How to Measure Intelligence Freshness in Competitor Monitoring

Intelligence freshness is the share of the competitor information you track that has been verified inside a time window you fixed before you started counting. It is not the data freshness an engineering team measures, which is a lag between an event and its arrival in a system. It answers one question, and only one: can what we hold about our competitors still be relied on today?

The only common metric that falls while you work

Almost every number a competitive intelligence function reports is cumulative. Alerts delivered, competitors tracked, battlecards published, briefings circulated: each one grows with effort, and none of them can decline while the team is doing its job. Freshness is the exception. It drops on its own, with nobody doing anything wrong, because the market it describes carries on moving after the work is filed. That property is the entire reason to track it.

It also means the metric behaves differently from the ones around it. A quarter of heavy output can lower the score rather than raise it, because everything published becomes something that now needs maintaining. Programmes that have never measured freshness are often surprised by this the first time, and the surprise is the finding.

Not volume, not coverage, not quality

It is not a volume metric. Publishing more does not improve it, and usually does the opposite for a period.

It is not coverage. Coverage asks whether intelligence reached a deal before the decision. Freshness asks whether what reached it was still true. Those fail independently, and a function can be strong at one while being useless at the other.

It is not a quality score. The metric records that somebody checked an item, not that the item was worth holding or that the check was any good. A tracker full of claims nobody needs can hold a perfect score indefinitely, which is the failure mode worth understanding before you adopt the number at all.

The arithmetic is trivial. The definitions are not, and the denominator is where this measurement is normally won or lost.

Why the denominator is not everything you track

The obvious version of this formula divides by every row in the tracker. It produces a number, and the number is misleading, because tracked items do not decay at the same speed. A competitor’s incorporation date does not need checking this month. Their pricing page might need checking this week. Dividing by the whole tracker punishes a team for holding durable facts and rewards one that holds none.

So the denominator is the set of items whose review cadence came due inside the window. Every row carries a cadence; the rows whose cadence expired during the period are the ones you were supposed to check, and they are the only fair thing to divide by.

What counts as verified

A verification is a check that produced a dated outcome, recorded as one of two values: confirmed unchanged, or updated. Both count in the numerator. An item somebody looked at and left without recording anything does not, because there is no way afterwards to tell it apart from one nobody opened.

The formula applied to a real tracker

One tracker can produce almost any freshness score depending on the cutoff chosen, which means a figure quoted without its window is not a figure at all. Pick the window per item type rather than for the whole tracker, because the honest cadence for a pricing page and the honest cadence for a funding history are months apart.

Once cadences are set, the single number is a weighted roll-up rather than a fact about the whole tracker, and it should always be reported with the split underneath. A function at 80% overall can be at 40% on pricing, which is the only part of that sentence anyone needs to act on.

No industry benchmark for this metric has been published. Not a low one, not a contested one: the figure does not exist, because measuring it would require access to the internal trackers of a representative sample of competitive intelligence functions, and no organisation has collected that. Any percentage presented to you as the standard for intelligence freshness was made up.

What does exist is evidence about how common the underlying problem is, and it is worth looking at precisely because it is so often mistaken for a target.

Why 13.0% is evidence and not a target

That figure counts reviewers, not items. It tells you what share of one platform’s customers were annoyed enough by out-of-date content to write it down in a public review. The share of content that was actually stale is a different quantity, it was never measured, and it is almost certainly higher, because the gap between being mildly inconvenienced and posting about it is wide. Reading 13.0% as a ceiling gets the direction of the error backwards.

Used properly it settles a different argument: whether this is a real problem worth spending a metric on. Every one of those reviewers was a paying customer of a product sold on the promise of currency, and staleness was still the sharpest thing they had to say about it. That is the case for measuring freshness on your own material whatever you have bought, and it is the extent of what the number supports.

What to compare against instead

Your own previous cycle, and nothing else, until you have four of them. The first measurement is a baseline rather than a grade, and the direction over the following quarter carries all of the information. A function that moves from 48% to 61% has learned something real; a function comparing 48% to an invented industry average has learned nothing.

The one target worth setting in advance is not a percentage at all. It is a rule: nothing in the battlecard set is older than its own cadence. That is binary, it is arguable in a room, and unlike a percentage it cannot be satisfied by improving the average on material nobody uses.

Both halves of the formula come out of the same artefact, and most functions do not have it yet in a usable form. A competitor tracker built for reading is not a tracker you can measure: a document with a section per competitor holds the same information and cannot produce either number.

What the tracker has to carry

One row per claim, not one row per competitor. Each row needs an owner, a review cadence, a last-verified date and the verification outcome. Those four columns are the whole requirement, and our competitor tracking spreadsheet template already carries them, so the metric is calculable from the first cycle rather than after a rebuild.

Where the verification itself happens

Every row points at a source that has to be re-read to close it. For the fastest-decaying rows that means a rival’s published pricing, checked against what you recorded last time. Where a page has changed and you need to know exactly what moved, archived snapshots of the old site turn a vague sense that something is different into a dated diff you can put in the tracker.

Exporting it, and connecting it

The calculation itself belongs in a spreadsheet. Export the tracker to CSV, filter to rows whose cadence expired inside the window, and count the ones with a verification date in that window. Two filters and a division, which is deliberately small: a freshness number that takes an afternoon to produce will be produced once.

A competitive intelligence platform changes which rows a person has to close by hand rather than whether the metric works. For the item types a platform watches, the last-verified date is maintained as a by-product of the watching, so the denominator shrinks to the rows that genuinely need judgement. The columns, the cadences and the formula stay exactly as they are.

The order below matters more than it looks. Most attempts at this metric fail at step four, by measuring before finding out which rows are unverifiable in principle.

1. 1List what you actually hold, claim by claim. One row per assertion a rep could repeat in a meeting. “Competitor A” is not a row. “Competitor A’s entry tier includes SSO” is.
2. 2Give every row a cadence and a named owner. Unowned rows do not get verified and should not sit in the denominator pretending otherwise.
3. 3Add the two fields the metric needs: a last-verified date and an outcome of confirmed unchanged or updated.
4. 4Run one full cycle without calculating anything. This is the step that gets skipped. It surfaces the rows nobody can close, usually because the source was never recorded or the claim is too vague to confirm, and those rows have to be fixed or retired before a number means anything.
5. 5Calculate on the second cycle. Rows due in the window, rows verified in the window, divide.
6. 6Report the percentage next to the list of what failed, grouped by competitor and item type. The list is what gets acted on; the percentage is what gets remembered.

What you end up holding is a one-page freshness report: a figure, the window it was measured at, the split by item type, and the named rows that went unverified. It takes about twenty minutes a month once the tracker carries the right columns, and it is the smallest artefact that will change anybody’s behaviour.

Rubber stamps look exactly like verifications

A cycle in which every check returned confirmed unchanged is not evidence of a stable market. It is more often evidence that the checks were cursory. Recording the outcome rather than only the date is what makes this visible: a period where nothing anywhere changed deserves suspicion, not congratulation.

The denominator is editable

Anyone can improve the score by archiving the rows that keep failing. That is sometimes the right call and sometimes the metric being quietly managed, and the two are indistinguishable in the percentage. Keep a visible count of rows archived per cycle beside the score, and the question answers itself.

It says nothing about what you never tracked

The metric is bounded by the tracker. A competitor who entered your market last month, and who nobody has added, cannot lower your freshness score by a single point. This is the limitation to state out loud whenever the number is presented well, because a high score is most reassuring exactly when the tracker is most incomplete.

Depth of verification varies and is not captured

Re-reading a pricing page and re-running a full teardown both close a row. One takes four minutes and the other takes a morning, and the metric treats them identically. That is acceptable for an operational number as long as nobody starts optimising it, which is another argument for reporting the failed rows alongside the total.

The uncomfortable thing about this metric is that most of the work it measures produces nothing. In a typical cycle the large majority of checks come back confirmed unchanged, and all of the value sits in the few rows that moved. A person spends the afternoon re-reading pages that are identical to last month in order to find the two that are not, and that ratio does not improve with practice.

Being late has a specific cost here rather than a general one. The window between a competitor changing something and your tracker knowing about it is the window in which your team is confidently wrong in front of buyers, and the reps carrying the worst information are the ones working the most contested deals.

This is the job competitive intelligence platforms exist to absorb: continuous re-checking of the item types that decay fastest, so a last-verified date is a by-product of the watching rather than the output of somebody’s afternoon. Flares tracks competitor pricing, product and messaging on that basis and dates what it finds, which collapses the denominator to the rows that genuinely need a human.

What no platform decides for you is what belongs in the tracker, what a change means, or which rows matter enough to argue about. Those are the judgements the metric exists to protect, and they stay yours. Automation buys back the re-reading, not the thinking.