Navigating the

Technology Landscapeof Innovation

Some companies are better off

making incremental improvements

to their products. Others that must

compete on their ability to innovate

focus on breakthrough inventions.

Either approach requires the explora-

tion of a specific type of "technology

landscape" and the right strategy for

searching across the terrain.

Lee Fleming and Olav Sorenson

organization has its strengths — and weaknesses. To eniphasizethe former while minimizing the latter, companies often devote consider-ahle resources to their corporate strategies, for example, crafting the per-fect plan to outmaneuver the competition in an emerging market. At thesame time, those same companies may give short shrift to the essentialtask of determining exactly what their strategy for product innovationshould be. The result: R&D projects that are out-of-synch with the rest ofthe organization.

Developing the right strategy for product innovation is hardly a simple mat-ter. In fact, it is a complex undertaking that first requires a fundamental

understanding of how technical modularity affects R&D efforts. In amodular design, a change in one component of a product (the

heating element of a coffeemaker, for instance] has relativelylittle influence on the performance of other parts or the sys-tem as a whole. \n a nonmodular, or "coupled," design, thecomponents are highly interdependent, and the resuh isnonlinear hehavior: A minor change in one part can causean unexpectedly huge difference in the functioning of the

overall system. With semiconductors, for instance, a minus-cule impurity (jusi 10 parts in a hillion) can dramatically alter

silicon's resistance hy a factor of more than 10,000. Generallyspeaking, modular designs make R&D more predictable, but they

tend to resuh in incremental product improvements instead ofimportant advances. Coupled designs, on the other hand, are riskier towork with, but they arc more likely to lead to hreakthroughs.

This trade-off hetween predictability and innovation can be visual-ized as a "technology landscape," with gentiy sloping hills corre-

sponding to incremental product improvements that are based onmodular components and soaring, craggy peaks representing break-

Lee Fleming is an assistant professor at the Harvard Business Schooi in Boston.Olav Sorenson is an assistant professor at the Anderson Graduate Schooi of Manage-

ment at the University of California at Los Angeles. They can be reached at ifleming@hbs.edu and olav.sorenson@anderson.uoia.edu.

WINTER 2003 MIT SLOAN MANAGFMENT REVIEW 15

through inventions tbat rely on tightly coupled parts. Develop-ing new products requires a search across such technology ter-rain. For a company like Dell Computer that can compete on itsefficient manufacturing and superb supply-chain management,avoiding the rugged peaks and instead traversing tbe slopinghills is an effective strategy. Other corporations like Apple Com-puter need to scale tbe high peaks to maintain their competitiveadvantage. For such expeditions, one approach is to minimizerisk by developing a "map" of the topograpby — tbat is, by gain-ing an understanding of the underlying science of the technolo-gies being used.

Tbe concepts of "technology landscapes" and "maps" serve aspowerful metaphors for understanding why some companieshave profited from their R&D efforts whereas others have stum-bled. Indeed, the framework is a valuable tool for helping or-ganizations make important decisions about tbeir innovationprocesses and resource allocation. For instance, a t1rm that ishaving great difficulty tnoving products through R&D intomanufacturing could be exploring an area of the technology

landscape that is too rugged. Instead, the company might try towork in less precarious terrain by tising more modular compo-nents (perhaps standardized parts). Or it might greatly benefitfrom a large investment in basic science to develop a map of thelandscape that will help researcbers avoid technological pitfalls.By addressing such issues, an organization can ensure tbat itsstrategy for product innovation is best suited to its competitivestrengths.

Technology LandscapesWe start with a simple and classic idea: Inventions combine com-ponents — whether they be simple objects, particular practicesor steps in a manufacturing process — in new and useful ways.An inventor can create novel products eitber by rearranging andrefining existing components or by working with new sets ofthem. For example, tbe steamship unites the steam engine withthe sailing ship. The automobile merges tbe bicycle, the horsecarriage and the internal combustion engine. DNA microarrays— devices that enable scientists to investigate tbe efficacy or

About the ResearchOur research on product innovation is

based on an extensive analysis of more

than two centuries of U.S. patent data.

Specifically, we investigated how the in-

novativeness of an invention might be

related to tbe number of its component

parts and the degree to which those

pieces are Interdependent.

For our work, we used key insights

from the "NK" model of complexity theo-

ry.' In tbe model, N is the number of

components in a system and K repre-

sents the richness of interactions among

them. Biologists, for example, might de-

fine N as the number of genes in an or-

ganism, each of which influences K other

genes. In our study, the measures of N

and K came from the subclass assign-

ments of each U.S. patent. These cate-

gories represent a fine-grained classifica-

tion of technology (the U.S patenting

system has defined more than 100,000

subclasses). We determined the interde-

pendence of these subclasses by analyz-

ing how they had been recombined in

more than 200 years of past inventions

— an analysis that required nearly 350

billion calculations.

In a technological context, interde-

pendence implies "coupling" between

components. When coupling is exten-

sive, a change in one component can

have a dramatic effect on the perform-

ance of the overall system, whereas in

a decoupled {or modular) invention the

effect is comparatively small. Thus our

NK analysis of the patent data provided

a quantifiable way to determine how

coupled each invention was. We then

assessed the relative innovativeness of

each of those patents by looking at the

number of times they had been cited

by subsequent inventions. Extensive re-

search has shown that more highly cited

patents stimulate more important future

research and engender greater financial

rewards. Next, we used statistical models

to explore how component coupling

tends to affect innovation."

Implications from the results of our

research can best be visualized as

"technology landscapes" (see accom-

panying examples). In such terrain, the

peaks represent different inventions —

the higher the peak, the more Innova-

tive the product. In addition, the

ruggedness of the topography indicates

the amount of component coupling —

the rougher the terrain, the greater the

coupling. Developing new products

requires a search across such technolo-

gy landscapes, with inventors looking

for tbe peaks. To understand how these

concepts play out within companies, we

undertook dozens of case studies, both

by using archival data on the history

of inventions and by interviewing many

inventors directly.

Our research has focused on the de-

velopment of physical products, but we

believe that our results likely apply to

service-based industries as well. There,

interdependence can occur between

the activities involved in providing a

service, similar to the way in which the

components of a physical invention in-

teract. For example. Wal-Mart's ability

to distribute products efficiently to

small and large towns across the United

States depends on highly tuned and

tightly coupled routines related to its

ordering, inbound and outbound logis-

tics, and marketing. These interdepen-

dencies have made it difficult for

16 MiT SLOAN MANAGEMENT REVIEW WINTER 2003

toxicity of ;i trial drug— rely on sfmiamductors, fluuresccnt dyeand nucleotides. And the emerging field of niiiionianufacturingcombines techniques from the semiconductor, mechanical andbiotcch industries.

In a technology landscape, the summits correspond lo inven-tions that have successfully merged different components, andthe valleys represent failed combinations. (The concept of tech-nology landscapes was tested in an extensive study of more than200 years of data from the U.S. Patent Office. For details, see"About the Researcb.") Metaphorically speaking, inventors seekout the peaks on this landscape, while trying lo avoid the chasmsin between.

When researcbers do not understand the components they areworking with nor how those parts interact, they search,blindly ina fog, unable to see tbe surrounding peaks and valleys. Because ofthis, the researchers (and their managers) prefer cautious forays,that is, small adjustments to proven concepts — a process knownas local, or incremental, search. Such an approach is attractive be-cause researchers know from experience that the minor enbancc-

ments will likely work and receive at least reasonable acceptancein tbe market. And local searching works particularly well on ter-rain dominated by a single peak like Mt. Fuji. For sucb a land-scape, inventors simply need to travel uphill to discover the nextinvention. A disadvantage here, though, is that even the personwho is first to reach the top gains, at best, a tleeting technologicaladvantage over competitors.

Now consider a different landscape like the Alps that is char-acterized by a multitude of crests. Here, inventors searcbing intbe fog will miss most ot the great inventions becaiise they are sit-uated beyond an abyss of tecbnologicai dead ends. Proceedingslowly upbill will typically leave people stranded on some localhill far below tbe soaring beights of a Mt. Rlanc. And oncestranded on a local peak, any direction I bey move will lead down-hill. Because of the fog, they cannoi see the heights beyond theneighboring valleys so they do not know in which direction tohead or whether to quit altogether. Thus, on such rugged land-scapes, local searching will almost certainly fail, even after enor-mous investments of time and effort. Instead, such explorers

competitors to mimic Wai-Mart's suc-cess.'" Service industries can also bene-fit from the application of science, buthere firms wiil iikely draw more onmathematics, statistics, and operationsresearch than on advances in bioiogy,chemistry, and pbysics. Our future workwili investigate how our theories of

product development migbt extend toservice industries.

i. For an overview of complexity theory and its ori-gins, see M. Waidrop. •'Compiexity" (New York:Touchstone/Simon & Schuster, 1992). Our workbuilds on research described in S. Kauttman, "TheOrigins of Order" (New York: Oxford UniversityPress, t993).

ii. For details of this research, see L. Fleming and0, Sorenson, "Technology as a Complex System:Evidence From Patent Data." Research Policy 30(August 2001): 1019-1039: and L Fleming and 0,Sorenson, "Science as a Map in TechnologicalSearch," working paper 02-096, Harvard BusinessSchool. Boslon, Massachusetts, 2002.

Hi. J.W. Rivkin, "Reproducing Knowledge: Repli-cation Without Imitation at Moderate Complexity,"Organization Science 12 (2001): 274-293,

The terrain of technoiogy landscapes can vary greatiy depend-ing on the type of parts that inventors use to buiid products.The three-dimensionai curves shown here are highiy simpli-fied representations depicting inventions with just two con-stituent parts (components A and B), but the concepts illus-trated would be similar for products with hundreds orthousands of parts. When the components are modular (ploton left), the terrain created by possible combinations of vari-

ants is relatively smooth, enabling inventors to easily developuseful products, which are represented by the numerouspeaks on the plot. When the components are "coupled," orhighly interdependent (plot on right), the peaks can be higher— denoting more technically successful inventions — but lesslikely to be found because the terrain is more jagged. Deadends and failures are more likely and inventors have greaterdifficulty developing products.

Component

WIMTER 2003 MIT SLOAN MANAGEMENT REVIEW 17

Summary of Strategies for Product Innovation

Type of Strategy

IntellectualProperty (IP)Protection' Benefits

Modular Strategy High

Coupled StrategyWith ShotgunSampling

Low

DrawbacksManagerial Prescription and StrategyImplications

Components can bemixed and matchedeasily.

Design costs can beminimized throughreuse of components.

Distributed develop-ment is easier.

Stability and reliability.

Barriers to entry arelower.

Tougher to differen-tiate products.

Competitive businesslandscape is morecomplex.

lower performance.

Provide adequate IP protection and/or comple-mentary capabilities, such as rapid productintroduction, efficient manufacturing, strongbrand names, efficient supply chains or superiorcustomer service.

Compete as system architect, standards owneror module maker.

Difficult for followersto imitate, especiallysmall and mid-sizecompanies thatlack technologicaldiversity.

Trials require mean-ingful and stringenttests.

Difficult to manageor predict.

Less effective ifsequential decisionsare needed in thedevelopment process.

Increase variance and number of trials.

Physically juxtapose multiple diverse and previ-ously uncombined technologies.

Keep slack resources available for unexpectedand tangential exploration, resulting from inter-nal opportunities or external threats.

Encourage informal networks and tinkering.

Simulate large systems if the interactions of indi-vidual components can be modeled.

Characterize manufacturing processes carefully.

Coupled Strategy Medium Minimizes technologi-With Mapped cal uncertainty.

Searching Doubles the technicalpayoff of modularstrategy.

Risk of appropriation of IP and consequent required levels of IP protection

Can require expen-sive investment inscience, thus increas-ing commercial risk.

Greater technologicalcompetition.

Seek IP protection.

Physically juxtapose technologists and scientists.

Reward scientists internally.

Keep abreast of published scientific literatureand maintain contacts with academia.

need some sort of map — even an approximate one. With it, theycould anticipate the rough terrain and decide how best to reach asummit. Having finally arrived there, they would have gained astrong technological advantage over rivals, particularly those thatlack a map.

Obviously, exploring rugged terrain (that is, working withtightly coupled systems) is an unpredictable and risky undertak-ing. But, as mentioned earlier, the payoffs can be huge for thosewho make landmark discoveries that others have difficulty repli-cating, That said, trying to achieve those breakthroughs can be amassive endeavor that not every organization should attempt. In-deed, many companies are better suited for exploring less ruggedterrain (that is, working with modular components).

Smooth Landscapes With Sloping Mountains A modular recombina-

tion strategy offers numerous benefits.' (See "Summary ofStrategies for Product Innovation.") It reduces design costs andexpenditure of time because people can reuse proven compo-

nents to quickly build complex .systems that work. Further-more, because the components have little interaction, compa-nies can deploy efficient distributed-development approachesin which various teams of researchers work in parallel on thecomponents.

Exploring a smoother technology landscape, however, has itsdrawbacks. Although modular recombination decreases the riskof a failed invention, it also limits the upside potential becauseblockbuster Innovations become far less likely. Also, modular sys-tems take a performance hit because they cannot fully exploit thepotential synergies among components. For example, in com-puter workstations, Sun Microsystems was able to develop work-ing systems relatively easily because of its use of modular micro-processors, but for years the computational power of thecompany's workstations lagged those of Apollo Computer, whichrelied on customized (and more coupled) designs. Eventually,though. Sun eclipsed Apollo as the latter found it difficult to sus-tain its pace of innovation.

18 MIT SLOAN MANAGEMENT REVIEW WINTER 2003

Excessive modularity also makes it difficult tt) maintain .1competitive advantage. Because highly modular componentscombine easily, rivals can easily copy a modular product. As thistechnical know-how spreads, tbe industry can become commod-ity-driven, resulting in a low barrier to entry, numerous competi-tors and little differentiation among them. Tbe personal comput-er market is a case in point. Ironically, as the technologylandscape of innovation becomes simpler to navigate, tbe com-petitive landscape can become treacberously complex.

Competing on a smootb tecbnology landscape tbus requiresstrong legal protection of intellectual property, complementary

To pursue a modLilar strategy for product innovation, a com-pany must first direct its R&D group to use existing standards,particularly if tbey are widely accepted. If standards have yet toemerge, engineers should develop their own standardized inter-laces and protocols before embarking on a new product design.And after establishing such proprietary standards, tbe organiza-tion should actively pursue their widespread adoption in the in-dustry, especially if competing standards are emerging. Oneeffective tactic is to outsource any manufacturing of tbe stan-dardized components to encourage suppliers to develop partsthat tbey can sell to other companies, thus accelerating the pro-

Ironically, as the technology landscape of innovation becomes simpler to navigate,the competititive landscape can become treacherously complex.

capabilities {low-cost manufacturing, for instance) or a combi-nation of both. Patents can provide some security from copycatcompanies, but it typically erodes as the tecbnology matuies andoutsiders learn bow to design around the patents. Without astrategy for safeguarding its intellectual property, a firm mustrely on complementary strengths, such as fast product develop-ment, lean manufacturing or superior customer service. Dell, forinstance, benefits not from revolutionary product developmentbut through exemplary supply-chain management and strongexecution of its business model. Those strengths enable tbe com-pany to succeed in a market in which standard interfaces andprotocols have allowed even consumers to build their own PCsby mixing and matching parts (monitors, microprocessors, diskdrives and so on).

Another example of a successful modular approach is theWalkman. To develop that landmark product, Sony engineersfirst built a library of standard, intercbangeable components.Doing so sbifted its product searcb to a smooth landscape, allow-ing it to find useful combinations witb relative ease. Sony wasable to maintain its competitive advantage not because its initialposition was unassailable, btit because tbe company was alwaysable to stay at least one step abead of the competition. Modularrecombination enabled rapid innovation, allowing Sony to inun-date tbe market continually with product changes and improve-ments, making it difficult for competitors lo keep pace. Sony alsobad tbe manufacturing expertise and marketing muscle to keeprivals at bay. The success of the Walkman also holds another im-portant lesson: Although modular designs tend to result in incre-mental product improvements, they can sometimes result in a se-ries of inventions that become commercial blockbusters.

lilcration of the standards throughout the industry. (Incidentally,Sun Microsystems pursued exactly this approach to outmaneuverApollo Computer.)

The structure of an R&D group can also be a major factor. Toreinforce modular-design principles, companies should allocateproducl development responsibilities to independent teams, eachdealing witb its own component of the system. This approacb ofdistributed development can subtly motivate researcbers and en-gineers to adopt standards and pursue modular designs. Other-wise, ihey will struggle with great inefficiencies as the compo-nents tbey develop have trouble working with the parts that otherteams have built.

Rugged Landscapes With Croggy Peaks As noted earlier, a coupled

strategy lor product innovation is a high-risk/bigh-return en-deavor. Companies tbat are fortunate enough to succeed withsuch an approach will occupy an almost ideal position. Discover-ies from this process typically involve much tacit knowledge, anda competitor will find it difficult to replicate those results becausethey will likely require an understanding of the underlying prin-ciples. Hence, intelleclual property protection and complementa-ry capabilities matter less tban they do for a modular strategy,(hideed, keeping an invention secret might protect it more effec-tively than a patent, because tbe patent application process re-quires disclosLU'e ol technical details.) To pursue coupled designs,however, companies need an effective strategy for dealing withthe daunting complexity of rugged technology landscapes.

Shotgun sampling. One approacb is to generate an enormousnumber of random trials and then subject those to rigorous

WINTER Z003 MIT SLOAN MANAGEMENT REVIFW 19

selection criteria. Instead of understanding the interdependen-cies among components that generate the landscape, inventorssimply apply the magic of large numbers in a manner similar tothe evolutionary process of natural selection. The successfulimplementation of this strategy of shotgun sampling requirestwo things: methods to generate variation cheaply and accuratetests to assess the value of those variations.

Thomas tidison's laboratory in Menlo Park, New Jersey, oneof the most successful research facilities in history, operatedunder the principles of shotgun sampling. F.dison brought to-gether a host of gadgets, chemicals, compounds and technolo-gies under one roof and staffed his inventor's dream house witha wide variety of technical and craft professionals. The layout ofhis lab further facilitated rapid shotgun sampling by juxtaposingseemingly unrelated technologies in the same work area and byencouraging his staff to search far and wide for strange combi-nations. In developing his most famous invention, the light bulb,Edison tried some 1,000 myriad combinations of filament mate-rials, vacuum pressures, voltages, bulb shapes and so on, runningeach against an ohvious and rigorous test: How long does thebulb stay lit?

In today's world, computer simulation offers a powerful toolfor improving the efficacy and reducing the costs of runningmyriad trial-and-error tests. Inventors can take relatively inex-pensive, virtual samples of large swathes of the technology land-scape instead of having to perform costly physical trials. BMW,for instance, builds computer models of prototype cars and then"crashes" them in virtual simtilations. Through this work, thecompany has been able to investigate new designs for improvingthe safety of its vehicles. BMW then builds physical prototypesto verify the iinal designs.̂ To be effective, computer simulation

Stanford University and the University of California at Berke-ley. Engineers had to simulate the effects of increasing a micro-processor's cache memory size or of implementing certaininstructions in hardware, because they could not predict howsuch changes would affect the chip's ahility to run variouscomplex software programs. As a result of this work, the annu-al rate of improvement in microprocessors increased from 35%to 55%.

An organization that performs shotgun sampling across arugged technology landscape can deploy a variety of tech-niques to increase its odds of success. First and foremost, it canwork with many diverse technologies, and it can physically jux-tapose researchers from seemingly unrelated fields to ensurethat inventors are not straitjacketed into technological andfunctional silos. It can also promote a relatively unstructuredworkplace. For example, 3M's policy of allowing its researchersto spend one day per week on their own projects encouragesthem to pursue random and unexpected tangents. A companyalso needs to be constantly on the lot)kout for unexpectedbreakthroughs, even if they appear far from the original target.Lastly, before a company releases the results of a shotgun-sampling strategy into production, it must ensure that its man-ufacturing process is reliable over a wide range of parameters.One caveat: If people do not understand how a technologyworks in the lab, identifying the root cause of problems inmanufacturing will prove nearly impossible, as has been thecase with many semiconductor companies and their fabrica-tion facilities.

A crucial point to remember is that shotgun sampling re-quires clear and stringent test criteria. Otherwise, the processcould result in terrible inefficiencies and frustrated researchers.

Shotgun sampling requires clear and stringent test criteria. Otherwise, the processcould result in terrible inefficiencies and frustrated researchers.

requires an accurate means of modeling the interactions betweenindividual components as well as reliable tests to determinewhich designs are superior.

The real payoff comes when engineers simply cannot man-ually predict the complex behavior of a system that emergesfrom the interactions of thousands or millions of components.For example, the success of microprocessors based on reducedinstruction set computing [RISC) in the 1980s hinged on theuse of computer simulation — a massive research undertakingthat required the efforts of Hewlett-Packard, IBM, MIPS,

In the pharmaceutical industry, for example, researchers facedaunting challenges hecause of the interactions among a drugmolecule, the system used to deliver that compound to its tar-get, and the idiosyncratic phenotype of each individual patient.Rather than attempting to understand all of those intricate,coupled interactions, Merck and other pharmaceutical firms in-stead have developed methods to test huge arrays oi randomcombinations of potential drugs simultaneously against multi-ple targets. Such targets could include, for example, leptin re-ceptors for obesity, complement receptors for infiammation

2 0 MIT SLOAN MANAGEMENT REVIEW WINTER 2003

and interleukin for allergies. The companies increase the vari-ance of their trials by constantly seeking out new compounds toexpand their libraries. Bui this approach has been slow to yieldnew drugs, partly because it lacks meaningful and effective tests.Researchers often have a limited understanding of the targetdiseases and must represent them by using simple cell-based as-says that do not model accurately how a potential drug willwork in humans. Because of that, companies often do not learnthat a certain drug is a failure until much later, in expensiveclinical trials. Thus, in summary, shotgun sampling works bestwhen a company has lots of cheap ammunition, multiple tar-gets and clear signals indicating that one of those targets hasbeen hit.

Mapped searching. Rather ihan perform shotgun sampling, acompany can instead obtain a fundamental understanding of thedifferent coupled components (and their associated technolo-gies), including how certain parts affect one another. With suchknowledge, researchers will be able to narrow their field of ex-ploration and arrive at useful inventions more quickly. This ap-proach is analogous to "mapping" the technology landscape. Typ-ically, although such a strategy requires an expensive andlong-term investment in basic or applied science, the potentialpayoffs are huge. Research on patent citations demonstrates thatthis approach generates inventions of twice the import, on aver-age, than the modular recombination strategy.-"* Translating thisinto financial rewards promises an even greater payoff becausethe commercial value of patents that are highly cited increases ex-ponentially. Specifically, recent work on breakthrough innova-tions has found that a patent with 14 citations has 100 times thevalue of a patent with eight citations.'*

undertake a mapped search on their own. In addition, smallercompanies and startups can greatly extend their in-house knowl-edge by building an informal outside network of scientists acrossa variety of fields, perhaps by locating close to universities andthen establishing relationships with the researchers there.-̂ What-ever the means, providing researchers with even an inexact mapof their rugged technological landscape can pay off handsomelyas long as the information is accurate enough for finding valuableinventions more quickly.

The power of mapped searching is amply illustrated by H-P'sinvention of the inkjet printer. Its legendary dominance of theinkjet market makes it difficult to imagine that the companystruggled greatly to develop the technology, eventually succeed-ing where legions had failed. In fact, the basic concept of inkjetprinting was first published more than a century ago hy LordKelvin in the 1867 Proceedings of the Royal Soci'̂ ty. Since then,IBM, Sperry Rand, Stanford University and Xerox each investedmillions of dollars in the technology, yet the resulting productsconsisted of fragile parts and operational sensitivities that defiedreliable operation. The inkjet printer challenged inventorstremendously because it involved so much coupling among thedifferent components, including the chemistry of the ink, thephysical layout of the resistors providing the heat pulse to spraythe ink, the shape of the electrical signal through the resistorsand so on.

lohn Vaught, a technician at H-P labs, finally succeeded inputting together the winning combination of intricate compo-nents that led to the company's blockbuster product — anachievement made possible only through a basic understandingol the complex physics of superheated liquids.*^ Vaught's storyholds several lessons. First, the scientific literature can provide

Recent work on breakthrough innovations has found that a patent with 14citations has 100 times the value of a patent with eight citations.

An important point to remember is ihal companies don't nec-essarily have to conduct basic research to benefit from it. Simpleawareness of and familiarity with the scientific literature can pro-vide a powerful advantage when dealing with highly coupledtechnologies. Alternatively, companies can pool their risk by par-ticipating in an industrial consortium like Sematech, in whichmember firms (Intel, Motorola, IBM, Texas Instruments and oth-ers) jointly develop key semiconductor manufacturing technolo-gies. Such approaches are particularly crucial for small firms thatdo not typically have ihe substantial resources necessary to

a wealth of valuable information. After Vaught made an initialbreakthrough with microjels, his H-P colleagues Howard Tauband lohn Meyers read the appropriate physics journals to gainfrom others' knowledge of bubble nucleation, the phenomenonthat is the basis for all inkjel printers. Second, researchers cantap into their personal network of colleagues for access to thenecessary expertise. Taub and Meyers contacted their friends atthe California Institute of Technology lo gain the latest infor-mation on microjets and high-temperature vaporization. Third,scientifically capable inventors then can do the remainder of the

WINTER 2003 MIT SLOAN MANAGEMENT REVIEW 21

necessary basic research. Meyers and his colleagues devisedoriginal simulations and numerical techniques to tmderstandhow best to apply the mechanism of bubble nucleation to thespecific context of their invention. This work helped map H-P'stechnology landscape, enabling the company to understandhow nucleation occurs in ink that is heated to extremely hightemperatures.

A technology map can be particularly useful for companiesthat have hit a dead end but don't yet realize it. Consider the caseof Lord Corporation, a Gary, North Carolina, company thatmanufactures adhesives and dampers. In the 1980s, Lord wastrying to develop a "smart damper" that would be able to, for ex-

repeated failure. When David Wong of Eli Lilly was looking for anew drug lo treat depression, he continued investigating a par-ticular set of Benadryl compounds, even though that groupinghad previously failed to show any promise. But Wong persistedbecause he had a theory of serotonin uptake — that is, an un-derstanding of the mechanism by which certain brain cells re-lease and absorb the powerful neurai transmitter. Eventually,thanks to that knowledge, Wong discovered the antidepressantdrug Prozac.

Eor all its advantages, though, mapped searching can he ex-tremely tricky to implement- Specifically, companies that de-ploy mapped searching must pay special attention to how they

Companies are not passive bystanders; they can migrate across thetechnology landscape to find the terrain that suits them best.

ample, change its resistance to give the rider of an exercise hikethe impression of climbing uphill. At the time. Lord [like mostof its competitors) had been focusing on using "electro-rheolog-ical" (ER) fluids, which thicken (thus increasing their resistance)when an electric field is applied to them. The company struggledwith ER materials until Dave Carlson, a company engineeringfellow, did a crucial calculation showing that the technology,even at its best, was essentially impractical — an analysis that re-quired a fundamental understanding of the chemistry andphysics involved.

Given that realization, Lord quickly switched its search to in-vestigate magneto-rheological (MR) materials, which are similarto ER fluids except that their damping qualities change when amagnetic (not electric) field is applied. This shift to a different re-gion of the technology landscape proved fortuitous. Lord re-searchers experimented with a variety of MR materials and dis-covered that a particular oil with iron filings and additivesseemed to work best. To investigate that finding. Lord returned toscience and did original research on the breakdown of iron par-ticles under mechanical and magnetic stress. This work has led tothe development of a variety of products, including an earth-quake damper for buildings and bridges, smart shock absorbersfor cars and a prosthetic leg that provides dynamic resistance onthe basis of the wearer's weight, gait and physical surroundings(for example, whether the wearer is walking on a fiat surface orclimbing stairs).

A technology map is also invaluable for researchers who thinkthey're at a technological dead end but really aren't. Indeed, sci-entific theories can encourage an ongoing search in the face ol

manage their scientists. Engineers tend to seek career recogni-tion within their firms and respond well to both compensa-tion- and career-based incentives. Scientists, on the otherhand, often seek professional recognition through externalpublication and respect within the wider scientific community.Recent evidence even suggests that scientists will accept lowersalaries for the opportunity to achieve such external recogni-tion.'' Keeping a scientist's work proprietary, then, can provedifficult when that person measures career success by the num-ber of articles published and the prestige of the journals inwhich they appear. Eurthermorc, the fundamental norms ofscience and the very basis for its success stem from open pub-lication and discourse. Lastly, good science codifies and articu-lates knowledge. Thus, even if a firm's scientists conduct theirresearch without the rewards of external publication, the codi-fication of tbeir knowledge makes that information less tacitand hence more easily appropriated by other firms.^ For exam-pie, researchers who leave a company can more easily recon-struct their work for a rival.

These prickly issues are illustrated by IBM's experience withits breakthrough on copper-interconnect technology.^ The in-vention, which enables the transistors on a chip to be connectedal very low electrical resistances, was a major coup. When IBMannounced it, the company's stock price rose 5%, and when thetechnology was released into production, the stock price rose anadditional 6%. Wisely, IBM allowed its researchers to publishonly general ideas about the tecbnology, keeping the specific de-tails proprietary. And it shrewdly refrained from applying forpatents to avoid disclosing that valuable information. During its

2 2 MIT SLOAN MANAGEMENT REVIEW WINTER 2003

downturn in the early 1990s, however, IBM suffered drastic cut-backs and, as a result, several researchers left the company. Withihis diffusion of knowledge, the rest of the industry gained a bet-ter grasp of the true advantages of copper-interconnect technol-ogy, and other semiconductor manufacturers increased theirefforts in that area, catching up with IBM's expertise within acouple of years. But in the chip industry, two years represents ageneration in technology, enabling IBM to ship more than a mil-lion chips before its competitors could respond.

The main lesson of IBM's success with its copper-intercon-nect innovation is that mapped searching across a rugged tech-nology landscape can yield incredible returns. To reap thosebenefits, however, companies must actively manage and closelymonitor their innovation strategies. Generally speaking, amapped strategy requires greater intellectual property protec-tion and stronger complementary capabilities than the ap-proach of shotgun sampling. On the other hand, it also requiresless IP protection and fewer complementary capahilities than amodular approach.

Finding the Right GroundCompanies that pursue a modular strategy to product innova-tion should find that their components mix and match easilywith little impact on system performance. They should also beable to easily substitute raw materials, vendors and manufactur-ing processes. In contrast, organizations that choose to competewith a coupled strategy should find just the opposite. Their in-ventors should find it difficult to explain exactly why innovationswork, why they only work in small "sweet spots," or why a smallchange can crash an entire system,

An important point to remember is that industries are dy-namic entities, so organizations need to continually reassesswhether they are exploring the right technology landscapes. As amarket matures, for example, a company that has built a strongbrand might elect to focus on lowering its manufacturing costs,perhaps by using more standardized parts to develop its nextgeneration of products. A complicating factor is that differenttechnologies and projects at the same company can represent dif-ferent landscapes, each requiring a tailored strategy.

To move to a more rugged landscape, a company should mo-tivate its engineers to work with new and more-sensitive compo-nents — a move that many may initially resist not only becausethey prefer using what's familiar but also because of the greaterinherent frustrations of working with coupled technologies. Re-gardless, an organization can support the transition by giving re-searchers the time and resources to explore ideas that mighi aifirst seem impractical and by not penalizing them for a tempo-rary reduction in productivity.

Going in the opposite direction — that is, moving to asmoother technology landscape — involves a relatively easier and

more conventional managerial challenge. Companies that wantto make this transition need to stress increased reliability, pro-ductivity and the reusability of components and designs. Ofcourse, migrating to a modular strategy of product innovationonly makes sense for companies that possess the complementarycapabilities necessary to prosper in commodity-like markets.

Either type of transition across a technology landscape pres-ents its share of challenges. But the crucial point is that compa-nies are not passive bystanders; they can migrate across the ter-rain to find the topography that suits them best. y\nd by doing so,they can capitalize on their strengths •— and avoid having theircompetitors exploit their weaknesses.

ACKNOWLEDGMENTS

We thank Corey Billington and Ellen King of Hewlett-Packard fordonating their patent database and for their generous support of ourcomputing needs,

REFERENCES

1. See also C. Baldwin and K. Clark, "Managing in an Age of Mod-ularity," Harvard Business Review 75 (September-October 1997):84-95.

2. S. Thomke, "Enlightened Experimentation: The New Imperative forInnovation," Harvard Business Review 79 (Eebruary 2001): 66-75.

3. L, Fleming and O. Sorenson, "Science as a Map in TechnologicalSearch," working paper 02-096, Harvard Business School, Boston,Massachusetts, 2002.

4. D, Harhoff, F. Narin, F. Scherer and K. Vopel, "Citation Frequencyand the Value of Patented Inventions," Review of Economics and Sta-tistics 81 (August 1999): 511-515,

5. L. Zucker, M. Darby and M. Brewer, "Intellectual Human Capitaland the Birth of U.S. Biotechnology Enterprises," American EconomicReview 88 (March 1998): 290-306.

6. L. Fleming, "Finding the Organizational Sources of TechnologicalBreakthroughs: The Story of Hewlett-Packard s Thermal InkJet," Indus-trial and Corporate Change 11 (November 2002): 1059-1084; and, foran account of how Inchiro Endo also invented the inkjet at Canon, in-dependent of H-P's efforts, see A. Robinson and S, Stern, "CorporateCreativity: How Innovation and Improvement Actually Happen" (SanFrancisco: Berrett-Koehler, 1997).

7. S, Stern. ''Do Scientists Pay To Be Scientists?" working paper 7410,Nationai Bureau of Economic Research, Cambridge, Massachusetts,1999.

8. O. Sorenson and L. Fleming, "Science and the Diffusion of Knowl-edge," working paper 02-095, Harvard Business School, Boston,Massachusetts, 2002.

9. For further details on IBM s story, see: K. Lim, "The Many Faces ofAbsorptive Capacity: Spillovers of Copper Interconnect Technology forSemiconductor Chips," working paper 4110, MIT Sloan School of Man-agement, Cambridge, Massachusetts, 2000.

Reprint 4422Copyright <•(? Mtiffachusclls Inslhuic of Technology, 200.}. All liglils rcicrvi'd.

WINTER 2003 MIT SLOAN MANAC]EMe\IT REVIEW 23