Cohort builder: benchmarking a custom subset of offices

What it is

The cohort builder is a scratchpad benchmark: pick any subset of your network's active offices, choose one metric, choose a comparison universe, and Verinode computes a fresh group-versus-rest distribution on the spot. It answers a question the standing benchmark pages cannot, because they only ever show your whole network against the industry. What if you only look at the Mountain region? What if you compare the five offices that finished onboarding first against everyone else? What if you want to sanity-check a coaching hunch about "the offices that took the new training" before you say it out loud in a leadership call?

Nothing about a cohort you build here is saved. There is no "cohorts" list, no history of past runs, and no record written anywhere that you ran this comparison. The selection lives only in your browser for as long as the page is open. Refresh the page, navigate away, or come back tomorrow, and you start from a blank slate: no operators selected, no result. If you want to keep a cohort's math around, take a screenshot or write the numbers down. This is the tool for "let me check something right now," not for tracking a standing group over time.

As with every HQ surface, the cohort builder never opens a single office's underlying jobs, invoices, or financial line items. Every number it produces is a rank, a percentile, or a median built from your offices' own contributed benchmark facts, the same anonymized layer every other Verinode benchmark reads from. Picking an office into a cohort does not give you a window into that office's private books, it only lets that office's already-aggregated facts sit on one side of a comparison instead of the other.

Where to find it

Open it directly at hq.verinode.ai/benchmarks/cohort. It sits alongside the standing Benchmarks tab (hq.verinode.ai/benchmarks), which shows your whole network's per-metric distributions against the industry, an always-on view built once for everyone. The cohort builder is the ad hoc counterpart: a page you visit when you want to slice the network yourself instead of reading the network-wide numbers.

Building a cohort

1. Pick your operators

The left-hand panel lists every active office in your network that has a linked operator profile. The header reads "Operators (N selected / M active)", so you always know how many you have picked against how many are available. Each row shows the office's location name, with its city and state underneath (a dash when city or state is missing). Check any row to add it to your cohort, uncheck to remove it.

Above the list:

• Search box ("Search by location, city, or state…"), filters the visible list as you type. It matches against the location name, city, and state together, so typing a region or a partial name both work.
• Add visible, adds every office currently showing in the filtered list to your selection in one click, useful after narrowing the search to a region or a set of names.
• Clear, empties the whole selection. Grayed out when nothing is selected.
• Quick-add by state, a row of state chips (only shown when your network spans more than one state) that adds every office in that state to your selection with one click, regardless of what the search box currently shows.

If the filtered list comes up empty, it reads: "No operators match this search."

Only offices with active status and a completed operator profile appear here at all. An office still in "invited" or "seeded" status, meaning it hasn't finished setting up an operator account yet, has no benchmark facts to contribute and is left out of the picker entirely. It becomes available automatically once it goes active.

2. Choose a metric

The metric dropdown offers twelve benchmarks:

| Metric | What it measures | |---|---| | Gross margin | Revenue left after direct job costs, as a percentage | | Net margin | Revenue left after direct costs and operating expense, as a percentage | | COGS ratio | Direct cost of goods sold as a share of revenue | | Labor ratio | Labor cost as a share of revenue | | Material ratio | Material cost as a share of revenue | | Equipment ratio | Equipment cost as a share of revenue | | Operating expense ratio | Overhead (SG&A) as a share of revenue | | Revenue per employee | Revenue divided by headcount, a productivity read | | Cycle time | Days from job open to job close | | Days to pay | Days from billing to payment received | | Supplement response | Days a carrier or adjuster takes to respond to a supplement | | Operator score | The composite 0-100 Verinode Score |

Each metric already knows which direction is good. For a ratio like labor cost, lower is better; for margin, revenue per employee, or operator score, higher is better. That direction is what decides whether the result tile reads green or red later, you never have to remember it yourself.

3. Choose a comparison universe

Two radio options decide what your cohort is measured against:

• Rest of network, everyone outside your cohort across the whole cross-network intelligence layer, the same industry-wide comparison the standing Benchmarks tab uses.
• Rest of this group, only the other active offices in your own network that you did not select into the cohort. Use this to ask an internal question, "how does my top-5-by-revenue group compare to the rest of my own network," without pulling in outside operators at all.

4. Run it

Click Run benchmark. While it computes, the button reads "Computing…" If you haven't picked any operators yet, it shows "Pick at least one operator." instead of running. A cohort is capped at 500 operators in a single run, a ceiling you are very unlikely to hit given a typical network's roster, but the button will tell you if you do.

Reading the result

Once the run completes, a result row appears headed "Result · [metric name]" with three tiles:

• Cohort, labeled "Cohort · n=[count]". The headline number is your cohort's median (P50) for the chosen metric. The line underneath reads the cohort's P25 and P75, the boundaries of its middle half, so you can see the spread inside your own selection, not just its center.
• The comparison side, labeled either "Rest of network · n=[count]" or "Rest of group · n=[count]" depending on which universe you chose. Same shape: median headline, P25/P75 spread underneath.
• Cohort sits at, the headline reads your cohort's median expressed as a percentile of the comparison side, for example "72nd". The sub-line spells out which universe that percentile is against ("Percentile of rest-of-network" or "Percentile of rest-of-group"). This tile is colored to match the metric's direction: green when your cohort's position is favorable, ember red when it is not, and neutral gray when there isn't enough comparison-side data to place a percentile at all.

All dollar figures are formatted the way you'd expect (thousands as "$42k", millions as "$1.2M"), percentages to one decimal, days to one decimal with a "d" suffix, and the operator score as a plain rounded number.

Reading "n="

The "n=" beside each tile's label is the count of distinct offices or operators whose data went into that side of the comparison, your cohort's count on one tile, the comparison universe's count on the other. It is not a job count or a data-row count, it is how many separate businesses are represented. A cohort of five offices always reads "n=5" even if those five offices logged thousands of jobs between them, because the benchmark is built one value per operator so that no single high-volume office can pull the median around by sheer job volume.

The coverage badge

Each result tile carries a small badge in its corner: Early signal, Indicative, Observed, or Verified. This is a confidence label, not a privacy control, it tells you how much data is standing behind the comparison, running from a handful of contributing offices on either side (Early signal) up to a broad, well-populated comparison (Verified). Read it the way you'd read a sample-size note on any survey: a Verified result is safe to lean on in a leadership conversation, an Early signal result is directional and worth another look once more offices have reported.

When the comparison is withheld

Separately from the confidence badge above, there is a hard floor beneath every cohort comparison: if either side, your cohort or the comparison universe, does not have enough distinct contributing operators, Verinode withholds the whole comparison rather than publish a number built from too small or too identifiable a group. When this happens, the percentile tile shows an em dash for its headline and reads "Not enough rest-side data" underneath, and its badge reads "No data." The median and spread tiles for both sides still show whatever real values exist, only the comparison itself is suppressed.

This floor exists to protect the offices on the other side of the comparison as much as your own cohort. A "benchmark" built from one or two contributing operators is not a benchmark, it is close enough to that operator's actual private number that comparing against it would defeat the purpose of anonymizing it in the first place. Financially derived metrics, margin, the cost ratios, revenue per employee, need a noticeably larger contributing cohort to clear this floor than operational metrics like cycle time, days to pay, supplement response, or operator score, because a P&L figure is more identifying than an operations timing figure. You will not see the exact count required either way, the floor is a qualitative safety line, not a number to reverse-engineer.

Cutting your own cohort down to a very small handful of offices (comparing "these two offices" against the rest, for instance) makes the cohort side itself more likely to hit this floor, since your own selection is subject to the same rule as the comparison side.

The demo and real boundary

If your network is a demo or training environment, the cohort builder's comparison universe is drawn only from the demo peer universe, never from real operator data. A live production network, in turn, is only ever compared against real, reporting operators, never against seeded demo data. This split is automatic and always in effect: you cannot accidentally rank a real cohort against demo numbers or the reverse, regardless of which comparison universe or metric you pick.

Empty states

• No active operators in your network at all: the whole page reads "No active operators yet. Cohort math needs at least one active member with an operator profile." Neither the operator picker nor the metric and comparison controls render until at least one office is active.
• A search with no matches: the operator list reads "No operators match this search."
• No operators selected when you try to run: "Pick at least one operator." appears under the Run benchmark button, and the run does not fire.
• Cohort too large: selecting more than 500 operators and attempting to run shows "Cohort capped at 500 operators."
• A data fetch problem behind the scenes: if the underlying benchmark facts can't be read, the result renders with zero counts on both sides and dashes throughout rather than a wrong or partial number. Verinode fails toward "no result" here, never toward a distorted one.

Best-practice example

Suppose you want to check a hunch before raising it at a regional call: do the four offices that completed the new estimating training outperform the rest of the network on cycle time? Open the cohort builder, search for and check-add those four offices (or use quick-add by state if they happen to cluster in one region), set the metric to Cycle time, leave the comparison on Rest of network, and click Run benchmark. If the result reads a favorable percentile with an Observed or Verified badge, you have a real, if informal, data point to bring to the call. If the badge reads Early signal, or the comparison is withheld outright for not enough data, that is Verinode telling you it is too soon to draw a conclusion, not that the training didn't work. Either way, nothing about the check is saved. Run it again next month as more jobs close to see if the picture holds.

Related articles

• Benchmarks at HQ
• Network Health: the network home
• The office leaderboard and composite ranking
• What HQ sees: the network privacy boundary
• HQ overview