Scope: This is a metric-selection test, not a company review. It runs one ratio, flow-through, across 16 audited company-years at four U.S.-listed computer-systems makers: Hewlett-Packard, Digital Equipment Corporation (DEC), Unisys and IBM, fiscal 1993 to fiscal 1996. Every input comes from each company’s own SEC filing, linked below. Nobody is graded and no company is ranked. The four were picked because their audited figures were already reconciled for an earlier Stagnation Assassins case and because, in the same industry and the same years, one grew fast while the other three shrank, stalled or swung on restructuring charges. That spread is the point. All figures are in millions of U.S. dollars. HP’s fiscal year ends in October and DEC’s in June, so their years don’t line up exactly with the calendar-year reporters.
Flow-through is one of the most useful numbers a turnaround team can put on a dashboard. Take the change in operating profit, divide it by the change in revenue, and you know how much of each extra sales dollar actually reached the operating line after overhead took its share.
It is also the easiest of those numbers to misread. In the sample below, the same formula produces 12%, 191%, 3,681% and minus 338%. Some of those readings mean something. Others are arithmetic noise. A CFO choosing what goes on the monthly pack needs to be able to tell which is which before the board does.
The question: when should you trust a flow-through percentage, and when should you stop reading the ratio and build an operating-profit bridge instead?
Short answer: trust flow-through when revenue grew by a meaningful amount, the company’s perimeter didn’t change, and the reported and charge-adjusted versions tell the same story. When revenue was flat or fell, when an acquisition or disposal changed what’s in the numbers, or when one definition gives a positive reading and another a negative one, put the ratio aside and walk the bridge line by line. Four checks below make that call mechanical.
What Flow-Through Measures, and Why It Breaks
[Definition] Flow-through = (this year’s operating profit − last year’s operating profit) ÷ (this year’s revenue − last year’s revenue). Grow revenue by $100 and operating profit by $30, and flow-through is 30%.
Todd Hagopian’s forthcoming book 10-Minute Transformation treats flow-through as the third of its three velocity metrics, the one that shows whether overhead is leveraging growth or absorbing it. The same chapter carries an explicit warning, paraphrased here: because flow-through divides one change by another, a small year-over-year revenue change makes the ratio jumpy, and when revenue and profit both fall, two negatives can divide into a large positive number that has nothing to do with leverage. The book’s remedy is to smooth the reading over three years and to treat any reading on a declining revenue base with explicit care. It says the metric is most dependable on a business that is actually growing.
This article tests that warning on real filings. It also adds one check the warning implies but doesn’t spell out: how much a single point of operating margin moves the ratio.
The Test Set: 16 Company-Years, Two Versions Each
Each company-year is computed twice. “Reported” uses operating income exactly as filed. “Ex charges” adds back the restructuring and special charges each company itemized in its own filing: DEC’s restructuring charges and a 1994 intangible write-down, Unisys’s special charges (the operating-line portion only), IBM’s 1992–1993 restructuring charges and its 1995–1996 write-offs of purchased in-process research and development from the Lotus and Tivoli acquisitions. HP reported no such items in the summary used. The last column is explained in the next section.
| Company, fiscal year | Revenue change | Flow-through, reported | Flow-through, ex charges | FT points moved by 1 pt of margin |
|---|---|---|---|---|
| HP FY1993 | +23.8% | 12.2% | 12.2% | 5 |
| HP FY1994 | +23.0% | 14.3% | 14.3% | 5 |
| HP FY1995 | +26.1% | 15.6% | 15.6% | 5 |
| HP FY1996 | +21.9% | 2.3% | 2.3% | 6 |
| DEC FY1993 | +3.2% | 431.6% | 90.7% | 33 |
| DEC FY1994 | −6.4% | 191.2% | 26.4% | 15 |
| DEC FY1995 | +2.7% | 581.2% | 162.4% | 38 |
| DEC FY1996 | +5.4% | −20.3% | 45.3% | 19 |
| Unisys 1993 | −9.1% | −1.7% | −1.7% | 10 |
| Unisys 1994 | −0.2% | 3,681.0% | 2,084.5% | 526 |
| Unisys 1995 | +4.0% | −337.8% | −69.9% | 26 |
| Unisys 1996 | +0.4% | 3,154.3% | 152.1% | 226 |
| IBM 1993 | −2.8% | 22.0% | 171.4% | 35 |
| IBM 1994 | +2.1% | 1,021.1% | 351.6% | 48 |
| IBM 1995 | +12.3% | 32.8% | 56.1% | 9 |
| IBM 1996 | +5.6% | 25.1% | −10.0% | 19 |
[Calculation] All ratios are mine, computed from the filed figures in the sources list. The underlying revenue and operating-income series are: HP FY1992–FY1996 revenue $16,410 / $20,317 / $24,991 / $31,519 / $38,420 and earnings from operations $1,404 / $1,879 / $2,549 / $3,568 / $3,726. DEC FY1992–FY1996 revenue $13,931 / $14,371 / $13,451 / $13,813 / $14,563 and operating income −$2,136 / −$237 / −$1,996 / $108 / −$44. Unisys 1992–1996 revenue $6,715.6 / $6,107.1 / $6,095.5 / $6,342.3 / $6,370.5 and operating income $688.2 / $698.7 / $271.7 / −$562.1 / $327.4. IBM 1992–1996 revenue $64,523 / $62,716 / $64,052 / $71,940 / $75,947 and operating income −$8,239 / −$8,637 / $5,005 / $7,591 / $8,596.
Only HP’s four years give stable, comparable readings. Most of the rest are not measurements of anything a board could act on.
Check 1: How Fragile Is the Ratio? One Margin Point Tells You
The last column answers a simple question: if operating profit had come in one percentage point of revenue higher or lower, how many points would flow-through have moved? The arithmetic is this year’s revenue divided by the revenue change. It’s an editorial check I’m proposing here, not a formula from the book, and it doesn’t need a threshold to be useful. It just shows the reader how much weight the ratio can bear.
- HP FY1995: one point of margin on $31,519 of revenue is about $315. Spread over a $6,528 revenue increase, it moves flow-through by about 5 points. A 15.6% reading means roughly 15.6%, give or take a rounding decision.
- Unisys 1994: one point of margin on $6,095.5 is about $61. Spread over an $11.6 revenue change, it moves flow-through by about 526 points. The 3,681% reading isn’t a finding. It’s what happens when you divide a $427 profit drop by a revenue change smaller than a rounding error on a $6 billion business.
[Calculation] In this sample, the readings where one margin point moves flow-through by about ten points or less are HP’s four years, Unisys 1993 and IBM 1995. Those are also the years where revenue moved by roughly 9% or more. Every reading with a revenue change under 7% in either direction moves by 15 points or more per margin point.
[Inference] Put the sensitivity number next to the flow-through number on the dashboard. If one point of margin can swing the ratio by more than the ratio’s own size, the ratio shouldn’t be driving a decision that month.
Check 2: Did Revenue Actually Grow? Two Negatives Make a Positive
The book’s sharpest warning is about a falling top line, and IBM’s 1993 is a clean example of why.
[Fact] IBM’s 1993 revenue fell $1,807, from $64,523 to $62,716. Reported operating income moved from a loss of $8,239 to a loss of $8,637, a $398 decline, after restructuring charges of $11,645 in 1992 and $8,945 in 1993.
[Calculation] Reported flow-through: −$398 ÷ −$1,807 = +22.0%. On a dashboard, that reads as an ordinary, even respectable conversion rate. It is the year IBM recorded an operating loss of more than $8.6 billion. Strip out the charges and operating income fell from $3,406 to $308, which gives +171.4%. Neither number describes leverage. Both are two declines dividing into a positive.
DEC’s fiscal 1994 shows the same trap, and it’s worth walking through the bridge to see what the ratio was hiding.
[Fact] DEC’s FY1994 revenue fell $920, from $14,371 to $13,451. Operating income fell $1,759, from −$237 to −$1,996. The filing’s footnote says FY1994 SG&A includes restructuring charges of $1,206 and a $310 reduction in the carrying value of intangible assets.
[Calculation] Reported flow-through is +191.2%. Ex charges it is +26.4%. That second number looks like a perfectly ordinary reading for a mature business. Here is what the operating-profit bridge shows instead:
| DEC FY1993 → FY1994 operating-profit bridge | $M |
|---|---|
| Operating income, FY1993 | −237 |
| Gross profit lost to lower volume (−$920 revenue at the FY1993 gross margin of 39.9%) | −367 |
| Gross profit lost to a lower gross margin (39.9% → 33.7%) | −834 |
| Lower research and engineering expense ($1,530 → $1,301) | +229 |
| Lower SG&A before charges ($4,447 → $3,718) | +729 |
| Restructuring charges and intangible write-down | −1,516 |
| Operating income, FY1994 | −1,996 |
[Calculation] Gross margin is total revenue less the combined “cost of product sales, service and other revenues” line in DEC’s eleven-year summary. Volume and rate effects are my split of the $1,201 gross-profit decline. Lines reconcile to the reported −$1,759.
[Inference] The bridge tells a board three things the ratio can’t. The biggest operating problem was the gross-margin rate, not the revenue drop. Cost cuts of $958 offset about 80% of the $1,201 gross-profit loss. The charges were larger than either. A “26% flow-through” summary would have hidden all three.
[Inference] The large-decline case deserves one caveat. Unisys’s 1993 revenue fell 9.1% while operating income edged up $10.5, a −1.7% reading that genuinely says costs were cut about as fast as revenue fell. On a big decline with profit holding, the ratio can carry information. But it is then a decremental margin, a different question from “how much of each new dollar converted,” and it should be labeled that way rather than shown in the same column as a growth-year reading.
Check 3: Do the Definitions Agree? When One Adjustment Flips the Sign
A dashboard metric that changes sign depending on which line you start from can’t anchor a decision. Two years in the sample do exactly that.
[Calculation] DEC FY1996: reported flow-through −20.3%, because operating income fell from $108 to −$44 while revenue rose $750. Add back the $492 FY1996 restructuring charge and it becomes +45.3%.
IBM’s 1996 is the more instructive case, because the reported number looks better than the underlying one.
[Fact] IBM’s 1996 revenue rose $4,007 to $75,947 and reported operating income rose $1,005 to $8,596. Operating expenses included purchased in-process R&D of $435 in 1996 (Tivoli) against $1,840 in 1995 (Lotus).
[Calculation] Reported flow-through is +25.1%. Before those write-offs, operating income fell from $9,431 to $9,031, which gives −10.0%. The bridge shows where the reported gain came from:
| IBM 1995 → 1996 operating-profit bridge | $M |
|---|---|
| Operating income, 1995 | 7,591 |
| Gross profit from higher volume (+$4,007 revenue at the 1995 gross margin of 42.2%) | +1,691 |
| Gross profit lost to a lower gross margin (42.2% → 40.2%) | −1,519 |
| Higher SG&A ($16,766 → $16,854) | −88 |
| Higher research, development and engineering ($4,170 → $4,654) | −484 |
| Smaller in-process R&D write-off ($1,840 → $435) | +1,405 |
| Operating income, 1996 | 8,596 |
[Inference] Almost all of the reported $1,005 improvement came from a smaller acquisition write-off. Volume added gross profit, a lower gross margin gave nearly all of it back, and R&D rose. That’s not a verdict on IBM’s 1996. More R&D can be a sound choice, and IBM said currency cut its reported revenue growth from 8.6% to 5.6%. It’s a demonstration that a 25% headline would have pointed a board at the wrong mechanism.
[Calculation] IBM 1995 shows the same problem as a range rather than a sign flip. Reported flow-through is 32.8%. Before the Lotus write-off it is 56.1%. If you also strip the other items IBM itemized inside SG&A (a $382 gain from a 1994 business sale, a $626 workforce and asset charge and a $175 one-time gain in 1995), it is about 67%. Three defensible definitions, a 34-point spread. The book’s own answer is to pick one basis, GAAP, and hold it across every company. Whatever you pick, show the adjustments beside it.
Check 4: Did the Perimeter Change?
Flow-through assumes this year’s revenue and last year’s revenue come from the same business. Acquisitions, disposals and changes from owned to licensed operations break that assumption before any of the arithmetic problems above start.
[Fact] IBM acquired Lotus in July 1995 and Tivoli in 1996, and sold a business in 1994 at a $382 pre-tax gain. Each moves both the numerator and the denominator. The perimeter test is also why the recent Stagnation Assassins Starbucks analysis didn’t lean on consolidated flow-through: the company tied much of its falling consolidated revenue to converting its China business into a licensed joint venture. That walk is in Starbucks’ Oct 29 Results: Traffic, Ticket, and the China Accounting Bridge.
Does Three-Year Smoothing Rescue the Bad Readings?
The book recommends smoothing over three years. Here is the simplest version of that, taking the endpoint-to-endpoint change over three fiscal years. It’s one reasonable implementation, not necessarily the exact averaging the book has in mind.
- [Calculation] HP FY1992→FY1995: 14.3%. FY1993→FY1996: 10.2%. Smoothing works well here. It confirms the annual readings and shows the FY1996 drop as a real change in direction, because operating margin did fall from 11.3% to 9.7%.
- [Calculation] IBM 1993→1996: 130.2% reported, 65.9% before the in-process R&D write-offs. Better than the annual chaos, but the three-year window still starts at a restructuring trough, so the reading mostly measures recovery from 1993, not leverage on new revenue.
- [Calculation] DEC FY1992→FY1995: revenue fell $118 over three years, so the ratio is −1,902% reported and −631% ex charges. FY1993→FY1996: revenue grew $192 over three years and the ratio is 100.5% reported, 356.8% ex charges.
- [Calculation] Unisys 1993→1996: revenue grew $263.4 over three years on a $6 billion base, and the ratio is −141.0%.
[Inference] Smoothing fixes noise. It doesn’t fix a missing denominator. If revenue didn’t grow over three years either, the three-year ratio is still dividing by close to nothing.
The Decision Rule for Your Monthly Pack
This is an editorial checklist drawn from the test above, not a named framework from the book. Run it before flow-through goes on a slide.
- Growth check. Did revenue grow? If it fell, don’t show flow-through. Show the bridge. If you want a ratio for a large decline, label it a decremental margin and keep it out of the flow-through column.
- Fragility check. Divide this period’s revenue by the revenue change. That’s how many flow-through points one point of margin moves. Print it beside the ratio. If it is larger than the reading itself, lead with the bridge.
- Definition check. Compute flow-through on your chosen basis and with itemized charges and gains removed. If the two disagree in sign, or sit far apart, lead with the bridge and show both.
- Perimeter check. If an acquisition, disposal, licensing change or extra fiscal week changed what’s in either year, rebuild both years on the same perimeter or don’t show the ratio.
[Inference] When all four pass, flow-through is a fast, honest read of whether overhead is keeping pace with growth, and HP’s run shows how stable it can be. When any one fails, the bridge is the measurement and the ratio is decoration.
What a Usable Bridge Contains
A bridge only beats the ratio if it splits the change into mechanisms someone can act on. At minimum:
- Gross profit from volume, at last year’s gross margin.
- Gross profit from the margin rate, split further into price, mix and cost where the data allows.
- Each operating-expense line, before charges.
- Each itemized charge, write-off and one-time gain, on its own line.
- Currency translation and perimeter changes, on their own lines.
Every line should reconcile to reported operating income. Anything you can’t explain goes on a line labeled “unexplained,” not folded into the friendliest-sounding bar.
Where This Test Could Be Wrong
Disconfirming evidence. The “ex charges” column depends on what each company chose to call a charge. DEC recorded restructuring charges in fiscal 1990, 1991, 1992, 1994 and 1996. Unisys took special charges in 1991, 1994 and 1995. When “one-time” items recur that often, adding them back flatters the business, and the reported column may be the more honest one. That’s why the checklist asks for both versions rather than declaring one correct.
Alternative explanation. A fair objection is that these four companies in 1993–1996 were an unusually violent sample. IBM’s own 1993 filing described slowing demand, excess capacity and falling mainframe revenue and margins, and three of the four booked restructuring or special charges in the period. In a calmer industry, small revenue changes might come with small profit changes, and flow-through would misbehave less often. That’s possible. The fragility check still applies, though, because it is pure arithmetic: a small denominator magnifies any profit change, large or small.
Sample limits. Four companies, one industry, one period, chosen because their filings were already reconciled, not drawn at random. The percentages here describe this sample only. They aren’t estimates of how often flow-through misleads across companies in general.
Related Analysis
Each of this week’s case studies hit one of these checks. Nintendo’s fiscal 2015 operating profit rose ¥71.2bn while sales fell, so the growth check failed and the piece used a bridge instead: Iwata’s FY2014–2015 Nintendo: Operating Recovery Is Not a Complete CEO Grade. Xerox had no revenue growth to divide by at all: Xerox 2000–2003: Three Bridges Before You Call It a Turnaround. For the IBM years graded against a fixed rubric, read Gerstner’s First Three Years: Grade the Decisions, Not the Legend. To fit these checks into a wider dashboard, see Measuring Transformation Progress: KPIs, Milestones, and Success Metrics.
Limits
This is a reconstruction from public filings, read from outside the companies. Figures are as filed in the documents listed below. IBM’s 1994–1996 figures are from its 1996 10-K, which reclassified 1995 and 1994 to its 1996 presentation; its 1992–1993 figures are from its 1993 and 1994 10-Ks, and the 1994 10-K matches the 1992–1994 revenue and operating income used here. HP and DEC fiscal years don’t match the calendar year. Unisys says its 1996 results include reversals of 1995 restructuring reserves offset by new charges, which aren’t separated in the figures used here. Gross-margin splits are my calculations. No company or executive is graded, and nothing here describes anyone’s motives. This is educational analysis, not investment advice.
Run the Test on Your Own Business
- Book Todd to speak: boards and finance teams can bring this evidence-first approach to turnaround metrics into the room. toddhagopian.com/speaking
- Read the books: Stagnation Assassin, plus the forthcoming 10-Minute Transformation (Koehler Books, February 23, 2027), where the three velocity metrics, flow-through included, are laid out in full. toddhagopian.com/book/
- Free tool: find out whether your business is an ATM or a money pit with the ATM Test Scorecard.
- Want to read 10-Minute Transformation before launch? Join the launch team.
Sources
- Hewlett-Packard Company, Form 10-K for the fiscal year ended October 31, 1996: Selected Financial Data, net revenue and earnings from operations FY1992–FY1996. SEC EDGAR
- Digital Equipment Corporation, Form 10-K for the fiscal year ended June 29, 1996: Eleven-Year Financial Summary (revenue, cost of revenues, research and engineering, SG&A, operating income FY1992–FY1996) and footnote on restructuring charges and the 1994 intangible write-down. SEC EDGAR
- Unisys Corporation, Form 10-K for the year ended December 31, 1996: Six-year summary of selected financial data; MD&A note allocating the 1994 and 1995 special charges by income-statement line. SEC EDGAR
- International Business Machines Corporation, Form 10-K for 1993: revenue and operating income 1991–1993, restructuring charges 1992–1993. SEC EDGAR
- International Business Machines Corporation, Form 10-K for 1994: revenue and operating income 1992–1994; 1994 business-sale gain. SEC EDGAR
- International Business Machines Corporation, Form 10-K for 1995: 1995 SG&A charge and one-time gain; Lotus acquisition. SEC EDGAR
- International Business Machines Corporation, Form 10-K for 1996: Consolidated Statement of Earnings 1994–1996 (gross profit, SG&A, RD&E, purchased in-process R&D, operating income); MD&A on 1996 currency effect. SEC EDGAR
- Todd Hagopian, 10-Minute Transformation, Chapter 2, “The Three Velocity Metrics” (flow-through definition and denominator warning; three-year smoothing; single-basis consistency). Koehler Books, forthcoming February 23, 2027. Paraphrased, not quoted.
Calculations are mine, from the filed figures cited. Percentages are rounded to one decimal place and sensitivity figures to whole points. Educational analysis only.
About the Author
Todd Hagopian is VP of Product Strategy and Marketing and a business transformation author, speaker, and executive. His research is available as SSRN preprints. He is the author of the Turnaround Code Trilogy: The Unfair Advantage, Stagnation Assassin, and 10-Minute Transformation (February 23, 2027). More about Stagnation Assassins: About.
