Do not chase one perfect company-wide number. Watch whether the median is improving, whether the tail is widening, and which source produces the slowest completed trips.

Segment by Signal Type

Sales objections, churn evidence, product changes, support patterns, and leadership decisions do not share the same clock or consequence. Combining them into one company average erases the operating difference.

A changed product claim may need to reach an active launch immediately. A repeated objection may be useful in a daily or weekly content review. A broad change in customer language may be meaningful only after enough evidence accumulates to show a pattern.

Report latency by source and signal class. Keep the underlying record available so the team can inspect why a group moved slowly instead of debating what one blended score means.

Pair Latency With Coverage

A fast latency score can be false comfort. If the route captures only the easy signals, every completed trip may look quick while the most valuable evidence remains invisible.

Track the share of eligible source events that were captured, classified, and delivered. Then show backlog age for events that have not reached the next state. Latency describes speed among completed trips. Coverage describes how much of the intended signal set made the trip.

This is the same reason a collection of recordings is not yet a marketing system. The guide to starting AEO in sales calls explains how to identify the commercial signal, while the internal answer inventory shows how to preserve its source and turn it into usable evidence.

Find Which Stage Holds the Delay

Consider a call that ends Monday at 10:00 a.m. The transcript is available at 10:30. The signal is classified Tuesday at 9:00 a.m. A weekly digest reaches marketing Friday at 9:00 a.m.

The total latency is 95 hours. Transcription used 30 minutes. Classification waited 22.5 hours. Routing into the weekly digest used the remaining 72 hours. Buying faster transcription would barely change the outcome. The routing cadence is the constraint.

Break p90 latency into the same stages. The largest segment shows where a change can matter: source access, classification cadence, ownership, or delivery schedule. This is more useful than demanding that every team simply move faster.

Set Targets From Decision Half-Life

There is no responsible universal benchmark for this metric. A useful target depends on how quickly a signal loses decision value.

Ask when the next decision will be made and when the evidence becomes too old to change it. A product change tied to tomorrow's launch has a short half-life. A pattern used in a quarterly positioning review has more time. The service target should arrive before the decision window closes, with enough margin for marketing to evaluate and apply it.

Define the target by signal class, record it beside the route, and review misses at p90. This keeps the objective tied to work the company can change instead of an arbitrary speed score.

Build the Metric Into the Working Loop

Start with the sources already central to revenue work: sales calls, churn interviews, support records, product decisions, and leadership priorities. Define eligible events, add the timestamps, and keep the source attached through classification and delivery.

Northstar's Listening Engine stamps capture, classification, and routing. The Glass Dashboard shows p50, p90, coverage, backlog age, and the records behind each measure. This closes the core system. No additional strategic layer is required to see whether useful company intelligence is reaching marketing in time.

The metric belongs next to the work, not in a quarterly diagnostic. When latency rises, inspect the stage. When coverage drops, inspect the source. When backlog age crosses the decision window, route the item before it expires.

Marketing does not need every signal instantly. It needs each important signal before the decision it can still improve.

Frequently asked questions

What is marketing intelligence latency?

Marketing intelligence latency is the elapsed time between a commercially relevant event and the moment marketing receives the resulting signal in a usable, attributable form. Northstar defines it as marketing_ready_at minus occurred_at.

When does the marketing intelligence latency clock start?

Start the clock when the source event occurs. That may be when a sales call ends, a churn interview is completed, a product decision is approved, or a support pattern is recorded.

When should the latency clock stop?

Stop it when marketing receives the signal with its source, meaning, and destination or owner. Track the later time from handoff to a changed page, brief, or campaign as a separate signal-to-applied measure.

Why use p50 and p90 instead of an average?

The median shows the typical delivery time, while p90 exposes the slow tail. A simple average can look acceptable even when a consequential share of signals arrives much later.

What is the difference between latency and freshness?

Latency measures the completed trip from the source event to a usable marketing handoff. Freshness or backlog age measures how old a signal is now when that trip has not been completed.

What if a signal never reaches marketing?

A missing signal has no completed latency value, so latency must be paired with capture and routing coverage. Otherwise the metric can look fast because it excludes everything the system failed to deliver.

How does Northstar Stack measure marketing intelligence latency?

The Listening Engine records event, capture, classification, and handoff timestamps. The Glass Dashboard shows p50, p90, source coverage, backlog age, and the stage where delay is accumulating.