What Distinguishes Successful R&D Projects
Eleven numbered findings from one multi-firm study of 211 projects — including a null result at initiation that undercuts any gate criterion rewarding crisp early goals, and four findings that reverse sign between lines of business.
What counts as a success here, before any finding
R6 is a 1986 practitioner-journal report on a large multi-firm academic research programme. Its database covered 211 R&D projects across 21 companies in four lines of business — flat steel, industrial chemicals, processed foods and agricultural chemicals — built over the preceding nine years. All are parameters of that one study.
The four-source uncertainty framework the paper is best known for is treated on the sibling page, sources of uncertainty in R&D projects. This page covers the machinery underneath it: how projects were classified, how the analysis was cut, and what each of the eleven numbered findings says.
That conjunctive definition does a great deal of work. A technically excellent project that never earned its keep is a failure here; so is a commercially useful one that missed its technical objective. Approximately 50 percent of the database was judged both technically and commercially successful — 110 successes and 101 failures, both findings of this study.
Two more partitions: major or incremental, product or process
Two further classification rules were applied to every project. Both are checkbox rules with an explicit collapse step — worth copying if you are building a comparable dataset, because the rule takes the judgement call away from the analyst and puts it in a stated procedure.
R6'S TWO CLASSIFICATION RULES, AS PRINTED
| Partition | What the respondent selected from | The collapse rule |
|---|---|---|
| Major / incremental (MI) — one or more categories checked | 1. Totally new product · 2. A major change in a product · 3. An incremental change in a product · 4. Totally new process · 5. A major change in a process · 6. An incremental change in a process | If 1, 2, 4 or 5 is checked → major. If only 3 or 6 is checked → incremental. |
| Expected outcome (OUT) — what the project was intended to do | 1. Establish a new product · 2. Make recognisable changes to an existing product · 3. Establish a new process · 4. Make recognisable changes to an existing process · 5. Other | Only 1 and/or 2 → exclusively product. Only 3 and/or 4 → exclusively process. (1 or 2) and (3 or 4) → product and process. |
Definitions given elsewhere in R6: major = dramatic or radical changes to the basic nature of an existing product or process, or a completely new one; incremental = minor improvements or modifications to existing products or processes.
The resulting distributions are findings of this study. Expected outcome: 208 exclusively product, 81 exclusively process, 128 product and process. Major/incremental: 151 major, 70 incremental.
The analytical design: four partitions, two orders of result
Four design partitions were applied to every project — success/failure, line of business, major/incremental, and expected outcome. Applying all four to the same project set permits an analysis of variance on each relationship between an independent variable and success, with the other partitions available as conditions.
That produces two orders of result, and R6 keeps them separate. Main effects are robust enough to be detected while ignoring line of business and the other partitions. Second-order effects hold only conditional on line of business or expected outcome. Their evidential weight differs, and so do their significance thresholds.
The instrument was substantial: eight questionnaires, four per project and four per firm. That yielded 422 possible project-level observations (two respondents on each of 211 projects); 419 were obtained, of which 417 could be uniquely categorised by expected outcome. Between 35 and 45 individuals per firm supplied data — all parameters of this study.
What this design supports
- Separating effects that survive across lines of business from those that do not
- Distinguishing uncertainty at initiation from realised outcome — instrumented separately
- Findings about the receiving internal user, surveyed on their own form
- Testing whether a relationship holds for major and incremental projects alike
What it cannot support
- Any claim about R&D unit success or failure — R6 states it addresses projects only
- Causal identification; these are analyses of variance on retrospective data
- A population success rate, for the sampling reason above
- Confidence that initiation conditions were reported as they were seen at the time
The main effects — findings 1 to 4
R6 states that six statistically significant main effects were found, and that four of the six measured some dimension of project uncertainty. Three are reported in detail as findings 1 to 3; finding 4 comes from a separate analysis. Two of the six are never described.
MAIN EFFECTS — FINDINGS OF R6, NOT GENERAL RULES
| # | Finding as stated | Direction | Status |
|---|---|---|---|
| 1 | The more experienced the firm was in producing and selling the product, or in using the process, the more likely the project was to succeed | Positive | Main effect, .01 or less |
| 2 | General management involvement, although slight, made a difference: projects were more likely to be coordinated with other business functions, and to succeed | Positive | Main effect, .01 or less |
| 3 | Goal definition at initiation was not significantly related to success or failure; late in the project's life the relationship was | Null at initiation, positive late | Main effect, no significant interactions |
| 4 | Successful projects had significantly greater impact, and made larger contributions to the firm and its business goals, than failures | Positive | From a separate analysis |
All four are results of this one study on its own 211-project database, not general rules about R&D projects.
The practical consequence is a change of measurement, not of ambition. A gate criterion rewarding a crisply written objective at initiation scores a variable this study found unrelated to outcome. A criterion comparing goal clarity now against goal clarity at the last gate scores the variable that was related — a trajectory test, described on the sources of uncertainty page.
Where the answer changes with the line of business — findings 5 to 8
These findings are why R6 refuses to publish a universal success-factor list. Each relationship was tested in all four lines of business, and in every case the answer differed. Two reverse sign outright.
SECOND-ORDER EFFECTS BY LINE OF BUSINESS — SIGNIFICANT AT .07 OR LESS
| # | Relationship tested | Agricultural chemicals | Processed foods | Flat steel | Industrial chemicals |
|---|---|---|---|---|---|
| 5a | Better fit with the firm's established business practices | Predicts success | Predicts success | No relationship | No relationship |
| 5b | R&D unit more experienced with the underlying science area | Predicts success | Predicts success | No relationship | No relationship |
| 6 | Project complexity — cost, time to complete, number of S&T disciplines involved | Successes less complex | No relationship | No relationship | Successes more complex |
| 7 | Expected contribution at initiation | Successes expected to have less impact | Not related | Successes expected to have greater impact | Not related |
| 8 | Cost of doing R&D in the science area (distinct from project cost) | Third variable intervenes — see below | Successes in less costly S&T areas | No relationship | Successes in more costly S&T areas |
Findings of R6 within four named lines of business, at a .07 threshold — weaker than the .01 used for main effects. Not properties of those industries in general.
Finding 8 in agricultural chemicals splits again on the major/incremental partition: major successes sat in more costly S&T areas than major failures, while incremental successes sat in less costly ones. That is a three-way interaction inside a second-order effect, and about as far as this dataset can be pushed.
Where the answer changes with expected outcome — findings 9 to 11
The last three findings condition on what the project was intended to produce. Two carry the study's most repeated percentages — both findings of this one 211-project database.
Findings 9 to 11
Process projects succeeded more often than product projects
Projects expected to result exclusively in new or modified processes were successful in 69 percent of cases, against 48 percent for new or modified products. R6's mechanism: a process project faces technical uncertainty, while a product project must clear both technical and commercial uncertainty to satisfy the conjunctive success rule.
R&D-originated new-product ideas fared worst
New product projects were successful in 56 percent of cases when marketing, distribution, sales and/or the customer first suggested the project — alone or jointly with R&D — against 35 percent when R&D was the sole source. The variable is the origin of the suggestion, not who ran the project.
Process successes showed rapid, large-step technical progress
Successes among new or modified process projects were characterised more by rapid, large-step progress in the underlying science and technology than were failures of those projects. No such difference separated successes from failures among product projects, and no percentages are attached.
What these findings can and cannot carry
R6 is unusually candid about its own limits, and the caveats are load-bearing. Carry them with any figure you quote.
- All data are retrospective. Respondents judged conditions at initiation only after the project had completed and success or failure was known. R6 notes nearly all research-management studies share this limitation.
- Projects and respondents were selected by a single firm-appointed liaison, who also made the success/failure classification.
- Second-order findings use a .07 threshold, considerably weaker than the .01 applied to main effects.
- Several findings reverse sign between lines of business, so no second-order result travels on its own.
- Interpretation is acknowledged as problematic. R6 states its published interpretations came from a joint industry and university assessment process, not from the statistics alone.
- The framework is explicitly not a selection or rejection tool. R6's position is that there are many reasons to undertake risky projects.
Before you quote a figure from this study
- State the success definition it rests on — technical and commercial success, conjunctively
- State the denominator: 211 projects, 21 companies, projects completed before 1982
- Say whether the figure is a main effect or a second-order effect, and give the threshold
- For a second-order figure, name the line of business or outcome type it is conditional on
- Never present the roughly 50 percent success rate as a rate — it is designed into the sample
What the study contradicts
Several findings run against standard practice hard enough to be worth stating as disputes. What they establish is that the conventional position is not universal — not that its opposite is.
CONVENTIONAL POSITION AGAINST R6'S RESULT
| Conventional position | What R6 found in its own data |
|---|---|
| Crisp, widely agreed goals at initiation mark a well-founded project | Goal clarity at initiation did not predict outcome. Its trajectory over the project's life did |
| New products are where the value is | Exclusively-process projects succeeded at 69 percent against 48 percent for product projects |
| Technically originated ideas are the purest form of innovation | R&D as sole source of the first suggestion was the worst-performing origin: 35 against 56 percent |
| Keep projects simple to improve the odds | Complexity had no universal direction — helpful in one line, harmful in another, irrelevant in two |
| Business fit and prior science experience are general success factors | They mattered in the two lines where the customer was least active as a source of ideas, not in the others |
Left column is the conventional position as R6 characterises it. Right column is that study's finding on its own data, not a demonstrated general truth.
One boundary is worth naming to close. R6 is about which projects turn out well; it is not a termination model. Which factors should drive a continue-or-kill decision at a specific moment belongs to a different empirical tradition — see making better project termination decisions and applying termination criteria by project stage. The retrospective judgement R6 depends on is also the raw material of post-project reviews in R&D.
What to carry forward
- Read the success definition before the success rate. A conjunctive technical-and-commercial rule produces a much lower number than a rate computed on developed products.
- Goal clarity at initiation did not predict outcome here; the change in goal clarity over the project's life did. Measure the movement.
- Second-order findings are conditional by construction. A result that reverses between lines of business cannot be lifted into your gate criteria.
- Process projects and commercially originated product ideas outperformed — 69 against 48 percent, and 56 against 35 percent — in one study, on projects completed before 1982.
- The database success rate is designed in by the sampling protocol. Quote it with that attached, or not at all.
- R6 declines to let its own framework be used as a kill screen. These are diagnostic findings, not selection criteria.
Frequently asked questions
Why is this study's success rate so much lower than other published rates?
Because of its definition. A project counted as a success here only if it achieved both technical and commercial success, and the denominator is R&D projects rather than fully developed products. Studies using a looser success criterion or a later denominator report much higher rates, and the difference is definitional rather than empirical.
Does the null result on goal definition mean initial goals do not matter?
It means that in this dataset, how well defined and widely recognised the goals were at initiation was not statistically related to eventual success. The same measure taken late in the project's life was related. The reading R6 supports is that resolving goal uncertainty over the project's life is what separates successes from failures, not starting with perfect clarity.
Can I use the 69 percent process versus 48 percent product figure to shift my portfolio?
Not on its own. It is a finding of one study on 211 projects completed before 1982, conditional on the expected-outcome classification and on that study's conjunctive success rule. It is a reason to ask whether your product projects are carrying commercial uncertainty that your process projects are not, which is a question you can answer with your own data.
Which findings here are the most transferable?
The main effects, because they were detected while ignoring line of business and the other partitions and were tested at a .01 threshold. Finding 3 in particular is recorded as having no significant interactions, meaning it held across all lines of business, all outcome types and both major and incremental projects within this dataset.
Why are there eleven findings when the paper reports six significant main effects?
The eleven are the numbered findings as printed, spanning main effects and second-order effects. Of the six significant main effects the paper says it found, only three are described in detail, plus a fourth from a separate analysis. Two of the six are never described, which is a limitation of the reporting rather than of the study.
How should I treat the counts that do not add up?
Record them as printed and flag the discrepancy. The major and incremental counts sum to 221 against a stated 211 projects, and the expected-outcome counts sum to 417 while being introduced as covering 422 responses. Silently correcting either would misrepresent the source.
References and source attribution
- R6 — why R&D projects succeed or fail. Practitioner-facing report of a large multi-firm academic research programme, in a journal for research management, November–December 1986; 6 printed pages as printed; 2 references. Database of 211 R&D projects across 21 companies in four lines of business, developed over nine years; eight questionnaires totalling 1,177 questions, four administered per project and four per firm. Sections used here: the success/failure, major/incremental and expected-outcome classification rules; the analytical design; the numbered findings; and the stated limitations.
- R7 — most new products do succeed. Practitioner-facing report of a firm-level survey study, in a journal for research management, November–December 1983. Used here only for the contrast in success definitions and denominators.
- The published innovation life-cycle model used to select the four lines of business, and the two prior articles drawn from the same database, are cited within R6 and were not supplied to this library.
- 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 R&D project success factors.
Suggested questions for Ask KEVOS
- Rewrite our gate criteria so they score the change in goal clarity rather than its level at initiation.
- Classify our current portfolio using the major/incremental and expected-outcome rules from this study.
- Which of these eleven findings are main effects and which are conditional on line of business?
- Draft a project initiation field that records who first suggested the project and whether a commercial function was involved.
- Explain why this study's success rate differs from the new-product success rates on the sibling page.
- What would we need to collect to test finding 9 on our own project history?