Six weeks to a forecast that outperforms yours See how
Back to planning notes

Half your benchmark gap is a definition and a product mix

Nov 10, 2026, 10 min read

In short

An external benchmark stacks three things on top of each other: how each participant defined the measure, what business each participant is structurally in, and how well each one is run. Only the third is worth funding a programme against, and it is usually the smallest. Plausible in-scope choices on supply chain cost as a percentage of revenue move the same company between roughly 6 and 12 percent. Reweighting your own per-segment performance to the peer group's channel mix separates the structural part of a gap from the operational part in an afternoon. The comparison carrying the most usable information is the one between your own sites, on your own definitions, where you can ask the people involved why.

The slide has two bars on it. Yours at 2.1 inventory turns, a peer group median at 3.8, and a gap converted into something like eleven million of released cash. By the time the deck reaches the operating review, the number has been written into next year's plan and given an owner.

Then somebody asks which companies are in the peer group. The provider does not disclose participants. Somebody asks how those companies counted inventory, and the answer is that it follows the standard definition. Nobody in the room can name the standard or say whether their own submission followed it.

What that bar chart contains is three quantities added together: how each participant defined the measure, what business each participant is structurally in, and how well each one is actually run. Only the last is worth spending money to close, and it is usually the smallest. Pulling them apart takes about a week and changes the size of the prize more often than it confirms it.

What a percentile is ranking

Start with the sample, because the sample decides the percentile and almost nobody asks about it.

Participation in a benchmarking study is voluntary and usually mediated by a consulting engagement. The firms that submit are the ones already measuring themselves. So the median describes the middle of the companies that chose to fill in the form, a narrower and better-organised set than your industry.

The second thing to ask is how many are left after the filters. A study with 400 respondents sounds like a distribution. Filter to your industry, then to your region, then to a revenue band that resembles yours, and the cut you are being compared against can be eleven companies. The median of eleven is the sixth one, so a single participant with an unusual accounting treatment moves your quartile.

Vintage matters as well. Submissions in a published study frequently span two or three years of fiscal periods, so a top quartile threshold can be drawn partly from a demand environment that no longer exists, which makes the comparison one of calendars.

Composite rankings add a fourth problem, which is that the weights decide the order. Gartner's Supply Chain Top 25 methodology, published each year alongside the ranking, combines opinion components from a peer panel and an analyst panel with financial measures including inventory turns and return on physical assets. The weighting between opinion and financials is stated, and it is a choice. Move it and the list reorders, so the ranking answers a question about the weights as much as about the companies.

None of this is an argument against benchmarking as Robert Camp described it in his 1989 book, which came out of Xerox studying how another company actually ran a process, in detail, on site. That activity and a percentile table share a name and very little else.

The definition moves the number further than the performance does

Take supply chain cost as a percentage of revenue, which APQC has published in its Open Standards Benchmarking work for years and which turns up in nearly every executive comparison.

Here is a business with 800 million of gross revenue and the following annual costs: outbound freight 24 million, warehousing 19 million, inbound freight 14 million, planning and customer service payroll 7 million, import duty 9 million, obsolescence write-offs 5 million, and inventory carrying cost 12 million.

Add everything and you get 90 million, which is 11.25 percent of revenue. Now take the narrower scope that many finance teams use, where inbound freight and duty are landed into cost of goods sold and never appear in a supply chain cost line, and where write-offs sit with the commercial team and carrying cost is not booked anywhere at all. That leaves 50 million, which is 6.25 percent.

Then decide whether revenue means gross or net of trade spend. At 720 million net, the two scopes give 6.9 percent and 12.5 percent.

The same business, the same year, the same ledger, and a range of roughly six to twelve percent, settled by an accountant who was not thinking about benchmarks. Your position against a peer median of 8.4 percent is decided inside that range. If your submission used the wide scope and half the peer group used the narrow one, the gap on the slide is an accounting artefact with a project plan attached.

The same effect runs through every composite measure. Order fulfilment cycle time depends on which event starts the clock. Cost to serve depends on how overhead is allocated (W4). On-time delivery depends on whose date you measure against, which is the argument that never ends between a supplier and a retailer (BB4). The Supply Chain Council built the SCOR metric definitions from 1996 onward specifically so that these comparisons would mean something, and ASCM maintains them now, but a definition only helps when both sides used it and can prove which fields they mapped.

Mix explains more of the gap than management does

Definitional noise is the first layer. Structure is the second, and it is bigger.

Two grocery suppliers of similar size. One sends 75 percent of its volume to national accounts in full pallets on scheduled routes. The other sends 60 percent to independents and wholesalers in mixed cases with weekly drops. Every physical cost per case differs between those two profiles for reasons no planner controls.

Put numbers on it. Cost per case to serve national accounts is 3.10 for the peer group and 3.30 for you. Cost per case to serve independents is 8.40 for the peer group and 8.10 for you. Their mix is 75 percent national, yours is 40 percent.

Their blended cost per case is 0.75 times 3.10 plus 0.25 times 8.40, which is 4.43. Yours is 0.40 times 3.30 plus 0.60 times 8.10, which is 6.18. On the slide, you are 40 percent more expensive.

Now hold the mix constant. Apply your own per-segment costs to their mix: 0.75 times 3.30 plus 0.25 times 8.10 gives 4.50 against their 4.43. The real operational gap is under two percent. And when you look inside it, you are about six percent worse than them on the pallet business and about four percent better on the mixed-case business, which is the segment that carries most of your volume.

The whole apparent gap was the channel mix, and the corrective action implied by the original slide, which was to attack distribution cost, points at the part of the business you are already better at. This is an ordinary index number problem and it has an old name in statistics, from Simpson's 1951 paper on contingency tables, where an effect visible in every subgroup reverses in the aggregate because the subgroups are weighted differently.

Mix is the usual culprit, and it is not the only structural difference that behaves this way. Ownership boundaries move cost across the line between two companies without changing anything physical: a peer that uses a third party logistics provider has warehouse cost in a service fee rather than in payroll and depreciation, and a peer that sells ex-works has no outbound freight at all. Make versus buy does the same to manufacturing measures. Echelon count does it to inventory. Before comparing any measure, write down what would have to be true about the other company for the comparison to be fair, and check whether you believe it.

The comparison you control is inside your own business

Everything above argues for spending most of your benchmarking effort internally, where the definitions are yours, the data is auditable, and you can walk over and ask why.

The number worth computing first is the internal spread. Take one measure, calculated the same way from the same source, across every site or category or planner. In most multi-site businesses that spread is wider than the external gap on the same measure, and unlike the external gap it is fully actionable, because the top performer is reachable by phone and has no reason to hide their method.

That also gives a better target than a percentile does. A target set at what your own best site achieved last year, on your own definition, in your own conditions, is defensible in a way that a quartile threshold is not, and the person being handed it can go and look at how it was done. Setting the target level and its action trigger is a separate discipline with its own failure modes (BB12).

Two other internal comparisons earn their keep. Your own history over 36 months with control limits on it, in the form Shewhart set out in 1931, tells you how much the measure moves on its own, which is the only way to know whether this quarter's change is a signal. And where sites differ on several inputs at once, data envelopment analysis, from Charnes, Cooper and Rhodes in 1978, ranks them on the efficiency frontier rather than on any single ratio, which stops a site being praised for a low cost per case it achieved by holding four extra weeks of stock.

What internal comparison cannot do is tell you whether an improvement you made was caused by the thing you changed, which is a different problem with its own designs (AX2).

Questions that make an external number usable

External data still has uses, and what separates a usable figure from a decorative one is how much you can find out about it. Before a benchmark enters a plan, get answers to these.

What is n after every filter you applied to produce my comparison group? If the answer is under twenty, treat the median as an anecdote with a decimal point.

What is the field-level definition, and which of my cost centres map into it? Ask for the mapping in a spreadsheet. Do the mapping yourself from your general ledger rather than accepting the one your own team submitted last time.

Which fiscal years are the submissions from?

Is any of it audited, or is all of it self-reported?

What is the interquartile range and the shape of the distribution? A median with a wide, skewed spread behind it means the sample contains several different business models, and the median is an average of things that should not have been averaged.

Then use whatever survives for the thing external data is genuinely good at, which is generating hypotheses. Treat a large, persistent, mix-adjusted gap on one measure as a reason to go and look at how someone else runs that process, rather than as a target or as a line in a savings plan.

Where this stops

Internal benchmarking has a serious blind spot that external comparison covers. Comparing your sites to each other cannot reveal a practice that none of them has ever tried, and when a whole industry has moved to a different operating model, your internal spread will be tight and uniformly behind, which looks like control. Site visits and industry bodies are how that gets found, which is the original Camp version of the activity.

External data also works better on structural questions than on operational ones. How many distribution centres companies your size typically run, what proportion outsource transport, how much of the industry has moved to a particular replenishment model: these are counts of choices rather than measures of performance, and definitional variance does much less damage to them.

The internal spread has its own trap. Pick your worst-performing site to run a pilot on and it will improve regardless, because you selected it on a low reading and low readings are partly noise. Whatever the pilot does, part of the gain belongs to the selection.

And a mix-adjusted gap that comes out near zero should not be read as reassurance. It tells you the comparison has run out of things to say, and leaves entirely open whether you and your peer group are both slow.

Take the one benchmark figure your plan is currently built on, and spend a day rebuilding it from your own general ledger against the provider's field definitions, line by line. By the end of the day you will know how much of the gap is arithmetic, and that number is usually large enough to change what you do next.

More on this in evaluation on the DemandLabs platform.

See what DemandLabs could do for your plan

Request demo

More planning notes

View all

Aug 23, 2026

Six questions that tell you whether an AI agent is an agent

Gartner coined the term in June 2025 and applied it to supply chain planning in May 2026. Agent washing, meaning conventional automation relabelled as agentic, and their warning was specific: most current offerings do not independently re-sequence objectives, negotiate trade-offs or adapt execution logic. Most of what is being sold as agentic improves the user experience through query interpretation, recommendations and conversational support, without changing decision quality or how decisions get made.

Aug 23, 2026

The 2027 deadline that turns a planning decision into an ERP project

Mainstream maintenance for SAP's Advanced Planning and Optimization suite ends on 31 December 2027. After that, extended support is available at a premium on the maintenance fee and comes with a commitment to move to S/4HANA.

Aug 23, 2026

Run the bake-off on your own history before you sign anything

Vendor accuracy claims are not comparable across vendors, because each one was measured on a different dataset at a different grain with a different baseline. A quoted number is an artefact of somebody else's data.