Core insight
A product that does not pay is not the same as a product worth retiring. The second finding needs a number almost no one has: what goes with the product when it goes, meaning the wider spend of the customers who leave with it.
Why a product's loss is not the decision
The case for retiring a product usually starts with the one number everyone already has: what it loses a year, and what stopping it would free up. Stated alone, that invites a decision to retire rather than a decision.
The other half of the comparison lives in two systems that no report joins. The product's customer list is in billing; what those customers spend on everything else is in the sales ledger. When the two are put side by side, the wider spend can be several times the product's own revenue, and some share of those customers leave entirely when the product goes.
Customers still use it and it does not pay. What does retiring it cost?
Kill criteria: decide the ending before you need it
Kill criteria are the conditions, agreed in advance, under which a product or project will be stopped: a usage floor, a loss ceiling, a date by which a milestone must be met. Set early, they turn a contested ending into a scheduled one. Set late, they are usually written to justify a decision already made.
Good kill criteria name more than the product's own numbers. They should include the customers who would be affected, what those customers buy elsewhere from you, where they would be pointed, and who decides. A criterion that only watches the product's loss will fire on products that are cheaper to keep.
The break-even loss
What keeping the product costs over the years you would keep it, against the one-time cost of ending it plus the wider spend of the customers who would leave with it, less what a replacement earns from those who move. The result is one figure: how much more the product would have to lose a year before retiring it becomes the cheaper choice.
An illustration, from the fictional company used in the Sunset Strategy Pack's worked files: a product installed at 1,800 customers brings in $6.4 million a year and costs $10.6 million to run, losing $4.2 million a year. That gets it onto the agenda. But those customers spend $31.0 million a year with the company on everything else, an estimated 22% would leave with the product, and retiring it would cost $25.8 million. Keeping it is cheaper by $9.5 million.
How to end a product well, and when to wait
The retirements that held were done in the same order: the destination named, the date announced once, and what the product had built kept.
Name the destination on the day
Intuit announced the end of Mint in October 2023 and pointed its roughly 3.6 million active users to Credit Karma in the same announcement.
Announce the date once
Google announced the end of Reader on March 13, 2013, gave a date of July 1, and kept it despite a petition with 154,283 supporters. SAP set end-of-support dates for 2017, 2019 and 2025 before confirming 2027; By the end of 2024 only 39% of its older-software customers had migrated, according to Gartner figures reported by CIO.
Keep what the product built
Apple discontinued the Newton in February 1998 but kept its stake in ARM, bought for a few million dollars and later sold down for about $800 million.
The evidence also shows when to wait. Nokia announced the move away from Symbian, a product that was still paying, in February 2011, before a Windows Phone device of its own had shipped. Over the following years its smartphone share fell from 33% to 3%. Customers price an ending the day they hear it, so announcing before the replacement exists can cost the product's remaining revenue early.
Nine retirements, read through what went with the product
Each case below is drawn from the Stratrix case library, checked against the filings and reporting listed under Sources. The last column says what each one does not prove.
| Product | What was decided | What happened | What it does not establish |
|---|---|---|---|
| Google Reader, 2013 | Retired, citing a loyal but declining following | Shut on July 1, 2013 as announced; never restored | What those users were worth to Google elsewhere |
| Nokia Symbian, 2011 | Ending announced before the replacement shipped | Smartphone share fell from 33% to 3% over the CEO's tenure; the phone business was sold to Microsoft in 2013 | That Symbian could have been saved; decline was already under way |
| Intuit Mint, 2023 | Retired, with users pointed to Credit Karma | Date moved once, to March 2024; Credit Karma revenue reached $2.3 billion in fiscal 2025 | How many Mint users moved rather than left |
| SAP's older ERP software | Support end dates set and moved, then confirmed for 2027 | Cloud revenue up 25% in 2024; about 39% of those customers had migrated by end of 2024 (Gartner, via CIO) | That the deadline caused the cloud growth |
| Atlassian Data Center, 2025 | End of life announced for March 2029, three years after saying there were no plans to end it | Too early to tell | Any outcome; none is yet on record |
| Sears catalog, 1993 | Discontinued alongside 113 store closures after a $3.9 billion loss | The fulfillment network and customer lists were not kept; e-commerce was rebuilt from scratch in 1997 | That keeping the catalog would have made Sears an e-commerce winner |
| Apple Newton, 1998 | Discontinued, but the ARM stake was kept | The stake was later sold down for about $800 million | What Newton users were worth to Apple beyond it |
| Ford sedans, 2018 | Phased out five sedans, still 17.5% of US volume | The planned crossover replacement was canceled within months; car sales fell 29.5% in Q3 2019 | Where the sedan buyers went |
| CVS tobacco, 2014 | Stopped selling tobacco, about $2 billion of revenue, on purpose | Front-store sales fell 4.5% in Q3 2014, about 480 basis points of it tobacco and the related basket; pharmacy sales rose 4.8% | That the lost basket was recovered elsewhere |
Running the decision in two weeks
A sequence for a product or portfolio team
Write the kill criteria down
If they do not exist, write them now and agree who decides, before anyone argues about this product.
Join the customer list to the sales ledger
For every customer on the product, what they spend with you on everything else. It is usually an afternoon's work and has usually never been done.
Estimate the share who would leave
It cannot be surveyed without announcing the ending, so treat it as a range and test the decision across it.
Price the ending
Migration, support, contracts, write-offs and the replacement's cost, once.
Find the break-even loss
How much more the product would have to lose a year before retiring is cheaper. Compare it with the trend of the actual loss.
If you retire it, announce once
Name the destination and the date together, and keep anything the product built that is worth more than the product.
What this method cannot tell you
It does not forecast the share of customers who would leave; that is your estimate, which is why it should be tested across a range. It does not price what announcing the ending does to a product that still pays before a replacement exists.
And it says nothing about what a product carries beyond its margin: an approval, a reference customer, or the reason a customer chose you in the first place. Those belong in the decision as their own lines.
Watch these decisions
Short videos built from the same Decision Records as this page.
Questions
What are kill criteria?
Kill criteria are conditions agreed in advance under which a product, feature or project will be stopped, such as a usage floor, a loss ceiling or a missed milestone by a set date. Setting them early turns a contested ending into a scheduled one.
When should you discontinue a product?
When retiring it costs less than keeping it, counting not just the product's own loss but the wider spend of the customers who would leave with it. A product losing money can still be cheaper to keep than to retire.
Our product loses money. Should we retire it?
Not on that fact alone. In the Sunset Strategy Pack's fictional worked case, a product losing $4.2 million a year is cheaper to keep by $9.5 million, because retiring it would cost $25.8 million, mostly in lost spend from customers who leave with it.
How should you announce a product sunset?
Once, with the destination and the date together, as Intuit did when it pointed Mint users to Credit Karma. Dates that move teach customers that the next date can move too; SAP set and moved several end-of-support dates before confirming 2027.
Does this apply to retiring a feature rather than a whole product?
Yes, if the feature has customers who can be listed and an upkeep that can be separated. The arithmetic is the same: what those customers spend with you elsewhere, what share would leave, and what the ending costs once.
Continue learning
Sources & further reading
- Wikipedia (citing Google's announcement) (2024). Google Reader.
- Change.org (2013). Keep Google Reader Running.
- Nokia (Form 6-K) (2011). Nokia and Microsoft sign definitive agreement ahead of schedule.
- Microsoft Corp. (2013). Microsoft to Acquire Nokia's Devices & Services Business (Form 8-K).
- Wikipedia (2024). Stephen Elop.Smartphone share, losses and stock decline over the tenure.
- Wikipedia (citing Bloomberg) (2024). Intuit Mint.
- Intuit Inc. (2025). Form 10-K, fiscal 2025.
- SAP SE (2025). SAP Announces Q4 and FY 2024 Results.
- CIO (citing Gartner) (2025). Nearly half of SAP ECC customers may stick with legacy ERP beyond 2027.
- Support Revolution (2024). SAP Support Deadline Extended Yet Again to 2027.
- Atlassian (2025). Ascend to the cloud: The next chapter for Atlassian and our customers.
- Atlassian Community (2022). What's new and what's ahead for Data Center: June 2022.
- Wikipedia (2024). Sears (1990s).
- Fortune (2017). Sears Could've Been Amazon. Here's How It Blew Its Chances.
- MacRumors (reproducing Apple's press release) (2023). Apple Discontinued the Newton 25 Years Ago Today.
- Tedium (2018). ARM Holdings history: from Acorn to giant tree.
- The Motley Fool (2018). Ford Dumps Its U.S. Sedan Lineup: Is It Making a Mistake?.
- NPR (2018). Ford Says Despite Trump's Tweet, Focus Active Won't Be Produced In U.S..
- Ford Motor Credit Co. (2019). Q3 2019 sales press release (Form 8-K).
- CVS Caremark (2014). CVS Caremark to Stop Selling Tobacco at all CVS/pharmacy Locations.
- CVS Health (2014). Q3 2014 earnings release (Form 8-K).
More to learn
Other foundational guides worth reading.
Make or Buy: How to Decide What Your Company Must Own
You can outsource anything whose interface you can specify, and nothing you cannot. The test is whether a supplier who has never seen the work could quote from what you hold today; the number is the probability of having to buy the capability back.
Mergers & AcquisitionsShould You Split the Company? How to Test a Spin-Off or Divestiture
A separation changes nothing operational on the day it happens. The prize is a re-rating that nobody can check in advance, and the cost, a one-time bill plus a corporate center that now has to exist twice, can be counted to the dollar.
Pricing StrategyPricing Model Change: How to Move to Subscription or Usage Pricing
A pricing model change moves up to four layers at once: the unit you charge for, the shape of the cash, the buyer who signs, and the promise the customer relies on. Name which ones you are moving, size the trough against carrying on unchanged, and run the case again with the volume uplift set to zero.
Mergers & AcquisitionsPost-Merger Integration: How to Tell Whether the Deal Is Paying for Itself
A synergy reported and a synergy in the ledger are different findings. An integration is working only when the promise reaches the accounts, net of the revenue the combination cost, before the year it stops paying for the premium.
Competitive StrategyHow to Respond to a New Competitor: Price Its Money, Not Its Product
Size a new competitor by its money, not its demo. Divide what it has for your market by what it costs it to take one of your customers, price what those customers would have earned you, and answer only if answering on every customer it reaches costs less than that.
Change & TransformationCost Reduction Program: What a Cut Buys and What Reversing It Costs
Every cost cut buys a smaller capability, and only the saving reaches the paper. For each cut, divide what rebuilding the capability would cost by the annual saving to get the reversal price in years, and divide the saving by the profit running through what was cut to see how small a loss would cancel it.
See this framework in the wild.
Follow it across the companies and lenses where it actually shaped the strategy.