Benchmarking Sites Fairly Across a Portfolio

Two buildings in the same portfolio score 82 and 68 on the same month's inspection. One is being run badly, or the other is easier to run — and until you know which, the comparison is a story rather than a measurement.

Benchmarking properties is the practice of comparing like with like: the same standard, the same sampling, the same scorer behaviour and the same normalising unit. Do that work first and the numbers become useful. Skip it and you get a ranking that damages a manager's credibility the first time someone asks how the score was produced.

Key takeaways

• Comparability is established before comparison: same checklist, same scoring scale, same sampling, same season, calibrated scorers.
• Normalise by a unit that describes the work — area inspected, occupants served, visits performed — and say which one you used.
• Report bands and trends rather than ranks, because a single visit cannot separate sites that are genuinely close.
• Never convert your own portfolio numbers into an industry average, and never accept an unsourced benchmark from a vendor.
• Re-baseline whenever scope or standard changes; an unbaselined comparison measures the change, not the site.

Why does benchmarking properties go wrong?

Because most differences between sites are differences in the measurement, not in the cleaning. The four biggest offenders are scope, standard, sampling and scorer.

A site with a tighter specification gets scored against a checklist written for a lighter one. A site inspected in the week after a tenant fit-out is judged on construction dust. A site where the inspector samples twelve washrooms is compared with one where they sampled two. A site inspected by a manager who scores 3 for a mark that another manager calls a 4 loses two points a year without anything changing on the floor.

None of that is statistical bias; it is ordinary operational variety, and it swamps the signal unless you control it deliberately. Where the programme sits in the wider operating model is covered in property management operations.

What makes two sites comparable?

Five conditions, all cheap to check before you publish anything.

Fix those five and a five-point gap means something. Miss one and the comparison is an opinion with a decimal point.

What should you normalise before comparing?

Divide by a unit that describes the work being done, and be explicit about which unit, because each one flatters a different kind of building.

For a performance score, normalising is mostly about sampling: the score is already a percentage of achievable points. For cost, normalising is the whole game. Compare a cost per square foot of gross floor area across four buildings of different use and you have compared the buildings' uses, not their management.

Should you benchmark quality, cost or both?

Benchmark both, report them separately, and resist the temptation to blend them into one number. A single composite score hides the trade-off an owner most needs to see.

For quality, use the score your inspection method produces and keep the method identical. For cost, use the contract value plus the extras you actually paid, divided by the normaliser you chose. Then look at the two together as a scatter rather than a table: high score with high cost, high score with low cost, low score with low cost — each quadrant asks a different question. Which metrics are worth reporting at all is covered in inspection metrics for managers.

Building condition and cleaning quality are different measures, and mixing them produces a figure nobody can act on. The scoring discipline itself is covered in building condition scoring; the comparison discipline is a separate step on top of it.

How do you build the comparison set?

Write a one-page definitions sheet, apply it to every site for a full cycle, and publish nothing until the cycle closes. The sheet is what makes the numbers defensible six months later.

• Define each metric in words, including what is excluded
• Fix the checklist version and weights for the comparison period
• Specify the sampling: which areas, how many, at what point in the visit
• Specify the time window: same weeks, same season, same occupancy conditions
• Name the inspector for each visit and record calibration dates
• Store the raw inputs, not just the score, so a later reviewer can recompute

That last point matters: a benchmark nobody can audit gets ignored the moment it produces an inconvenient answer.

Illustrative worked example

The figures below are invented to demonstrate the arithmetic and the traps. Substitute your own inspection scores, areas and contract values; nothing here is an industry reference.

The naive read is that C is the worst performer and B the cheapest. Both conclusions are wrong. C sampled six areas instead of sixteen, so its score rests on a thinner and probably harsher sample, and it ran on a lighter specification than the others.

Repair the comparison before concluding: bring C up to the same sixteen-area sample, or exclude it and flag the gap. Then set the band — A and D together, B slightly behind on score and ahead on unit cost, C unclassified until its sampling matches. A fair comparison, bought with one extra column of information.

How do you avoid false precision in rankings?

Use bands, report trends over at least three visits, and state the sample size next to every number. A rank implies a difference the data cannot support.

A practical convention: score differences under five points are treated as noise and reported as a shared band; five to ten points is a watch item; more than ten points justifies a site-specific review. Publish the convention before the first results, so the rule is not written to fit the outcome.

A site that moved from 71 to 78 across three quarters is doing something right, even if it still sits below a site holding steady at 80.

What should you never claim from your own data?

Never present your portfolio's numbers as an industry benchmark, and never accept one that arrives without a method. Both destroy trust in the programme.

Four specific prohibitions:

• Do not average four sites and call the result a market rate or an industry norm
• Do not compare your site to a published figure unless you can see its checklist, scale and sampling
• Do not benchmark cleaning cost against a building with a materially different specification
• Do not infer portfolio-wide performance from one cycle or one inspector

If a vendor quotes an industry average, ask three questions: which sample, which specification, which year. If those cannot be answered, the number is marketing. Where a genuine external standard helps, use a published one — APPA's levels of cleanliness give a shared vocabulary for what a level means, which is different from claiming a portfolio average.

That distinction is what separates internal benchmarking from competitive tendering: one improves your own operation, the other tests the market. Vendor benchmarking that holds up covers the tender side.

How do you present results without starting fights?

Report the method before the numbers, keep individual sites anonymous in group settings, and pair every score with the next action its site manager has already agreed to.

The pack that survives a portfolio meeting has four parts: how the comparison was built, the sample and its limitations, the bands with trend arrows, and one action per site with an owner and a date. Nothing about league tables, and no surprises — each site manager should see their own result before the meeting, never during it.

How often should you re-benchmark?

Run the comparison quarterly within a fixed cycle, review the method annually, and re-baseline immediately after any change in scope, standard or vendor.

A re-baseline means the old numbers stop being comparable and start being history. Say so in the report, because an unbaselined chart reads a legitimate improvement as a decline. When a contract is renegotiated or a specification is rewritten, that date becomes the new zero.

Frequently asked questions

What is the single most important step in benchmarking sites?

Matching the measurement. Same checklist, same scale, same sampling and calibrated scorers before any comparison is drawn. Everything else, including the arithmetic, is straightforward once that is done.

Can I compare sites with different cleaning specifications?

Only after re-scoring against a common subset of items, or by reporting them in separate groups. A site contractually required to deliver less cannot be fairly judged on a common score without noting the difference.

How many sites do I need before comparing?

Three is enough to see spread and four or more lets you separate a genuine outlier from ordinary variation. With two sites you have a conversation, not a benchmark.

Should benchmarking results affect vendor payments?

Only where the contract already ties payment to a defined service level and the measurement method was agreed in advance. Introducing financial consequences after the results are known turns a diagnostic tool into a dispute.

How do I benchmark properties without industry data?

Build an internal baseline: score every site with the same method over a full cycle and compare sites against each other and against their own history. That is enough to drive decisions, and it is honest about its limits.

Where a fair comparison lands

A benchmark is a measurement of your own operation, not a verdict on people. Build it from the same checklist, the same sampling and the same scale, normalise honestly, report bands and trends, and keep the raw inputs so anyone can check the arithmetic.

NeatScore gives you that consistent basis: the same checklist across every site, item-level scoring with critical items flagged, photos per finding, offline inspections that sync, and dashboards that show trend and site comparison without mixing unmatched data. Try it free for 14 days, or see pricing.

Not legal or financial advice. Contract terms, payment mechanisms and the use of performance data in vendor discussions vary by agreement and by province. Confirm with the owner and, where money or termination is involved, with the owner's adviser.