How to Calculate Competitive Intelligence ROI

Most competitive metrics count something that happened. This one compares a cost you know exactly against a benefit nobody can isolate, which makes it a business case wearing the notation of a measurement.

That is not a reason to refuse it. Budget holders ask the question, and a function that answers with “it cannot be measured” loses the argument to one that answers with a number. The workable position is a figure whose every assumption is stated, presented as a range, next to the two operational metrics that genuinely are countable.

The three terms, and which one is the problem

Programme cost is exact, if you count it properly. Contested revenue is countable from the pipeline. Margin is a rate finance already publishes. The fourth term, the share of that margin the programme can claim, has never been measured by anyone and cannot be, and it determines the answer more than the other three combined.

The 348% is real arithmetic on plausible figures and it is still not a fact about the world, because one of the four inputs was chosen rather than counted. The next four sections are about making each term defensible, starting with the one people get wrong first.

The denominator is the only term you fully control, and understating it costs you the room faster than anything else in the calculation, because finance can check it. It is also where most published returns are weakest: counting the software and nothing else produces a large multiple on a small base and an obvious question.

1. 1Fully loaded people cost. Salary plus employer costs plus benefits for everyone doing the work, at their real fraction of time. A product marketer spending 40% on competitive work is 0.4 of a loaded salary, not a rounding error.
2. 2Software and data. The competitive platform, research subscriptions, review-site data, any paid market sizing.
3. 3External spend. Commissioned research, win/loss interviewing when it is outsourced, analyst time.
4. 4Consumption time. The contested part. Reps reading briefings and attending enablement are a real cost of the programme, and excluding it entirely is as distorting as loading the whole sales org onto it. A defensible approach is to count the scheduled time only, and say that is what you did.

Revenue overstates what a won deal is worth by everything it costs to deliver. In software that gap is modest and in services it is enormous, and in both cases a finance team applies the margin rate before taking any return seriously. Applying it yourself is the difference between a number that gets accepted and one that gets corrected in the meeting.

Use the company’s published gross margin rather than a segment figure unless you can source the segment one, and say which you used. Where contested deals carry systematically different economics, heavier discounting being the common case, note it rather than adjusting for it: an unexplained adjustment reads worse than an acknowledged imprecision.

Here is where almost every published competitive intelligence ROI figure quietly stops being a calculation. Contested revenue is not programme revenue. Some of those deals would have been won with no competitive support at all, and the share that would not is the number nobody has.

Skipping the step means implicitly setting the attribution rate to 100%, which is the strongest possible claim made by omission. It is also why the resulting figures are so large and so uniform.

How to choose a rate you can argue for

Start from something countable: the share of contested won deals that received specific material before the decision. Then discount it, hard, because reaching a deal is not the same as changing it. Teams that do this carefully tend to land somewhere in the low tens of percent, and the important part is not the value but that you can describe how you got there.

What would be needed to do better

A holdout: contested deals deliberately left unsupported, compared against supported ones. Marketing eventually made real progress on attribution this way. No revenue leader will agree to it for competitive support, and any figure implying that kind of rigour without it is overstating what was done. Saying so plainly is more persuasive than pretending otherwise.

Since one input is chosen rather than measured, the output is a function of that choice and should be presented as one. This costs two extra rows and converts the most attackable part of the case into the most credible.

Every figure above uses the same $214,000 cost, the same $8.2M contested revenue and the same 78% margin. The only thing that changed is the term nobody can measure, and it moved the headline by a factor of thirteen. A single confident percentage hides that entirely.

Percentages for this metric circulate widely, often precise, often without a citation attached to the individual claim. They are not necessarily wrong. They are usually uncheckable, which for a number you intend to repeat in front of your own executives amounts to the same risk.

Three questions settle it, and they take about a minute:

1. 1What was in the cost base? If only the software licence, the multiple is inflated by however much the people cost.
2. 2What attribution rate was applied? If the answer is none, the figure assumes the programme caused every contested win.
3. 3Over what sample, and when? A return from one company in one good year is an anecdote with a percent sign on it.

A figure that answers all three is worth citing with its conditions attached. A figure that answers none is worth leaving where you found it, and this page does not publish one for the same reason.

A programme in its first year should not be defending a return. It should be demonstrating that the work reaches deals and that what it holds is current. A quarter is long enough to produce both, and neither rests on an assumption anybody can dispute.

Competitive deal coverage rising from 30% to 65% is a fact. A freshness score holding above target through a quarter of heavy publishing is a fact. Presented together they make the ROI conversation easier when it arrives, because by then the attribution argument is the only thing left to have, and you will have already shown you do not inflate the things that can be checked.

Assembling the inputs from systems you already pay for

Cost comes from finance, contested revenue from a CRM export, margin from the last reported accounts. Build it once in a spreadsheet with the attribution rate in its own clearly labelled cell, so anybody can change that cell and watch the answer move. That sheet is a better artefact than the percentage: it survives the meeting, and it makes the sensitivity obvious without an argument.

Wire competitive monitoring into the CRM and the contested-revenue half stops being a reconstruction, because the tags then carry dated evidence of what each competitor was doing. That does not make the attribution rate measurable. It does mean the inputs you can count are counted properly, which is where this calculation usually leaks.

The specific decay on this metric is in the cost base, and it runs the wrong way. Most of what a competitive programme actually costs is people re-checking things: prices that did not change, pages that were not rewritten, releases that did not ship. That work produces no output and it scales with the number of competitors rather than with the value of what is found.

Being late costs twice over. The hours are spent whether or not anything moved, and the one change that did happen is found a month after the deal it would have affected.

Reducing the denominator is what the competitive intelligence software category is for, and it is a more honest lever on this metric than arguing the attribution rate upward. Flares carries that re-checking across a competitor’s prices, packaging and public claims, which takes the largest recurring line out of the cost base and leaves the hours for work that produces something.

It does not make the return provable. Nothing does, and a page claiming otherwise would be doing the thing this one spends four sections warning against. What changes is that the two terms you can count become cheaper and more accurate, and the one you cannot stays visible, labelled and open to argument.