A software company that spends a fifth of revenue on engineering reports that spending as an expense in the year it happens. The code those engineers wrote does not appear anywhere on the balance sheet. Next year the company sells the product built from it, books the revenue, and the cost of building it has already been written off against a period when the product did not exist.
The accounting is defensible. Nobody can prove the code has value until customers pay for it. But if you are trying to work out whether a software business is expensive, using numbers that treat its main investment as a running cost and its main asset as nothing will send you to the wrong conclusion in both directions.
What expensing does to the two numbers you use
Two numbers get distorted and they get distorted in opposite directions, which is why this adjustment surprises people the first time they run it.
Operating margin is understated. Every dollar of this year's research is charged against this year's profit, including the portion that is building something which will not generate revenue for two or three years. A company growing its research spend fast is penalised hardest, because the gap between what it is spending now and what its past spending is earning now is widest.
Return on invested capital is overstated. The denominator is missing the accumulated research asset entirely. A company that has spent a billion dollars building a product over five years shows an invested capital base that contains none of it. The return on capital looks spectacular because the capital has been deleted rather than because the returns are high.
So a software business simultaneously looks less profitable than it is and more capital efficient than it is. Screens that rank on margin push these names down. Screens that rank on return on capital push them up. Neither ranking is telling you about the business.
The adjustment on one sheet
Pick an amortisation life. Five years for enterprise software and platform businesses is the convention I use, three years for anything consumer facing where the product cycle is genuinely short. The point is to pick one and apply it to every company you compare, because a life chosen while a specific name is on your screen is an opinion about that name.
Pull the research and development line from the last six annual reports. Build the capitalised asset by taking each year's spend and writing off a fifth of it per year. This year's spend is fully on the books, last year's is four fifths, the year before three fifths, and so on. Sum them and you have the unamortised research asset.
Then restate two lines. Add this year's research spend back to operating profit, since you are now treating it as investment rather than cost. Subtract this year's amortisation charge, which is a fifth of each of the previous five years' spend. And add the unamortised asset to invested capital.
What the restatement actually moves
Round numbers, made up, chosen to make the arithmetic visible. A software company with 2,000 of revenue, 400 of research spend, and 100 of operating profit, so a five percent operating margin. Research has been growing at fifteen percent a year, so the prior five years ran roughly 348, 303, 263, 229 and 199.
The amortisation charge this year is a fifth of each of those five years, which comes to about 268. Add back the 400 of current spend and subtract the 268 charge, and operating profit goes from 100 to 232. The margin goes from five percent to about eleven and a half percent. That is a business that looked marginal and now looks reasonable, on identical cash flows.
Now do the denominator, which is the part people skip. The unamortised asset is 400 plus four fifths of 348 plus three fifths of 303 plus two fifths of 263 plus a fifth of 229, which is roughly 1,011. If reported invested capital was 800, it is now about 1,811.
Return on capital before the adjustment, at a 21 percent tax rate, is 79 of after-tax profit on 800 of capital, so just under ten percent. After the adjustment it is 183 on 1,811, which is just over ten percent. It barely moved.
That is the finding worth carrying away, and it is not the one most write-ups lead with. Capitalising research transforms the margin and leaves the return on capital almost untouched, because both the numerator and the denominator grow by roughly the amount of research you have been running. The adjustment is a margin correction dressed up as a returns correction.
The composite that only partly moves with you
This matters for how you use a screen like this. The board is genuinely useful as a shortlist generator across a wide universe, and its own scoring is internally consistent. What it cannot do is reflect an adjustment you made in your own spreadsheet. The composite score, the verdict, and the sub-scores are all computed from the reported figures. When your restated margin says a name is reasonable and the board says overvalued, you have not found a bug. You have found the gap between reported accounting and your restatement, which is the entire reason you did the work.
The practical workflow is to filter to the technology bucket, take the names, and do the restatement offline. There is no research spend column and no adjusted margin column on this view, so nothing on the screen re-ranks when you capitalise. The ranking that changes is the one in your own file.
The names where this adjustment flatters a bad business
Capitalising research is not a device for making expensive software look cheap, and it will do exactly that if you apply it carelessly. Three cases to watch.
The first is a company that already capitalises internally developed software. Many do, partially. If you capitalise the reported research line on top of that, you have double counted, and the margin improvement you calculated is fictional. Check the cash flow statement for a capitalised software line before you start.
The second is research that is really maintenance. A company spending fifteen percent of revenue keeping an existing product working is not building an asset, it is paying the cost of staying in business. There is no clean disclosure that separates the two, so the honest approach is to look at whether revenue per unit of cumulative research spend has been rising or falling over five years. Falling means the spending is defensive and a five year asset life is too generous.
The third is the one that costs money. A company whose research spend is growing much faster than revenue will show a large add-back and a small amortisation charge, so the adjustment produces a flattering margin every single year without the business ever producing profit. The restatement is arithmetically correct and economically empty. If the adjusted margin is good but the cash flow statement shows the company has never generated cash, the adjustment has told you nothing you should act on.
The one check before you pay a growth multiple
Take the software name you are closest to buying. Write down two ratios. Research spend divided by revenue, and cumulative research spend over five years divided by current revenue. The first tells you how much the margin understates. The second tells you roughly how big the missing asset is relative to the business.
If the first ratio is under about eight percent, skip the adjustment. The distortion is inside the noise of everything else in your model. If it is above fifteen percent, the reported margin on that company is not a number you can compare to anything, and running the five year schedule is an hour that changes what you think you are buying. Do it before the position exists, not after, because this is the rare piece of analysis that is much easier to take seriously when you have no money riding on the answer.