How do you actually build a defensible peer set?

Knowing that comps matter is one thing; building a clean set is a repeatable craft. Here's the workflow professionals follow, and the traps at each step.

The four-step build

1. Start from the industry. Use the target's sector/industry classification as your raw pool — every company that makes money roughly the same way. The top of the pool is below.

2. Filter by size. Keep companies in a sensible range of the target's market cap (a common habit is same order of magnitude — don't compare a $500bn giant against a $2bn small-cap). Size drives risk, liquidity, and the multiple the market assigns.
3. Filter by business fit. Read what each company actually does. Two firms can share an industry code yet earn money completely differently — a payments processor and a retail bank are both "financials" but not comps. Cut the ones whose economics don't match.
4. Sanity-check growth and margins. Glance at each survivor's growth rate and profitability. Drop the ones that are wildly different; a hyper-growth outlier or a barely-profitable straggler will distort the median.

You'll usually land on somewhere between five and fifteen names. Too few and one company sways everything; too many and you've clearly stopped being selective.

A worked pass

Say your target is a mid-sized branded-food company. Your raw industry pool has 40 names. Filtering to a similar size cuts it to 18. Removing the ones that are really commodity-ingredient suppliers rather than branded-consumer businesses cuts it to 9. Dropping two that are in the middle of a restructuring (temporarily distorted earnings) leaves 7 clean comps. That set of 7 — each one there for a reason you could defend out loud — is what you compute a median multiple from.

The discipline that keeps you honest

Write down why each name is in or out. If you can't justify a company's inclusion in a sentence, it probably shouldn't be there. And decide your filters before you look at whether they make your target look cheap or expensive — deciding after is how motivated reasoning sneaks in. The set should reflect the business, not the conclusion you were hoping for.

In the data

The screen is the pool-builder, and it has one limit worth knowing before you trust it: it screens the latest day only. Here are the smallest US "Technology" names that clear a $50 billion floor, the size filter of step 2, with each one's industry:

The floor cut the pool to 86 US-listed names on 28 September 2026, and to 381 listings with every exchange counted, because the same company lists in several places. And the pool you filter today cannot be reconstructed as it stood a year ago: companies bought, merged or delisted since then have simply left the list.

Try it now

1. Start from the industry. The raw pool is the sector screen under step 1 of the build above, largest first. Name the five and say which of them you would not have filed under the same label.
2. Filter by size. Read the industry of each row at the edge of the $50 billion table above and say which of them belong in a peer set for a software or chip company, and which only share the sector label.
3. Filter by business fit. Two survivors filed under the same industry label:

Read both descriptions. Two firms can share an industry label and earn money completely differently; say in a sentence what each one owns and sells, and whether you would keep both in one set. 4. Sanity-check growth and margins. A shortlist of six survivors on the multiple, the margin and the growth rate:

Drop the wild outliers — a hyper-growth name or a barely-profitable straggler will distort the median — and write down which ones you dropped and on which measure. 5. Compute the median multiple of your shortlist by hand, then recompute it with a single name added or removed. That sensitivity is the whole lesson — and write down, in a sentence each, why every survivor is in the set. 6. Before you rely on the set, write down the date you built it. A screen has no history, so this pool is reproducible going forward and never backward.