R&D Project Management Quick Reference
One page to keep open while you work. Eleven papers reduced to their frameworks, criteria sets, formulas and classifications, each labelled by what kind of claim it is and linked to the page that treats it properly.
How to read what follows
The four labels used throughout
- Study finding
- A result of that one paper's own data. It describes the sample it was measured on and nothing else. Never a norm, a target or a benchmark.
- Worked example
- A demonstration figure, hypothetical or drawn from one reported case, used to show a method operating. Never a benchmark.
- Externally cited
- A number the paper quotes from somewhere else. The paper is not itself evidence for it, and the original was not supplied to this library.
- Regulatory value
- A binding threshold or ceiling of one framework, in one jurisdiction, for a period ending no later than 31 December 2006. Historical, not current law.
The eleven papers at a glance
THE SET, BY DOMINANT MODE
| ID | Subject | Dominant mode | Treated in depth on |
|---|---|---|---|
| R1 | Choosing R&D projects as a choice of what information to acquire (1991) | Proposes a formal model | Selecting R&D projects: an informational view |
| R2 | Valuing R&D as a call option on implementation (2000) | Proposes a valuation model | Real options valuation of R&D projects |
| R3 | Treating R&D as an investment and appraising it financially (1984) | Synthesis of existing tools | The financial frame for R&D management |
| R4 | Why compressing a schedule raises cost more than proportionally (1989) | Proposes a curve and a metric | Why costs increase when projects accelerate |
| R5 | Stage-specific discriminant analysis for continue-or-kill decisions (2002) | Proposes a method | Making better project termination decisions |
| R6 | What separates successful from failed R&D projects (1986) | Reports empirical findings | What distinguishes successful R&D projects |
| R7 | Measuring new-product success rates properly (1983) | Reports empirical findings | Measuring new product success rates |
| R8 | Post-project reviews: incidence, barriers, maturity (2003) | Reports empirical findings, and proposes a maturity model | Post-project reviews in R&D |
| R9 | Decision-analysis tools applied across strategy, selection and execution (1994) | Synthesis of existing tools | R&D project evaluation tools |
| R10 | A decade of regulatory practice on aid to large R&D projects (2006) | Regulatory analysis | State aid assessment of large R&D projects |
| R11 | One corporate R&D centre's transformation (2006) | Single-organisation case study | Transforming a corporate R&D centre |
Several papers do more than one thing; the column names the dominant mode. Across this set: four propose a named tool or model, three report empirical findings, two synthesise existing tools, one is a case study, one is regulatory analysis.
Selecting and valuing projects
SELECTION AND VALUATION RULES, WITH THEIR SOURCE AND CLASS
| Rule or formula | Statement | Source and class |
|---|---|---|
| Dominance rule | Prefer project A to B if A yields superior information at at least the same success rate, or equivalent information at a strictly larger probability or arrival rate | R1 · conditional result of a stated model, not an empirical finding |
| Project value, static | V(G,c) = π(G,c)·W(G) − c — expected reward times success probability, less cost | R1 · formal definition |
| Free disposal of information | Never penalise a project for producing extra information; reward is non-decreasing in the information acquired | R1 · formal result |
| Funding test | Fund if potential commercial value × probability of success > expected R&D cost to completion | R9 · proposed decision rule |
| Shareholder value | incremental expected NPV − expected R&D cost to completion | R9 · proposed measure |
| Expected return (productivity index) | potential commercial value × probability of success ÷ R&D cost. Rank on this when resources are constrained; fund all positive-value projects when they are not | R9 · proposed decision rule |
| Value at R&D completion | V* = max[0, R − K] — implement only if the NPV of revenues exceeds the NPV of production and marketing costs | R2 · proposed model |
| Value of the R&D itself | V = e^(−kt) · ∫₀^∞ x·f_X(x) dx — the truncated expectation of net cash flows, discounted back over the R&D phase | R2 · proposed model |
| Variance addition | With independent normal revenues and costs, σ_x² = σ_r² + σ_k², so cost uncertainty increases option value | R2 · formal consequence of the model's assumptions |
| Equivalence at the cutoff | A benefit-to-cost ratio of one corresponds to a net present value of zero, so the discounted methods agree near the hurdle rate | R3 · general rule as printed |
| Payback reciprocal | The reciprocal of years payback is a rough estimate of internal rate of return expressed as a decimal | R3 · rule of thumb read off one division's chart |
R1 contributes structure, not numbers: the paper contains no empirical figures at all, and its propositions are conditional on its stated model.
R3'S FIVE OBJECTIVE ASSESSMENT TECHNIQUES
| Technique | Definition as given | Stated limitation |
|---|---|---|
| Sales-to-development ratio | Uses sales as a proxy for earnings | A convenient early screen only; does not communicate to financial directors in their own language |
| Cash flow payback | Time required to recover an investment out of future earnings | Ignores product life after payback and slow market introduction; liable to subjective interpretation |
| Net present value | Algebraic sum of all cash flows discounted at the company's hurdle rate | No good method for ranking projects under capital rationing |
| Benefit-to-cost ratio | Money value of returns divided by project cost | Proper use requires discounting to an agreed hurdle rate |
| Internal rate of return | The discount rate at which total cash flow discounts to a present value of zero | Offered as the good tool, because capital expenditure procedures already use it |
R3's own empirical support is two regressions from a single division's single year of projects: internal rate of return = 13.9 + 1.9 × sales-to-development ratio (r = 0.969) and = 119.6 − 24.2 × years payback (r = 0.872). Study findings of that one division. The coefficients are not industry constants.
What a schedule costs
R4 — THE FOUR MECHANISMS THAT MAKE THE TIME–COST CURVE CONVEX
| Mechanism | How it works |
|---|---|
| Information dependency | R&D is heuristic; each step builds on information from previous tasks. Compression forces overlap, so each task begins with less information, producing mistakes and rework — and the penalty worsens as compression increases |
| Diminishing returns to added people | Communication and training burdens grow. With pairwise communication between groups the burden scales as n(n−1)/2. New staff absorb experienced members' time, and the faster you add them the worse the ratio |
| Parallel search | Serial search is cost-minimising: try approaches in sequence and stop when one works. Buying time means running approaches concurrently and paying for ones you would never have needed |
| Critical-path crashing | The cheapest task to accelerate is taken first, then the next cheapest. Further compression exercises progressively more expensive options, and the network grows denser as tasks overlap |
Proposed causal theory, not measured mechanisms.
R4 — THE ELASTICITY TABLE
| Project type | Percentage cost increase per 1 percent duration reduction | Class |
|---|---|---|
| Hardware projects | 1.75 | Study finding of one econometric study, recomputed into a common metric by R4 |
| Software projects, first source | 0.88 | Study finding of one study, recomputed |
| Software projects, second source | 2.00 | Study finding of one study, recomputed |
| The synthesised headline | 1 to 2 percent | A general rule R4 derives from those three point estimates — not a measured constant |
Every value is computed at one place on the curve: about 10 percent above the minimum possible completion time. Penalties are greater nearer the minimum and smaller at long durations. The two software estimates sit on both sides of the hardware figure, so no blanket claim about software's compressibility is supported.
- Five factors shape the curve. Steeper for projects near the state of the art, for larger firms, for firms with less relevant experience and for large-scale projects; shallower where professional work content is high — but only in the wages-and-salaries sense. Proposed factors, with mechanisms given and no measurement.
- The accountant's number is too low. R4 excludes the loss to other projects when talent is pulled away, and says explicitly that this would not appear in an accounting rendering of acceleration cost.
- Worked example. A five-year, $100 million project accelerated by six months is a 10 percent duration reduction, implying a 10 to 20 percent cost increase — on the order of $10 to $20 million. A hypothetical, not a case.
Uncertainty, success and the decision to stop
R6 — THE FOUR SOURCES OF PROJECT UNCERTAINTY
| Success condition | Corresponding uncertainty |
|---|---|
| A relevant business need, problem or opportunity has been clearly identified | Uncertainty about the relevance of the business objective |
| An appropriate scientific or technical approach has been matched to that need | Uncertainty about the fit between technical and business objectives |
| The project results can be transferred to an internal user | Uncertainty about transfer to an internal user |
| The internal user can produce, market, distribute and sell the result | Uncertainty about commercialisation |
Evaluate the project separately against each. The continuation test is whether uncertainty is actually falling over the project's life. R6 states the framework is a diagnostic, not a selection or rejection tool.
R6 AND R5 — FINDINGS AND CRITERIA
| Item | Content | Class |
|---|---|---|
| Goal clarity at initiation | How well defined and widely recognised a project's goals were at initiation was not significantly related to eventual success. Late in the project life the relationship was significant — successful projects resolved goal uncertainty, failures did not | R6 · study finding, 211 projects in 21 companies across 4 lines of business |
| Process against product | 69 percent of projects expected to result exclusively in new or modified processes succeeded, against 48 percent of product projects | R6 · study finding |
| Who suggested it | New-product projects succeeded 56 percent of the time when marketing, distribution, sales or the customer first suggested them, against 35 percent when R&D was the sole source | R6 · study finding |
| Success definition | A project counted as a success only if it achieved both technical and commercial success | R6 · classification rule |
| Twelve discriminating variables | Reduced from 41 candidate variables in 6 categories, by stepwise discriminant analysis run separately at three evaluation points in the development stage | R5 · method |
| The top discriminator | Priority placed on product quality relative to competitors: coefficient 0.676 at the initial stage (highest), 0.45 at the middle stage (highest), 0.269 at the final stage, ranking seventh of eleven | R5 · study finding, 135 valid questionnaires covering 217 projects across 17 industries |
| Where forecasts rank | Expected probability of commercial success ranked 10th of 12; expected probability of technical success ranked 3rd. Resources and budget variables did not survive at all | R5 · study finding |
| The comprehensive risk model | A derived measure of a project's comprehensive risk, claimed to avoid setting a threshold per variable | R5 · proposed method — the mathematics are not printed in the article |
Measurement and post-project learning
R7 — THE DISPUTED STATISTIC AND WHAT THE STUDY OFFERS INSTEAD
| Figure | Value | Class |
|---|---|---|
| The received new-product failure rate the paper disputes | 50 to 90 percent | Externally cited · characterised by R7 as originating in speculation, personal claims or studies of questionable merit |
| Success rate for fully developed products ready for commercialisation | 59.37 percent, standard deviation 24.95 points | Study finding · 103 firms |
| Post-launch commercial failure | 18.71 percent | Study finding |
| Pre-launch kill | 21.92 percent | Study finding |
| Efficiency of R&D spend | Mean $8.30 and median $2.67 of new-product sales per year per dollar of annual R&D spending | Study finding |
| Diminishing returns | $29.55 of new-product sales per R&D dollar in the lowest spending band against $0.91 in the highest | Study finding |
| Threshold effect | New-product output begins to rise only above roughly 2 percent of sales spent on R&D | Study finding |
R7's efficiency formula is E = (S_NP × S_V) ÷ C_RD — percentage of sales from new products introduced in the last five years, times annual company sales, divided by annual R&D spending. The paper states the true payoff is higher, because the correct quantity is profit over the product's life discounted to net present value, which it could not compute.
R7's transferable contribution is definitional, not numerical. Before accepting any published success or failure rate, establish what counts as a new product, at what stage the denominator is drawn, and what counts as success. A rate computed on product ideas will always look bad; a rate computed on fully developed products ready for commercialisation is a different quantity. That distinction is what makes R6's roughly 50 percent and R7's 59 percent non-comparable — see four myths about new product failure.
R8 — BARRIERS, MATURITY LEVELS AND INCIDENCE
| Element | Content | Class |
|---|---|---|
| Definition | A formal review of the project that examines the lessons that may be learned and used to benefit future projects | Definition given |
| Four barriers | Psychological (inability to reflect; memory bias) · team-based (reluctance to blame; poor internal communication) · knowledge-utilisation (difficult to generalise; tacitness of process knowledge) · managerial (time constraints; bureaucratic overhead) | Proposed framework · only one of the four clusters is procedural |
| Five maturity levels | 1 Initial, ad hoc reviews without guidelines · 2 Repeatable, guidelines for comparable reviews · 3 Defined, standardised reviews with identified output · 4 Managed, actionable failure tolerance and quantified goals · 5 Optimizing, the review practice is itself reviewed | Proposed model, explicitly adapted from a software capability maturity model |
| Incidence | 80 percent of R&D projects were not reviewed at all after completion; on average 19.4 percent of projects respondents had worked on were post-reviewed | Study finding · 63 respondents from more than 40 companies |
| Guidelines | 55.6 percent said their companies had established no formal guidelines; 6.3 percent applied sound and consistent criteria to every review | Study finding |
| How lessons actually move | Individuals moving to new projects, 52.4 percent, ahead of written documentation, 39.7 percent | Study finding |
| Corroboration | A benchmarking study of 79 highly regarded R&D organisations found fewer than a quarter made full use of post-project reviews | Externally cited |
R8 states its own numbers may be optimistically high: several managers claiming the top maturity level were found to be counting singular, non-repeatable learning events. The sample was self-selected attendees at two executive training events.
R8's six rules for conducting a review
- Run it like a mini-project: set a goal, allow divergence at the start, then apply discipline toward a tangible output.
- Use a trained, independent facilitator so project members can focus on the results.
- Require pre-review preparation, including the most unusual or surprising observation from the project.
- Prefer offsite; if held on site, cap it at half a day and split it into two meetings if needed.
- Invite stakeholders of current and selected future projects — including customers, marketing and sales, and the project administrator.
- Produce a summary document of concrete conclusions, warnings and recommendations, with responsibilities assigned.
Portfolio, decision quality and organisation
R9 — THE DIAGNOSTIC, THE TOOLS AND THE DISPLAYS
| Framework | Content |
|---|---|
| Six dimensions of decision quality | Frame · Alternatives · Information (including the appropriate range of uncertainty) · Values (time preference, risk preference, non-financial objectives) · Logic · Commitment. The sixth is organisational, not analytical: an analytically perfect evaluation nobody accepts fails the test |
| Three inputs every project decision needs | R&D cost and time to completion · probabilities of technical, implementation and commercial success · potential value given success |
| Five evaluation tools | Strategy tables · commercial and technical influence diagrams · sensitivity analysis · decision trees · expected value |
| Four portfolio displays | Portfolio grid (probability of success against potential commercial value, with quadrants named pearls, bread and butter, oysters, white elephants) · productivity curve (expected return against cumulative cost to completion) · segment return analysis · probabilistic new-product revenue forecast |
| Assessment convention | For each uncertain variable assess low at the 10th percentile, high at the 90th, and the median at the 50/50 value |
| Organisational prescription | Execute through functionally complete teams highly empowered by senior management, with both authority and resources; functional departments become reservoirs of competence but do not call the shots |
R9 is a consultant's synthesis of a decision-analysis toolkit illustrated with anonymised client displays, not an empirical study. It excludes basic knowledge-building research from its scope explicitly, and assumes shareholder value as the ultimate goal of industrial R&D.
Two of R9's rules are worth isolating. Do not fund a project if the team cannot clearly explain how its efforts can be expected to generate value. And when budgets tighten, do not cut every project uniformly — eliminate the weakest entirely if they cannot be dramatically improved. Both are treated on decision quality in R&D and R&D portfolio displays and strategy tables.
R11 — THE TRANSFORMATION MODEL AND ITS EVIDENCE
| Element | Content | Class |
|---|---|---|
| Four-direction model | Visioning and restructuring · excellence of product · excellence of process · excellence of people, run concurrently | Proposed framework from one case |
| Six drivers | Consensus on the need for change · top-team leadership · alignment with business units · stable investment · actionable planning and performance management · clear strategic direction and close division relationships | Asserted · no comparative case, no test of necessity or sufficiency |
| Commercialisation rate | 18 percent in 1997, 61 percent in 2002, 80 percent in 2004 | Study finding of one organisation · the two underlying cohorts are not comparable with each other |
| That the transformation caused the improvement | Claimed throughout | Asserted · no counterfactual, no control, no separation from the group's own growth |
Public funding, and where this set runs out
R10 — REGULATORY VALUES OF ONE FRAMEWORK, IN FORCE AT MOST UNTIL 31 DECEMBER 2006
| Item | Value |
|---|---|
| Individual notification trigger | Project costs above ECU 25 million and aid gross grant equivalent above 5 million. Both limbs |
| Base intensities | Fundamental research 100 percent of eligible costs · industrial research 50 percent · pre-competitive development 25 percent |
| Feasibility studies | 75 percent of study costs where preparatory to industrial research; 50 percent where preparatory to pre-competitive development |
| Bonuses, in percentage points | Small enterprise +10 · regional +10 or +5 · research-programme link +15, rising to 25 · dissemination or cooperation +10 |
| Absolute caps | 75 percent gross for industrial research; 50 percent gross for pre-competitive development |
| The decisive test | Incentive effect. For large firms it had to be proved, not presumed, and the state had to supply figures for the case where the project did not go ahead |
Historical regulatory values of one framework in one jurisdiction. Not current law. Full detail on R&D stage classification and aid intensity.
LIMITS REGISTER FOR THIS SET
| Limit | Consequence for use |
|---|---|
| Nine of eleven are practitioner-facing, not peer-reviewed | Frameworks are proposed rather than tested. Treat a named model as a hypothesis with a worked illustration attached |
| None of the eleven declares a research paradigm | You cannot read an ontology or epistemology off any of them. See declaring a research paradigm — or not |
| Papers span 1983 to 2006 | Technology, tooling and market commentary are of their date. The structural frameworks generally travel |
| Several contradict each other directly | On valuation, on whether goal clarity at initiation matters, on how success should be defined, on whether option pricing transfers. The disputes are unresolved |
| Single-source empirical bases are common | R3's regressions come from one division's single year; R11 from one organisation; R5 from one metropolitan area |
| The set was assembled for a literature review | It is not a systematic search and not representative. No count across these eleven describes the field |
What to carry forward
- Four labels, applied to every number: study finding, worked example, externally cited, regulatory value.
- The frameworks are the durable content. The thresholds, elasticities and rates belong to the samples they were measured on.
- R1, R2 and R3 give three incompatible answers to how an R&D project should be valued, and the set does not resolve them.
- R5, R6 and R8 converge on one point: what happens during a project matters more than what was written at its front end.
- R7's real contribution is the instruction to interrogate the denominator before accepting any success or failure rate.
- Nothing here is a benchmark, and no count across these eleven papers is a statement about the field.
Frequently asked questions
Why is every number labelled instead of just quoted?
Because these papers mix four very different kinds of quantity: results measured on one sample, figures invented to demonstrate a method, numbers quoted from elsewhere, and binding regulatory thresholds of an expired framework. Quoting them without the label is how a single study's finding becomes an industry benchmark in someone else's slide deck.
Which formula should I actually use to value an R&D project?
The set does not agree. R3 argues for discounted cash flow, principally internal rate of return, on the ground that it is the language the finance function already uses. R2 argues discounted cash flow is structurally wrong because it ignores the right to abandon at completion. R1 argues both miss what distinguishes projects, which is what each would let you learn. Pick deliberately and state which frame you are in.
Can I use the 1 to 2 percent acceleration elasticity as a planning figure?
Only with its conditions attached. It is one author's synthesis of three prior point estimates, all evaluated at about 10 percent above the minimum possible completion time. Penalties are greater nearer the minimum and smaller at long durations, and the two software estimates straddle the hardware one. Use it to argue that the relationship is convex, not to price a specific compression.
Why do two papers report such different success rates?
Because they measure different things. One counts a project as successful only if it achieved both technical and commercial success, across 211 projects in 21 companies. The other measures commercial success against a minimum acceptable profitability threshold, on fully developed products ready for commercialisation, at 103 firms. Different denominators and different definitions produce different rates, which is the point one of them is making.
Are the state aid figures still usable?
Not as law. They belong to one framework, in one jurisdiction, in force at most until 31 December 2006, with a replacement already announced when the paper was written. They remain useful as a worked example of how a classification-plus-ceiling funding scheme is built, and as a reminder that a stage classification can be a pricing decision.
What does this set not cover?
It was assembled as a reading list, not a curriculum, so its coverage is uneven by construction. There is no systematic treatment of R&D people management, of intellectual property strategy, of open or collaborative innovation, or of anything published after 2006. Nothing here should be read as a survey of the field.
References and source attribution
- Eleven copyrighted journal articles on R&D project management, supplied as a reading set assembled by a student for a literature review and profiled for this library: an economics conference paper on R&D project choice (1991); practitioner articles on option value (2000), financial appraisal (1984), acceleration cost (1989), termination decisions (2002), project success and failure (1986), new-product success rates (1983), post-project reviews (2003) and decision-analysis tools (1994); a regulatory practice review (2006); and a single-organisation case study (2006). Front matter, abstracts, framework sections, tables and figures were read; article bodies were not reproduced, and all content here is paraphrase.
- The framework of 1996 on state aid for research and development, prolonged several times and in force at most until 31 December 2006, as quoted within the regulatory practice review. Not supplied to this library; its values are historical regulatory values of their period.
- Figures quoted within these papers from other sources — the disputed 50 to 90 percent new-product failure rate, the benchmarking study of 79 R&D organisations, the three acceleration elasticities and the consulting review of complex-project overruns — were not supplied to this library and were not independently examined. The papers quoting them are not themselves evidence for them.
- Supplied teaching source for this library (research methods and research process materials). Used here for page conventions, voice and the provenance labelling scheme; it does not treat R&D project management.
Suggested questions for Ask KEVOS
- Build me a one-page selection rubric from the frameworks on this page, marked by which paper each criterion comes from.
- Which of these frameworks are safe to apply as-is, and which need re-testing in my organisation first?
- Explain the disagreement between the financial-appraisal paper and the real-options paper in terms my finance director would accept.
- Draft the provenance caveats I need if I quote three of these figures in a board paper.
- Which papers in this set contradict each other, and on what exactly?
- What questions about R&D project management does this reading set leave unanswered?