Honestly? To many people, its when they're doing as well or better than their peak. Which is an unreasonable expectation.
numbers are a funny thing.
do note, nothing below is, can be, or will be backed up by "facts". its only an example. likely a bad one

. I like using bad examples!
in the before times, if WOTC sold 8,000 units of X and paizo sold 1,000 units of Y, WOTC could rightfully claim to be the biggest seller.
Then, WOTC announces that X shall, at some future date release X version 2. as word gets out, sales of X version 1 decline from 8,000 units sold to 800 units sold ('cause, why buy a DOA X amirite?). Paizo continues to sell at their normal clip of 1,000 units. now, suddenly, Paizo can announce that they are #1. this is true, though there is no change in volumes sold.
Fast forward, Wotc X version 2 "blows up" and becomes the new hotness and a global sensation. X version 2 sales go from their "low" of 800 to 40,000. Paizo continues to sell at their normal clip of 1,000. Wotc is now the dominant leader with all others a far distant 2nd. and yet, Paizo continues to do "just fine" relative to their past performances and trends. Presumably their business model is predicated on selling 1,000 units so they are not in any way "in jeopardy" relative to their business plans and shareholder expectations.
if as many suggest, rising tides raise all boats and Paizo sees their product Y sales increase from their historical 1,000 units to 1,200 units - a modest increase when compared to Wotc, is still from Paizo's POV, "selling better than any product before".
I get the human need to compare 1 thing against another as a benchmark for success - but while products "compete" with each other, not every product must (or wants) to compete head-on using marketshare as their only metric. Rolex has no interest in selling more units than Timex - though they both sell "watches", some could argue that because Rolex doesn't out-sell Timex, they must be doing worse. Everyone would laugh at that - because its pretty obvious that their entire business model is entirely different from each other.
This is a case, IMO, where observing actions - and any changes in their actions - may be a stronger health indicator than any stat released by VTT or ICV2 counts. Has release schedules changed? quality of product? fan engagement (via conventions or organized play)? observable downward trends for staffing metrics (head count, contractor hours, Salary, etc)? quantity and quality of book art? quality of book production? those all are likely better leading indicators of a companies "health" than with a zero-sum market share mindset.
but that's just me
Cheers,
J.