This document explains how the method works, what it rests on, how it is tested and where it stops working. Not with the figures themselves, but with the reasoning behind them.
PAR stands for Planned Agency Resource. One PAR is a fixed unit of agency work, roughly the size of two agency days. It is a measure of size, not a price tag.
The reason for counting in work rather than in euros is simple: the same logo takes roughly the same amount of work whichever agency makes it. The price varies widely, the workload far less. Counting in work means comparing like with like.
What PAR expressly is not: a billing unit. The method does not ask agencies to quote in hours and does not pay them on that basis. How an agency organises and prices its work remains the agency's decision. PAR is a yardstick that brings scopes and quotes onto a common denominator.
This is where the most important design choice of the system sits, and it is often misunderstood. If you define a measure in terms of the time work takes today, that measure shrinks along with the market. If certain work becomes faster, the same website is suddenly worth fewer PAR, and every comparison with last year breaks.
| Layer | What it measures | Does it move? |
|---|---|---|
| Size | PAR: how much work this scope contains | No. Fixed measure, anchored to a base year. |
| Effort | how much time one PAR takes today | Yes. Moves with productivity, tooling and AI. Recalibrated annually. |
| Price | euros per PAR | Yes. Moves with the market. Recalibrated annually. |
If AI makes certain work faster, the effort per PAR falls and the price per PAR falls. The size of the assignment does not change. A scope from today therefore stays comparable with a scope three years from now, even if the way it is produced has changed in the meantime.
One derived measure makes the whole thing tangible: an annual contract translates into staffing, so you can see whether it amounts to half a person or three.
PAR is calibrated on the financial archive that IKAg and PitchPoint have built up over more than ten years of agency selection: pitch files and quotes from a range of sectors, from retail and automotive to banking, insurance, government and healthcare.
What makes that archive unusual is not its size but its structure. In almost every file, several agencies priced exactly the same brief. That allows a comparison you never get from standalone quotes: the assignment is fixed, only the agency differs. Some of those files also contain a written-out workload per deliverable, which is the most valuable material there is, because the agency itself states how much work something takes.
A quote is not a dataset. Lines refer to one another, totals sit next to components, and what is one line in one quote is spread across five in another. We therefore work with a subset we can demonstrate to be complete: cases where the individual lines add up exactly to the declared total. There we know for certain that nothing falls outside the count.
That approach immediately produced a warning about ourselves. Measure proportions between items on those complete scopes and you get a distorted picture, because meeting schedules add up neatly while campaigns with dozens of adaptations do not. The share attributed to meetings comes out more than twice too high as a result. Hence a fixed rule in the method: complete scopes are used for prices per deliverable, the full archive for proportions between items.
Not every line in the catalogue rests on equally solid ground, and we do not pretend otherwise. Every deliverable carries a code stating how its value came about.
Calibrated directly on stated workloads or on priced lines from the files. These are the values we are willing to build a benchmark on.
Derived from price lines via the average market ratio between price and workload. Reliable in order of magnitude, with more noise than K1.
Expert value based on market knowledge, not yet calibrated. Usable for counting a scope, not for communicating as a benchmark.
That the catalogue contains estimated lines is not a shortcoming that disappears over time. It is structural: agencies rarely quote specialised deliverables as a separate line, and that is their prerogative. More archive does not solve it; those values are calibrated through targeted enquiry, deliverable by deliverable.
The reason we show these codes rather than leave them out is straightforward. Anyone using a benchmark needs to know which part of it is solid. A system that sounds equally certain everywhere is less trustworthy than one that says where it is estimating.
A question we are often asked: why does PAR distinguish agency types and not advertiser types? Surely a multinational buys differently from an SME?
That is true, and it is in the system. Just on the other axis.
Languages, markets, approval layers, feedback rounds, sector screening. A corporate with four approvers and two languages buys more work than an SME with one decision-maker, even when the brief is identical. PAR counts that as countable characteristics rather than as a label.
That is a deliberate choice. A label is never as precise as a count, and a label gets exactly the interesting cases wrong: the SME working with an international network, and the multinational bringing in a specialised boutique. Both combinations occur and both can work excellently.
The price per unit of work is an agency characteristic, and that has proved measurable. In close to half the cases where the same agency quoted for different advertisers, it applied exactly the same rate. Place different agencies side by side within the same pitch, however, and the spread is real.
One exception is systematic: in public tenders, rates sit noticeably lower than for the same agencies in the private market. That is not an advertiser effect but a tender effect, and it is worth allowing for when comparing quotes from both worlds.
PAR therefore works with segments that differ in scale and in what they have on board: from compact teams, through agencies with a full range of capabilities in house, to international networks with access to several markets. Comparison happens within the right segment, because judging a network agency by the yardstick of a compact team is as pointless as the reverse.
This point deserves emphasis, because it is easily misread. The segments say something about the scale of the collaboration, not about its quality. A higher price per unit of work is not inefficiency: it often buys more strategic capacity, more specialisation, international coordination, or the assurance that a large volume will be handled reliably. A lower price per unit is equally not automatically a bargain.
Which segment suits an advertiser is a strategic consideration about ambition, complexity and geography. PAR does not answer that question and does not try to. It only ensures that a quote is judged against the right reference, so that no type of agency systematically appears too expensive or too cheap through comparison with a different kind of collaboration.
The price side rests on a broad and robust body of observations. The effort side, whether a more compact team handles the same deliverable with less workload, does not. The reason is mundane: agencies rarely write out their workload in a quote, and that is their prerogative. What the model states about effort is therefore an assumption and not a measurement, and it says so explicitly.
A benchmark nobody tests is an opinion with numbers on it. The method therefore has three built-in checks, and all three have found genuine errors.
Deliverable values and package prices come from different sources. Build a campaign out of individual deliverables and you ought to arrive at what agencies charge for such a campaign. The first time round it did not match: the build-up came out well over a third too high.
The cause turned out to be double counting. One source called the creative thinking a brand platform, the other called that same thinking a key visual, and we counted both as though they were successive steps. Once those levels were separated, the build-ups fall within a few per cent of the market values.
With every expansion of the dataset, we place randomly selected extracted lines next to the original cell in the source file. Simple, dull, and it catches errors no clever test finds.
That is how it emerged that a number of agencies in one tender had given their rates inclusive of VAT, making them appear a fifth more expensive. At one agency, however, that same note sat above amounts that were exclusive, and blind correction would have made that agency a fifth too cheap. Precisely the kind of error that upsets a segment classification.
Another find: what we initially read as an annual package for social media management turned out, on opening the source lines, to be a strategic deliverable produced once a year. Had that entered the catalogue as a package, a counting rule would have stripped every social component out of every scope.
When material arrives that was not in the set before, the existing catalogue is tested against it without being adjusted. In the most recent round, several values were confirmed within a few per cent of their existing level, including the three campaign packages and various meeting items. One value was revised downwards.
That we publish these three checks and not only their outcomes is not modesty. It is the only argument that counts: a system that never finds an error has no check built into it.
A measure is only usable if two people working from the same scope arrive at the same figure. That requires counting rules, and these four do most of the work.
That last rule is stricter than it sounds and regularly costs us data that looks usable. An item of a hundred thousand euros for "newsletter" is not a newsletter but an annual programme, and anyone who does not check that is building a benchmark on noise.
Once a year, and each time per layer separately. The PAR values themselves stay fixed on their base year, since that is the entire purpose of a measure. The effort per PAR and the price per PAR are recalculated on the past year's files.
Every change is traceable to a source: previous value, new value and origin are recorded for each adjustment.
Every value in the catalogue carries its reliability code, and where a parameter is an assumption rather than a measurement, it says so explicitly. Where the data cannot support an outcome, we take the cautious route: better a derived value than a measured one resting on too few observations.
What we do not do is force an outcome because it suits us better. In the most recent round we failed to meet one of our own targets, and that stands unchanged in the internal report with the reason alongside it.
Tick what your agency delivers, tell us about your organisation, and enter your quote. In two minutes you see how much work that scope holds, what the market asks for it, and how much of it goes to meetings.
You get an access link in your inbox. No password, no subscription.
PAR works best beforehand: one scope, expressed in work, sent to every agency you approach. They then all quote on the same thing, and comparing becomes reading instead of guessing.
Have your scope measured, your fees benchmarked or your entire agency selection handled objectively. The underlying values and benchmarks are shared within an active engagement.
Get in touch