Earlier today, I had four Super Intelligence tools read On Holding AG’s financial statements and infer what the company paid for the ON.com domain name. Three of the four landed on roughly CHF 15 to 16 million, about USD $16 to 17 million, by spotting an anomalous CHF 16.4 million of additions to “Patents, licences and other rights” in 2022 and backing out the routine spending.
This post gets to the same number by a completely different road. To my knowledge, it is a road nobody has taken before in domain name reporting: inferring the original purchase price from the amortization the asset leaves behind in later years. The result is $16.7 million, give or take a couple of hundred thousand, and it is worth a post of its own, for reasons that go beyond ON.com.
The idea, without the accounting
If a friend tells you their mortgage payment and the term of the loan, you can work out roughly what they paid for the house. The payment is the price spread over the years. Run it backwards and the price falls out.
Companies do the same thing with intangible assets. When On bought ON.com, it did not expense the price in one go. Under IFRS it capitalised the domain name and has been writing it off in equal annual instalments ever since. Those instalments are the amortization charge, and On discloses them every year, and every half-year, in Note 3.5 of its accounts. The term is disclosed too: in the year of the purchase, the maximum useful life in the category jumped from 15 years to 20, and the domain name is the only thing that was added that year that could carry it.
So the price is the instalment times twenty. The only difficulty is that the instalment is hiding inside a line item that also contains everything else in the category, mostly the Roger Federer licence rights and a churn of ordinary patents. You have to separate them. Fortunately On tells you how long each of those lives too, which turns the problem into arithmetic.
Four tools, round two
I gave the same four tools the new brief, with one extra fact they did not have the first time: DomainTools.com WHOIS history shows the domain name changed hands in December 2022. That pins the start of the amortization clock. Here is where they came out.
| Tool | Implied cost of ON.com | How it handled the old patents expiring |
|---|---|---|
| Claude | CHF 15.5–15.6M (band 15.4–15.8M); USD $16.5–16.9M | Reconstructed the run-off from the exact 2021 and H1 2022 figures, then fitted all seven annual and half-year windows from 2022 to 2025 at once |
| ChatGPT | CHF 14.7–16.3M at a 20-year life; central USD $17M | Modelled it, and explicitly warned that the naive shortcut gives CHF 8 million for exactly this reason |
| Gemini | CHF 12–14.9M; USD $13–16M | Guessed it at CHF 0.25–0.45M a year, too low, and assumed 5–10-year lives for routine items when the filings say 4 |
| Grok | CHF 5–9M; USD $5–10M | Ignored it entirely |
The dividing line is a single accounting fact. On’s ordinary patents have a four-year life, and a batch of them bought before 2020 was expiring in 2022 and 2023, at the same moment the domain name started amortising. Roughly CHF 0.6 million a year of old charges disappeared while CHF 0.78 million of new charge arrived. Net, the line went up by only CHF 0.4 million. If you take that 0.4 at face value and multiply by 20, you get CHF 8 million, and you are wrong by half. Grok did that, with a further haircut for good measure, and arrived at CHF 5 to 9 million. It never noticed that this contradicts its own first report: a CHF 5 to 9 million domain name would leave CHF 7 to 11 million of unexplained “other rights” in the second half of 2022, and a CHF 17 million fourth-quarter cash outflow that nobody can account for. It also stated that only annual amortization by category is reported, which is false; the half-year interim statements have it, and they are the key to the whole exercise.
Gemini saw the run-off but guessed its size instead of calculating it, and it assumed longer lives for the routine items than the filings disclose. Both errors push the same way, and its range came out CHF 3 million light. ChatGPT got the mechanics right and was careful to the point of refusing to assume the 20-year life, which is why its band is wider than mine. Kudos to it for naming the trap in plain language. My current leading tool, Claude, built the full model, which is what the rest of this post walks through.
For the nerds
Write the category’s annual amortization as
At = F + Rt + Bt + D/20
where F is the Federer licence, Rt is the routine four-year items, Bt is the pre-2020 base running off, and D is the cost of ON.com. Everything except D is either disclosed or can be built from disclosed figures:
- F = CHF 44,795k ÷ 15 = 2,986k a year, flat. The 2021 accounts give both the cost and the 15-year life, and the exact 2021 charge confirms it is straight-line.
- Rt: each half-year’s additions, amortised over four years. The additions are disclosed each period. The routine part of the second half of 2022 is 16,202k − D, which is the hinge: every franc attributed to the domain name is a franc moved from a four-year schedule to a twenty-year one.
- Bt: calibrated from the periods that are reported in thousands rather than millions. Subtracting F and R from the exact figures leaves 361k, 327k and 337k per half-year through H1 2022. At four-year lives the base is fully gone by the end of 2023, so 2024 and 2025 are base-free.
Take 2024, which has no base to worry about. The routine items that year are the 2020 remnant (41k), the 2021 additions (253k), the H1 2022 additions (50k), the 2023 additions (300k) and the 2024 additions, pro-rated (150k), plus (16,202 − D)/4 for the second half of 2022. Add Federer and the domain name:
A2024 = 2,986 + 794 + (16,202 − D)/4 + D/20 = 7,830 − 0.2 D
On reported CHF 4.7 million for 2024, which means anything from 4,650k to 4,750k. Solving:
4,650 ≤ 7,830 − 0.2 D ≤ 4,750 ⇒ 15.4 ≤ D ≤ 15.9 million
One year, one line of algebra, and the answer is already inside CHF 0.5 million. Doing the same for the H1 2024, H1 2025 and FY2025 windows gives 15.4–16.2, 14.9–15.9 and 15.2–15.7 respectively. Requiring all seven windows from FY2022 onward to hold simultaneously, with the base run-off left free, is satisfied only at D ≈ CHF 15.5 million, regardless of whether December 2022 contributes a full month, half a month or nothing to the 2022 charge. A least-squares fit across all seven gives D = 15.56 million, and at that value every window is inside its reported interval except H1 2023, which overshoots by 12k on a 100k-wide range.
The decomposition also explains the trap. From 2022 to 2023 the domain name added 0.71 million (a full year less December), the old base dropped by 0.52 million as it expired, and routine items grew by 0.23 million. Net change: +0.42 million. The reported change was +0.4 million. The shortcut, 0.4 × 20 = 8 million, throws away the 0.52 million of run-off and halves the answer.
Sensitivities, for completeness. Every post-2021 input is rounded to CHF 0.1 million, and a 0.1 million error in a full-year figure moves D by 0.5 million, so the method cannot do better than about ±0.2 million even with perfect assumptions. The honest band is CHF 15.4 to 15.8 million. A 15-year life for the domain name instead of 20 would cut the implied cost by a quarter, but then the 20-year maximum that appeared in 2022 would have to belong to some other addition, and there was none of any size. Each 100k of the second-half-2022 routine items that actually carried a longer life would lower D by about 0.1 million.
In dollars: the franc traded at 0.925 to 0.94 per US dollar during December 2022, so CHF 15.5 to 15.6 million is USD $16.5 to 16.9 million. A round USD $17 million closing late in the month, or USD $16.5 million early in it, would both book inside that range. The filings cannot tell those apart. They do rule out anything much below USD $16 million or above USD $17 million.
Two methods, two different line items, one shared assumption (the 20-year life), and they overlap at CHF 15.5 to 15.6 million. That is now my number.
Why this matters beyond one domain name
The additions method only works in the year of the purchase. If you miss that filing, or the purchase is small relative to the category, the anomaly is gone. The amortization method works every year afterwards, for as long as the asset is on the books, and it gets more precise as more years accumulate. ON.com will be throwing off a CHF 0.78 million charge until 2042. Anyone who wants to check this analysis can do so from On’s 2026 accounts, and its 2027 accounts, and so on. The same technique applies to any public company that has bought a domain name and left it in a disclosed intangible category with disclosed useful lives, which is a longer list than you would think.
It also says something about the tools. None of the four thought of this approach on their own in round one. A human who has spent years reading these notes did, and then the tools did the heavy lifting in minutes, two of them well and two of them badly. That is the symbiosis I described in the first post: the expert decides which question to ask and which answer to believe, the instrument does the rest, and the expert with the instrument beats either one alone. Round two made the second half of that sentence vivid. Two confident, well-formatted reports were wrong by a factor of two to three, and the only way to know which two was to understand the accounting. The human in the loop is not a courtesy. It is the error-correction.
There is a last twist. By publishing the method here, on the open web, with the arithmetic shown, it becomes part of what these tools will read. The next time somebody asks one of them what a company paid for a domain name, “back it out of the amortization” will be in the toolbox, with a worked example attached. That is a strange and new kind of contribution to make, and I am happy to make it, on the usual condition: link to where it came from.
A word for the copycats
Which brings me, as it always does, to the low-life parasitic copycat bloggers of the domain name industry. They have had the same tools I have for the better part of two years. Same models, same SEC filings, same free Sunday. Not one of them produced anything resembling the first post, let alone this one, because producing it requires knowing what to ask. They could not have prompted a Super Intelligence tool at this level if their traffic depended on it, and their traffic does depend on it, which is why they will be along shortly to rewrite this instead. That is the entire business model: wait for someone else to open the filing, pose the question, catch the errors, and publish; then paraphrase it under their own byline with a vague nod to “reports”. They do not innovate. They have never innovated. The technique in this post did not exist this morning, and four frontier models did not think of it on their own; a human did. Anyone who runs a copycat article on it will be copying a copy of something they were incapable of originating, and everyone who matters in this space knows exactly which sites those are. As I said in May, and in the first post today, I remain largely “on strike” with a backlog of unreported transactions for precisely this reason. Linking to the original is not a courtesy. It is the minimum.
Credit where due
After the first post went up, AbsReturn on X worked the additions line by hand and arrived at about CHF 15 million, using an organic run-rate of CHF 1.1 to 1.2 million a year. That is the same road as my first method, done manually, and it lands in the same place; at December 2022 rates CHF 15 million is about USD $16 million. Kudos to him for showing his working. He also offered a view on who the seller was, which I have not verified and will leave at the link.
If you appreciate this original content, be sure to engage with my “GeorgeKirikos” account on X. If you use this technique, or quote this analysis, link to this post. It is not a difficult standard.
Sources
All figures are from On Holding AG’s filings with the SEC, Note 3.5 (Intangible assets) in each case:
- Form 20-F for fiscal 2021 (2020 and 2021 movements in CHF thousands; useful lives of 4 years for patents and other rights and 15 years for the SLIA 2019 licence; licence cost CHF 44,795k)
- Form 20-F for fiscal 2022 (useful-life range of 4 to 20 years)
- Form 20-F for fiscal 2023 and Form 20-F for fiscal 2025 (2024 and 2025 movements)
- Half-year interim financial statements for 2022, 2023, 2024 and 2025
- 2022 annual report, parent-company accounts (year-end exchange rate of CHF 0.9252 per US dollar)
The December 2022 acquisition month is from DomainTools WHOIS history.