Measuring New Product Success Rates
Move the denominator from product ideas to fully developed products and the reported rate shifts further than any real difference in performance could. R7 is a paper about what that does to a statistic everybody quotes.
The statistic under dispute, and how R7 sources it
R7 is a 1983 practitioner-journal report of a firm-level survey, adapted from a longer academic marketing-management article published the year before. It opens on a controversy rather than a framework: the commonly accepted claim that new products fail at rates somewhere between 50 and 90 percent.
THE DISPUTED FIGURE AND ITS SOURCING, AS R7 REPORTS IT
| Claim | Provenance class | What R7 says about it |
|---|---|---|
| Failure rates as high as 90 percent, or even 50 percent | Externally cited figure — the received view in the literature, not R7's data | Characterised as the commonly accepted position, and disputed as an artifact of sampling, measurement and definition |
| Most reported failure-rate values, now accepted as fact, were originally based on speculation, on personal claims, or on studies of questionable scientific merit | Externally cited — the conclusion of a prior intensive literature review, quoted by R7 | Used as the ground for the dispute. R7 is not itself evidence for this claim |
| Of the commonly cited references, only seven reported results from empirical studies involving a reasonable sample of firms — and all seven indicate failure rates far below 50 to 90 percent | Externally cited count, quoted by R7 | The core of the argument: the received figure has a thin empirical base |
| One frequently cited data source covers strictly retail goods | Externally cited, quoted as an example | Offered as an instance of the industry-mismatch problem in quoted statistics |
| A study by a business research organisation reports a 61.5 percent success rate | Externally cited figure, quoted by R7 | Noted as remarkably close to R7's own result |
Every row here is a figure R7 quotes from elsewhere, not a result of its own survey. R7 is not evidence for any of them; it is evidence that they were in circulation and how they were sourced.
The figures R7 offers instead — one study's findings
Every number in this section is a result of R7's own survey of 103 industrial firms, with an effective response rate of 69 percent, drawn from a standing sample of almost 200 industrial firms known to be active in new products, running since 1974. None of them is a norm, a benchmark or a rate for new products in general.
The composition matters more than the headline. Of the 41 percent of developed products that were not successful, more than half were the result of a kill decision made before launch. Only about 19 percent actually reached the market and failed there. A kill and a commercial failure are different managerial events with different costs, and R7 counts them separately.
Why the definitions decide the answer before data collection
This is the part of R7 worth carrying into your own work regardless of what you think of its numbers. Three definitional choices set the answer, and studies that make them differently are not in disagreement — they are measuring different quantities.
R7's operational definitions
- New product (firm's point of view)
- A product that represented a significant departure from the company's current products in terms of either markets or the product itself, even though the product may not be new to the market. Minor modifications and style changes are ruled out.
- New product (developmental qualifier)
- One that had been fully developed and ready for commercialisation, although not necessarily launched yet. This is the denominator the success rate is computed on.
- Success
- Commercial success — the degree to which a product exceeds, or falls short of, the minimum acceptable profitability for that type of investment.
- Kill
- A decision to stop the product before launch, after development. Counted separately from commercial failure.
Against each of those, R7 sets out the alternatives that produce different statistics. On what counts as a new product, definitions range from radical innovations totally new to the market at one extreme to improvements to the existing product line at the other. On the denominator, published rates are computed on anything from a product idea through to a commercialised product. On success, the alternatives include technical or engineering success, the fact that the product is still on the market, and whether the product met management expectations.
The same mechanism explains an apparent contradiction elsewhere in this library. A study reporting that roughly half of R&D projects succeeded, using a rule that required both technical and commercial success on a denominator of projects rather than developed products, is not in conflict with a 59 percent rate on developed products — see what distinguishes successful R&D projects. Two definitions, two denominators, two quantities.
Before you quote or accept any success or failure rate
- Establish the denominator — ideas, projects, developed products, or launched products
- Establish the success criterion — commercial, technical, survival on the market, or expectations met
- Establish what counts as new — radical innovation, new to the firm, or line improvement
- Ask whether kills before launch are counted as failures or reported separately
- Ask what population was sampled, and whether it was selected for being active in new products
- Ask whether success was self-rated by the respondent or measured independently
The measurement model: one input set, three output constructs
R7's structural move — and the reason its findings hang together — is refusing to treat new-product performance as one thing. Output is split into three constructs that are measured separately, and they behave differently.
R7'S CONCEPTUAL MODEL OF THE NEW-PRODUCT PROCESS
| Class | Variable | How it was measured |
|---|---|---|
| Input | Annual R&D spending | As a percent of sales, to standardise for firm size |
| Input | Firm resources and skills in eight areas | Self-rated relative to domestic competitors on a −5 to +5 scale (+5 much stronger, −5 much weaker). The eight: financial; R&D; engineering; marketing research; management; production; salesforce and distribution; advertising and promotion |
| Output — effectiveness | Success rate, failure rate, kill rate, plus an overall programme performance rating | Rates as percentages; the programme rating self-rated on the same −5 to +5 scale |
| Output — magnitude | Quantity of output | Percent of corporate sales derived from new products introduced over the past five years |
| Output — efficiency | Return on the R&D input | New-product sales generated per dollar of R&D spending (formula below) |
| Moderator | Firm characteristics | Industry type; firm size by annual sales; ownership (domestic, foreign multinational, domestic multinational) |
Model structure and instrument design from R7. The eight resource areas and the −5 to +5 scale are that study's instrument, not a standard assessment framework.
Keeping effectiveness, magnitude and efficiency apart is what allows R7 to say that something can raise one and leave another untouched. That separation drives the paper's conclusions about R&D spending and about which corporate strengths matter — treated on the sibling page, four myths about new product failure.
The efficiency formula, and what R7 says it is not
Efficiency is computed as E = (S_NP × S_V) / C_RD, where S_NP is the percentage of sales contributed by new products introduced in the last five years, S_V is annual company sales, and C_RD is annual R&D spending. The output is dollars of new-product sales per year per dollar of annual R&D spending.
In R7's sample the median firm obtained $2.67 or more of new-product sales each year per dollar of annual R&D spending; the mean was $8.30, on a standard deviation of 20.00. The gap between mean and median is the signature of a heavily right-skewed distribution, and the same pattern appears in the sample's R&D spending — a mean of 3.22 percent of sales against a median of 1.67 percent. All are findings of this study.
What the ratio measures
- One year of new-product sales against one year of R&D spending
- Sales, not profit
- A ratio that is comparable across firms of different sizes
- An input-to-output relationship, with no causal claim attached
What R7 says it should have been
- Profits rather than sales
- Taken over the product's life, not one year
- Discounted to net present value
- Therefore: the true payoff is higher than the computed ratio
R7 is explicit about why it did not do the harder computation: differing accounting practices across firms, the difficulty of predicting sales over a product's life, and the choice of an appropriate discount rate together made a net present value calculation impractical. That is a candid statement of a measurement limit, and it is the same set of obstacles that shapes the techniques discussed in the financial frame for R&D management.
What it would take to generalise any of these figures
R7 disputes a received statistic on the grounds of sampling, measurement and non-comparable operational definitions. That critique applies to its own numbers with equal force, and the paper's value is diminished rather than protected by treating 59 percent as the new true rate.
WHAT WOULD HAVE TO HOLD BEFORE R7'S RATES TRAVELLED
| Condition in R7 | Why it limits transfer | What generalising would require |
|---|---|---|
| Denominator is fully developed products ready for commercialisation | The rate is not transferable across denominators; a rate on ideas or on projects is a different quantity | The comparison study must draw its denominator at the same stage |
| Success defined as commercial success against minimum acceptable profitability for that type of investment | Technical success, market survival and expectations-met produce different rates on the same products | An identical success criterion, and a stated profitability threshold |
| Unit of analysis is the firm and its programme, not the product or project | Nothing here supports a project-level go or kill decision | Project-level data if project-level conclusions are wanted |
| Firms drawn from a standing panel known to be active in new products | A selected population, not a cross-section of industry | A sampling frame that does not select on the behaviour being measured |
| Programme performance and resource strengths are self-rated on a −5 to +5 scale relative to self-nominated competitors | Self-assessment against a self-chosen comparison set | An independent performance measure, or at minimum a fixed comparison group |
| Cross-sectional correlations and analyses of variance | No causal identification is claimed anywhere in the paper | A design capable of supporting causal claims, if causal claims are to be made |
| Survey conducted for a 1983 publication | Market conditions, product cycles and accounting practice have moved | Contemporary data, and a statement of what has changed |
Conditions as stated or evident in R7. The right column is this library's synthesis of what generalisation would demand; the source does not set out such a test.
One further boundary is worth stating. R7 measures programmes, not projects. It cannot tell you whether to continue or stop a particular development, which is a different question with a different literature — see making better project termination decisions — and it says nothing about whether the uncertainty in a given project is being resolved, which is the subject of sources of uncertainty in R&D projects.
What to carry forward
- The 50 to 90 percent failure figure is an externally cited claim that R7 disputes on the grounds of its sourcing. R7 is evidence about that figure's provenance, not evidence for it.
- R7's own rates — about 59 percent success, 22 percent killed before launch, 19 percent failed after launch — are findings of one 1983 survey of 103 industrial firms, on very large standard deviations.
- Count kills separately from failures. More than half of the unsuccessful outcomes in this study were pre-launch kills, which is a different managerial event from a market failure.
- Three definitional questions decide any published rate: what counts as new, where the denominator is drawn, and what counts as success. Ask all three before you accept a number.
- Separate effectiveness, magnitude and efficiency. They move independently, and a measure that fuses them will hide the thing you wanted to know.
- The efficiency ratio is one year of sales over one year of spending. R7 states plainly that the true payoff is higher, and why it could not compute it.
Frequently asked questions
So do most new products succeed or not?
In R7's sample of 103 industrial firms, about 59 percent of fully developed products succeeded commercially. That is a finding about those firms, on that denominator, using that success definition. It is a serious challenge to the received failure figure, but it is not a general rate for new products, and R7's own critique of sampling and definition applies to it too.
Why does the paper count kills separately from failures?
Because they are different events. A kill is a decision made before launch, after development; a commercial failure means the product reached the market and did not meet the profitability threshold. In R7's data more than half of the unsuccessful outcomes were kills, so folding them together would nearly double the apparent market-failure rate.
Can I compare our internal success rate to the 59 percent figure?
Only if you build yours the same way: fully developed products ready for commercialisation as the denominator, commercial success against a minimum acceptable profitability threshold as the criterion, and kills reported separately. Change any one of those and the comparison is meaningless.
What is the efficiency ratio actually good for?
It gives a size-comparable reading of new-product sales generated per dollar of R&D spending in a year. R7 is explicit that it understates the true payoff, because the correct quantity would be profits over the product's life discounted to present value, which the paper could not compute. Treat it as an indicator, not a return.
Why is a self-rated programme performance score acceptable in a study like this?
It is a practical compromise rather than a strength, and R7's limitations reflect that. The resource strengths and the overall programme rating are self-assessments on a −5 to +5 scale relative to competitors the respondent nominates. Any finding that leans on those measures inherits the bias, which is one reason the effectiveness rates and the ratings should be read as separate constructs.
Where are the findings about R&D spending and marketing strength?
On the sibling page. R7 organises those results as four myths about new-product failure — covering firm uniqueness, the effect of R&D spending, and whether technological strength determines programme success. This page covers the disputed statistic, the measurement model and the definitions that underpin all of them.
References and source attribution
- R7 — most new products do succeed. Practitioner-facing report of a firm-level survey study in a journal for research management, November–December 1983; 6 printed pages; 13 references; four numbered tables. Adapted with permission from a longer academic marketing-management article published the previous year. Sample: 103 industrial firms, effective response rate 69 percent, drawn from a standing sample of almost 200 industrial firms active in new products maintained since 1974. Sections used here: the measurement model, the efficiency formula, the operational definitions, and the classified quantitative claims.
- The prior intensive literature review of failure-rate sources, the seven empirical studies it identified, the retail-goods data source and the business research organisation's 61.5 percent success figure are all cited within R7 and were not supplied to this library. They are recorded here as R7 describes them.
- R6 — why R&D projects succeed or fail. Practitioner-facing report of a multi-firm research programme, November–December 1986. Used here only for the contrast in success definitions and denominators.
- Eleven copyrighted journal articles on R&D project management, supplied as a reading set for a literature review and profiled for this library. Front matter, abstracts, framework sections, tables and figures were read; article bodies were not reproduced, and all content here is paraphrase. The set is a reading list, not a systematic or representative survey of the field.
- Supplied teaching source for this library (research methods and research process materials). Used here for page conventions and voice only; it does not treat new-product success measurement.
Suggested questions for Ask KEVOS
- Draft the three definitional questions to put to any success rate quoted in a business case.
- Set up an internal measurement of success, kill and failure rates using this study's definitions.
- Explain why two studies of the same firms can report success rates forty points apart.
- Compute our new-product sales to R&D spending ratio and list what it does not capture.
- What would we need to change about our data before comparing ourselves to a published success rate?
- Summarise the provenance of the 50 to 90 percent failure statistic for a sceptical executive.