The failure of system dynamics to become part of the management consulting profession’s standard toolbox

img_20160821_205651On December 9, McKinsey Quarterly did a reprint of a 1995 article by Jay Forrester, who died on November 16, about system dynamics.  As said in the intro to the article, Forrester “extended [feedback loops and other now-common techniques] from industrial operations to business strategy, employment cycles, social problems, and the fate of the world.”

For readers who are not familiar with the subject of system dynamics (hereinafter SD), SD is about modelling (typically non-technical systems) using the methods of ordinary differential equations.  It is in particular applied to systems with elements of human decision-making and long time delays.  See a causal-loop diagram of the situation facing COIN in Afghanistan produced by PA Consulting Group, below and to right (source:  Such diagrams are generally seen as great pedagogical tools, though the one below is on the complex side.


On a personal note, I like many other engineering-minded MBAs have always been intrigued by SD: i) I am engineer, with a degree in technical cybernetics and systems theory, and believe in the practices of engineering; ii) SD is just engineering practices applied to non-engineering phenomena; and iii) SD has proven and robust predictive powers, also in non-engineering domains (properly applied),

That said, SD has been and is controversial for a number of reasons: cross-disciplinary, sometimes obvious, sometimes counterintuitive (but correct), a technocrat’s solution to political issues, and often very complex.  It nicely explains the Kondratieff cycle (in the National Model), though economists would claim that such cycles have been studied and explained by economists since 1925.  It was the conceptual basis for Forrester’s book Urban Dynamics, which essentially and controversially stated that low-cost housing in certain areas in Boston increases poverty in same area, it does not reduce it.  It was also the basis for Limits to Growth, which was commissioned by the Club of Rome, and included a large number of quantitative simulations of a world with exponential economic and population growth in combination with finite resources.

A most insightful and elegant attack on the complexity of SD was indeed delivered by NY Times in an April 2010 article about the same SD model as referred to above, though NY Times somewhat incorrectly framed the issue as a PowerPoint issue, rather than as an SD issue.  Their heading was “We Have Met the Enemy and He Is PowerPoint”.  (See the original article here:, and PA Consulting’s response here:

Given the importance that for example McKinsey attaches to SD and Forrester’s work by republishing Forrester’s 1995 article, and given SD’s great predictive capabilities, also in in competitive situations, it is nevertheless a puzzle why SD is not more applied in the management consulting profession.  One could for example use it to predict industry end game structures, supply and demand cycles, or technology adoptions; explore alternative policies for resource allocation across business functions; or optimize supply chains; all stuff that serious strategy reports are built of.

Before exploring this issue in more detail, I should say that it is not all bleak.  I had in May 2015 the pleasure of attending a talk by John Stermann, Professor of System Dynamics and Engineering Systems at MIT, on “Interactive Simulations to Catalyze Science-Based Transformation” and the contribution of SD to “the implementation of sustainable improvement programs to climate change and the implementation of policies to promote a sustainable world.”  And in Norway we have a local SD tools vendor, Powersim, and Dr. Jørgen Randers, one of the authors of the Limits to Growth study, is professor at BI Norwegian Business School.  But again, regarding SD in business, we shall see that it has had limited impact on serious, hard-core strategic decision making.

Methodologically, I decided to approach the puzzle from three perspectives:

  1. Is SD taught at Norwegian universities?
  2. Is SD used in client engagements by major general consulting firms, say by McKinsey, BCG, Bain, PwC, Deloitte and PA Consulting?
  3. What are the fundamental reasons for the lack of interest in SD from the business community?

Questions (1) can be answered fact-based: If restricting ourselves to Norwegian universities, both BI Norwegian Business School and University of Bergen appear to have strong offerings in the area of SD, measured in terms of faculty and course offerings.  The University of Bergen has indeed a Master’s programme on System Dynamics (see  Other Norwegian universities do not appear to have similarly strong offerings in this area, though they often appear to refer to SD in courses on for example dynamic strategy.

Question (2) can also be answered fact-based: Deloitte (after the acquisition of Systems Dynamics, Ireland in 2015), PwC, McKinsey; and PA Consulting appear to have or have had practices, but concentrated in a few locations, dependent on a few individuals, and coming and going with these individuals.  McKinsey’s web site (including McKinsey Quarterly) lists only 12 articles including the term ‘system dynamics’, though I believe they used to use the term business dynamics.  Regarding Bain and BCG, I have found limited traces of SD, though specific individuals may be using their SD expertise in their daily work.  The above is corroborated by using advanced search on LinkedIn (# of current employees using key word ‘system dynamics’): Deloitte: 89; PwC: 62; McKinsey: 24; PA Consulting: 19; Bain: 9; BCG: 3.  All caveats apply.

Question (3) is vexing: First, the complexity issue of SD is real, see the PA Consulting infographic above.  However, no engineer would say that a fracture-mechanics based model of the development of a fracture in a weld in an oil pipeline should be simplified to make it more pedagogical, but less accurate.  Indeed, there is an accepted and respected technical and scientific discipline called CAE (Computer-Aided Engineering), comprising FEA (Finite Element Analysis) and CFD (Computational Fluid Dynamics), which is all about simulating engineering phenomena in extreme detail, and routinely on grids of size 10-100 million nodes, which is about 100 000 times larger than the largest SD models I have ever seen in use.  And CAE originated in the late seventies, while SD dates its origins back to 1961.  The CAE community used 20-40 years to overcome initial scepticism from their more practically inclined engineering peers, but the SD community has more than 50 years after the origins of SD still not been able to convince the business community about its merits.

Second, some of the problems with SD may be linked to typical issues addressed or problems solved using SD.  As an example, last issue of System Dynamics Review includes articles on: understanding decision making about faculty gender balancing; a competence development framework for learning and teaching SD; and quantifying the impacts of rework, schedule pressure, and ripple effect loops on project schedule performance.  Not exactly the stuff that high-flier consulting careers are made of.

Third, a more fundamental issue is the modelling of future outcomes in continuous time and as many small decisions, and based on (at least partially) known mathematical relationships.  Today’s business reality is that strategy value is typically generated in discrete time and concentrated in a few large bets characterized by competition, optionality, uncertainty and flexibility.  In such situations, game theory, real options theory, or multi-player decision trees may offer better modelling frameworks.

But again, there are situations in business where SD provides deep strategic insight and actionable recommendations and for which there are no applicable alternative and equally valid theoretical frameworks (except that you may believe that you have genius gut feel and rely on that).  Look for situations where you have already tried action A to improve metric M, only to see M deteriorate (= counterproductive policies) or strategies for evolving industry structures.  Feel free to contact me at if you believe that you are facing such situation and would like to explore the use of SD.

And if you just want to learn more about SD, there are a number of journals and blogs, see for example System Dynamics Review (of System Dynamics Society, see or SDwise (at

I wish all readers of CI Perspectives a Merry Christmas and a Happy New Year.  And I commit to increasing CI Insight publication frequency in 2017.



State of the nation regarding decision trees / game trees in big bet industrial decision making


Grim Gjønnes, General Manager Crisp Ideas

Today, I will endeavour to address an interesting conundrum: Why fairly sophisticated industrial organizations with outstanding in-house financial talent chose not to take into account uncertainty, the value of optionality and rational competitive behaviours in their big bet industrial decision making.

Not all organizations of course.  Indeed, in Dagens Næringsliv (DN; the leading Norwegian financial newspaper) on 04.02.2016 there was an article about Norske Skog, a globally leading paper producer HQed in Norway, and how they were taken to court in the US by a group of holders of secured bonds, headed by Citibank, regarding an agreement between Norske Skog and a group of holders of unsecured bonds, including the Blackstone companies GSO and Cyrus Capital.  Reading between the lines, both the Blackstone guys and the Citibank guys appeared to have taken full package of MBA courses on decision trees and sequential game theory.  (Norske Skog’s executive management appeared in comparison less prepared*.)  Furthermore, in preparation for this blog post, I had a chat with a senior executive in a large European energy company, who stated that (my translation, slightly edited): “Mostly anybody, including the CEO, in [my company] would be able to credibly contribute to a discussion about the output from [their in-house stochastic dynamic programming software package]”.

But for most organizations, DCF with uncertainty modeled through the addition of a risk premium to risk free interest rate appears to be the preferred analytical tool for valuation of most industrial projects (with the exception of multiple-based methods), despite the fact that sound theoretical tools for say valuing optionality and predicting competitive behaviours have been available to most MBAs and PhDs in finance for one or two decades.

The situation reminds me of the situation of fracture mechanics-based integrity assessment of offshore oil and gas pipelines 5-10 years ago.  The tools for calculating correctly were there, but the practitioner community and various standards-setting bodies chose to stick to old-fashioned standards, like DNV-OS-F101, complemented with proprietary calculations and safety factors rather than do a full finite element calculation.  The result: often overly conservative and costly designs, but also cases of significant under-dimensioning, fracture, costly repairs and possibly environmental damage.

Methodically, I started out my small conundrum-resolving project with mapping out standard practices within banking, venture funds, energy companies, oil services companies, and early-phase technology companies.

The mapping was done through informal interviews with a few people from the industries in question (but no claim of statistical significance).  Some typical applications discussed were: repair yes / no, extend life time yes / no, start-up valuation, distressed bond valuation, produce yes / no, drill yes / no, enter market yes / no, and conduct R&D project yes / no.

Some patterns emerged:

  • Most companies use variations of standard DCF, with uncertainty taken into account through the discount factor (plus often identification of a worst-case scenario), optionality not valued quantitatively, and in case of company valuation, corrections for various balance sheet items.
  • Some organizations leverage more sophisticated models, including stochastic dynamic programming / simulation / decision trees.   These are generally found in the energy sector and finance.
  • There are also a number of Norwegian universities and research organizations with advanced capabilities / practices in this area, including Norwegian School of Economics (NHH; Bergen), Norwegian Business School (BI; Oslo), Norwegian Computing Center (NR; Oslo), and SINTEF (Trondheim).

However, the conundrum remained unresolved: why would the practitioner community not want to calculate correctly**?  A number of explanatory factors were suggested: Established in-house practices, need for more tough-to-estimate parameters (including subjective success probabilities), lack of competence, and short-comings of possible theoretical frameworks.

On a general note, one may predict that the observed reliance on plain-vanilla DCF (plus some qualitative considerations on the last page of the ppt) would result in incorrect / suboptimal decision, too low valuations (by ignoring value of optionality), too high valuations (if ignoring predictable competitive behaviours, and including unjustified optimism), and the need for adjustment factors to compensate for un-modelled factors.  However, the real issue is that one precludes the solid thinking and the qualitative insights that come with proper quantitative modelling, which leads to incomplete exploration and assessment of the solution space, unknown and sometimes unnecessary strategic exposure, and strategic sub-optimality.

One may at this point wonder what tools would be available to a practitioner that decides to venture beyond DCF: In practice, there are two approaches: real options theory, often calculated by stochastic dynamic programming (from the field of mathematical finance), and decision trees (from the field of decision theory).  There are pros and cons for each of them:


Going back to the DN article, what tool would I use if I were say the CFO of Norske Skog?  Short answer #1: A multi-player decision tree (or in game theory parlance, an extensive form representation or a game tree) that represents the sequential choices of the stakeholders (administration, shareholders, holders of unsecured and secured bonds) plus key uncertainties (including the price path of newspaper paper and the outcome of certain foreseeable court cases).  If I had done that, would I understand that Norske Skog would be taken to court by Citibank over this agreement with GSO and Cyrus Capital?  Short answer #2: Yes, absolutely.

I tend to complete my blog posts with a call to action.  In this case, it is simple: i) Identify a pilot case for the application of quantitative methods for valuing a project opportunity with significant optionality and / or competitive behaviours.  ii) Do an evaluation of the quantitative and qualitative insights extracted from such case, benchmarked against a plain-vanilla DCF analysis.  iii.a) If a repetitive situation (say a trading strategy for an energy company or investing for a VC), build up internal capabilities in the area of quantitative analysis.  iii.b) If a singular event (say a one-off R&D project or an M&A scenario), get assistance from an external consultant.

And if you think that this blog post is just a trivial, but somewhat interesting example of technology adoption theory in action (and now we are at the chasm between early adopters and early majority), I will not disagree.  But, let me revert to my example from the offshore oil and gas pipeline industry, in which a small number of industry practitioners, consulting firms, and research organizations (supported by major oil and gas companies) over the years helped the pipeline industry move from standards-based integrity assessment to full 3D finite element analyses.  I am hoping that this blog post could in a similar, though obviously much smaller way, contribute to the acceptance of full-scale decision trees and game trees as useful tools in industrial decision making.


*) Disclaimer: I have no inside information whatsoever about Norske Skog and this court case, so this is my personal interpretation of some newspaper articles.  I believe the court case is still pending.  Norske Skog’s administration and BoD may have private information that makes their strategy perfectly rational.  Besides, Norske Skog’s set of feasible solutions when they did this deal with GSO and Cyrus Capital may, according to a number of newspaper articles, have been very restricted.

**) Once you start to venture beyond DCF, ‘calculating correctly’ is a non-trivial affair.  As indicated by any standard text book in finance, one generally has to introduce draconian simplifying assumptions.  In practice, there are two schools: calculating exactly right using binomial lattices (or similar), but introducing such simplifications and going very black box, or calculating approximately right, based on modelling full complexity, but using a somewhat arbitrary discount rate.


Quantifying the value of uncertainty and optionality in M&A, market entry, and R&D projects; and the business case for real-scale, real-world decision trees

OLYMPUS DIGITAL CAMERA(Disclosure: This blog post is based on an actual client engagement.  The blog post has to some extent been anonymized / obfuscated, and has been approved by the client in question.)

According to garbage can theory, as formulated by Michael D. Cohen, James G. March, and Johan P. Olsen back in 1972, organizational choices can often be seen as the outcome of seemingly unrelated problems, solutions and decision makers coming together in a fairly random way.  This blog post, which is about valuing high-tech start-ups facing massive uncertainty and optionality through the use of decision trees, is a case in point:  A windsurfing friend of mine was in the process of developing his high-tech start-up, I was in the process of starting up a quantitative practice as part of my strategy advisory firm, and the two of us were already working on another joint project, completely unrelated to start-ups, high-tech, quantitative methods, and valuation.

My friend, let us call him Andrew, is electrical engineer by training, with long background from the oil and gas industry, the last few years on contract with a control system provider.  Andrew has an entrepreneurial  mind, and has ongoing investments in real estate organized as a holding company, say ABI.  However, one day he told me that he had formed a small subsidiary of ABI, let us call it Subsidiary, to commercialize a specific invention of his in the area of predictive algorithms for condition-based maintenance.   Early technical tests based on real data provided robust, but probably not statistical significant evidence that the technology worked.  He had already received a Norwegian patent for the technology and an international patent was in process.  Furthermore, he had received some funding from a governmental scheme and from an individual with professional background from commercial air transport.  I, with a doctoral degree in engineering and 10 years + of experience from similar industries, could clearly see the technical credibility and commercial potential of this invention.

Andrew was now facing some serious decisions: whether or not to invest in the case, on what market to focus, whether to partner with what company, whether to seek additional funding from new investors and in case what type of investors, and generally how to go all in or not.  He was also facing serious some major uncertainties, including what was the probability of technical success, what was the market, and what was the size of the addressable market.

Ultimately, he also wanted to know what his share in Subsidiary was actually worth, with or without an industrial investor.  I suggested at that point that we could apply decision trees for valuating his share, as such trees explicitly models the uncertainty, the optionality, and the complexity inherent in the sequence of decisions and uncertainties that aggregate to form a commercialization process for a high-tech start-up.

Using decision trees for valuing staged investments with uncertainty and optionality is of course nothing new; I learned about decision trees whilst I got my MS in Management at MIT in the early nineties.  Furthermore, most MBAs today with a specialization in decision theory or corporate finance have learned about them at university, as they are described in most intermediate-level text books.  However, fact is that with some exceptions (typically companies doing oil exploration or multi-stage R&D) very few individuals in Norway can use them effectively for real-size problems.

There are of course a number of alternative approaches to valuing such project, typically based on deterministic discounted free cash flow (DFCF), various types of multiples, or in the extreme, real options theory.  None of them provided the capability for modelling the complex sequence of decisions and uncertainties that Andrew and Subsidiary faced.

Equipped with DPL Professional from Syncopation Software, Andrew and I produced a full decision tree for the situation at hand during two half-day workshops plus some individual work on my side.  The fully expanded tree had around 8-9m end nodes, each of them with a different evaluation function.  The software automatically calculates the optimal decision at any point in the tree, that is, the decision that maximizes expected future DFCF from Subsidiary for ABI.  Unlike in a spreadsheet analysis, the optimum decision depends on history.  The optimization process starts in the future and works backwards towards the current.

outcome distributionFigure to left: Distribution of ABI’s share of DFCF from Subsidiary, red = go for it; green = sell IP as is.  One can see the founder’s dilemma, either go all in and either lose significant money (= the left part of the red curve)  or make it big (= the long red tail to the right), or just sell the IP as is and get a small, but unknown amount of money for it (= the green curve just on the positive side).

Key conclusions were: If the objective ABI is to maximize ABI’s share of DFCF from Subsidiary, Andrew should continue his endeavour to build up Subsidiary.  Subsidiary should also team up with an industrial partner, as getting a small bite of a big cake is better than a big bite of a very small cake (and Andrew neither has the financial muscles nor the professional experience to do this alone).  If at the end of 2017, Andrew has not been able to compellingly demonstrate the technical and commercial merits of his invention, he should consider selling the IP.  Furthermore, the value of ABI’s share of Subsidiary are NOK Xm, with an uncertainty range of NOK A-Bm.  Not only that, the option value of the investment = the value of the investment with flexibility – the value of the investment with a fixed / tied to the mast plan (= the base case) (= NOK YM) = NOK Zm.

NOK Xm was probably on the low side and I observed during the engagement that Andrew became somewhat disappointed.  Not only that, a separate valuation based on DFCF in Excel resulted in a value that was not far-off from NOK Xm, at least after some fiddling with the numbers.  Why was the engagement then assessed a success, and what did Andrew then learn from this engagement, beyond getting a theoretically sound and accurate valuation of ABI’s share of the DFCF from Subsidiary?

Not unsurprisingly, I think what really caught his attention during this engagement were matters only indirectly related to the exact value of ABI’s share of DFCF of Subsidiary:

  • How to price Subsidiary’s products?
  • To whom in the value chain (manufacturers, control system providers, asset owners, or service providers) to offer these products?
  • What is the addressable market, and how big is it?
  • Is this an incremental or disruptive innovation?
  • What will over time happen if an industrial investor gets a significant ownership share?

Could we have had discussions about the above issues without the use of decision trees?  My hypothesis is yes, but that our process for creating such decision tree and our subsequent discussions about it significantly sharpened our joint insight into what are critical decisions, uncertainties, and parametric sensitivities facing Andrew and Subsidiary.

Why was this a particularly interesting engagement for me as an advisor?  Because it nicely illustrated some key issues and benefits related to the use of real-scale, real-world decision trees for M&A, market entry, and R&D projects:

  • In principle, one could have done the decision tree in Excel. But with 8-9m terminal nodes and history-dependent optimal decisions (at any point the decision is the one that maximizes expected value of outcome, given what is already known), one is lead to conclude that any claim to be able to do this in Excel would be bogus.
  • It is theoretically justifiable to use a lower cost of capital in a decision tree, based on explicitly modelling uncertainty, than in a deterministic DFCF model, in which uncertainty or risk is taken into account through a somewhat arbitrary adjustment of hurdle rate.
  • For this particular case, the value of optionality (the Z referred to above) was not massive. In practice, optionality is significant only if there is room for scaling up, scaling down, abandon, or defer.
  • It would of course possible to get the correct value for EV and MV from a traditional DFCF analysis, by arbitrarily adjusting some numbers, say a hurdle rate or discount rate. Personally, I like the good feeling of calculating correctly based on a sound model.
  • Decision trees appear in practice to be the only way to correctly value such cases. Multiple-based methods are not relevant; deterministic DFCF models do not really capture optionality and uncertainty (except through discount rate); and real options theory is applicable only for the simplest problems.

To sum up: correct EV (from a decision tree) = EV (from deterministic DFCF model) + value of optionality + value of reduced risk adjustment of cost of capital – cost of uncertainty – value of unjustified optimism.  If no optionality and value of reduced risk adjustment of cost of capital = cost of uncertainty, then we are back to correct EV (from a decision tree) = EV (from deterministic DFCF) – value of unjustified optimism.

Which is why it is better to get the model correct from the start, with a decision theoretical approach, and which is why I wanted to share with you my insights related to this engagement.  Do not hesitate to contact me for a discussion about how decision trees can be applied to model complex decision, valuation, and investment situations in your organization, typically in the context of M&A, market entry, or R&D projects.



The case for quantitative methods in big bet industrial decision making and policy formation, and why quantitative methods ≠ Excel

Grim Gjønnes, General Manager Crisp Ideas

Grim Gjønnes, General Manager Crisp Ideas

Secretary of Defence Donald Rumsfeld said in a news briefing on February 12, 2002: “There are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns – the ones we don’t know we don’t know”

For this feat, Mr. Rumsfeld was generally ridiculed and in 2003 awarded the recognized ‘foot in the mouth’ award.  However, I take the statement to be perfectly OK, though expressed in a slightly awkward and imprecise modelling language, and lacking clarity of thought.  The real problem with Mr. Rumsfeld, and for which he will be remembered, was his inability to instil the right decision-making processes in the various organizations he led, and the disastrous decision-making blunders he made in matters involving Iraq and Afghanistan.

There is a strong interrelationship between thinking and decision making, and I will later in this article explain the relevance of the above statement and Mr. Rumsfeld’s decision-making processes to the topic of today, which is quantitative models for big bet industrial decision making.  On a personal note, I decided some weeks ago to extend my niche advisory firm with a quant practice for big bet industrial decision making.  The decision resulted from some Excel-based financial projections for 2015-2020.  However, like many other such decisions in the annals of business, the real analysis starts after the decision, when having to create a credible plan for converting a good idea into serious money, and a strong narrative for communicating the value that my company could generate for my clients.  This blog post is part of such analysis.

Let us first start with the concept of quantitative methods.  I will in the following use the concept about the use of quantitative models for decision making that goes beyond simple spreadsheet-based models.  Typical examples are models based on decision trees, linear programming, system dynamics, probabilistic models, game theory, and real option theory.  Of course, mostly everything can be implemented in Excel, especially if expanded with VBA for Excel, but the point is that for a number of reasons these models go significantly beyond what is typically found in the spreadsheets used for decision making in business.

In principle, it should be easy to identify typical applications (e.g., M&A, new product introductions, entry into new territories), typical benefit areas (rationality in the decision process, more accurate valuations, better uncertainty management, and organizational learning), and thus the business case for quantitative methods.  Fact is, quantitative methods are not generally used in big bet industrial decision making (with some exceptions, like pricing complex commodity contracts and deciding on whether or not to drill oil wells).

I will in the following argue that they are not used due to five myths about quantitative methods in big bet industrial decision making:

  • Myth 1: Quantitative methods do not creative value beyond DCF-based decision making.
  • Myth 2: One cannot quantify the unquantifiable.
  • Myth 3: Quantitative methods are easy, and most MBAs use them on a regular basis.
  • Myth 4: Quantitative methods are too complex, and do not create value.
  • Myth 5: Quantitative methods = quantitative models = statistics = big data (and we do not have big data).

Regarding Myth 1 (that quantitative methods do not create value beyond DCF-based decision making): The underlying assumptions in DCF-based decision making is to calculate deterministic after-tax cash flow (probably prudency adjusted) based on all decisions being taken before project start, transform that into an NPV using a risk-adjusted discount rate, and accept all project proposals with NPV > 0.  In theory OK, in practice incorrect and here is why: i) Technically inaccurate: use of WACC instead of project-specific cost of capital, no assignment of value to optionality or flexibility (e.g., the option to abandon, exit, or expand; always positive and often significantly so), and no explicit modelling of uncertainty.  ii) Conceptually incomplete: no modelling of staff morale, market dynamics and competitive behaviours (after all, strategy without conflict or competition is not strategy, but planning).  iii) Organizationally less relevant: not conducive to epistemic learning, not easy to share with or communicate to others.  In contrast, there are established and well-founded quantitative methods for dealing with (i)-(iii).

Regarding Myth 2 (that one cannot quantify the unquantifiable): This is of course true, but in practice most qualitative or textual models can be mapped into a quantitative framework, and most qualitative variables / relationships have quantitative brethren.  Often, by doing so, we end up with a sharper qualitative model or a more precisely defined qualitative variable / relationship.  And, regarding uncertainty, this is an area where quantitative models excel, in the form of probability theory and statistics.  One may indeed wonder why McKinsey, the strategy consulting firm, recently republished an article from 2000 about strategic decision making under uncertainty (see without any reference to a quantitative approach to managing strategic uncertainty.

Regarding Myth 3 (that quantitative methods are easy and most MBAs use them on a regular basis): Yes, most MBAs come out of business school with a good basic understanding of quantitative methods, say in game theory, decision analysis, and real options theory.  My experience is however, that confronted with three business realities they rapidly after graduation revert to traditional spreadsheet-based DCF analysis.  Reality 1: Reality is complex and does not look like the 3-10 node decision tree in the standard decision analysis textbook.  Reality 2: Most organizations have standard decision-making processes, and standard spreadsheet-based DCF models tend to be the lingua franca in investment appraisals and similar processes.  Reality 3: One most probably needs specialized modelling software to do say a full-fledged decision tree analysis.

Regarding Myth 4 (that quantitative methods are too complex for practical use): The universal law in modelling is that of Occam’s razor, or that among competing hypotheses that predict equally well, the one with the fewest assumptions should be selected (source: Wikipedia).  Contrary to popular perception and even though say system dynamics or decision analysis has a certain learning threshold, models based on quantitative methods tend to be significantly less complex relative to say spread sheets or qualitative models (in text), for a given level of predictive power.

Regarding Myth 5 (that quantitative methods is somehow big data): I fully concur with the case for big data, but big data is not quantitative methods, as we understand them here.  In fact, for most big bet industrial decision making, the amount of data available and necessary to make a decision is relatively modest, and understanding structural relationships is more important than discerning patterns in massive amounts of multi-dimensional data.  On a side note, I have previously explored this issue in another blog post, see

Having read my attempt to debunk some myths that hinder the adoption of quantitative methods in big bet industrial decision making, the interested reader may wonder: What can we do to make effective use of quantitative methods in our organization?  I have four pieces of generic advice:

  1. Use quantitative techniques to drive qualitative insights, and qualitative thinking to drive quantitative modelling. Essentially: Use quantitative methods to drive deepness of thought (and way beyond your simplistic DCF models).
  2. Develop a standard framework in your company for modelling optionality, uncertainty / risk, feedback loops, time delays, and competitive behaviors; it may lead to significant adjustments to your NPV calculations.
  3. Don’t trust your CFO when it comes to these matters; through their formal training they tend to see strategy as planning, uncertainty as something that can be managed through discount rate / cost of capital, and optionality and feedback loops as something to be discussed in a qualitative way and with no financial value.
  4. Recruit quant talent, grow your quant capabilities, and invest in proper tools (you will not be able to do this in Excel).

(1)  may indeed be a good starting point for any use of quantitative methods, and I have just finished the compilation of a causal loop diagram / influence diagram of a set of marketing activities, including the market’s response to same, for a client in software space, with no explicit ambition beyond getting a better understanding of this client’s marketing strategy (including strategic levers, sensitivities, interrelationships, and uncertainties).

Going back to Mr. Rumsfeld’s quote in the introduction, I am not sure what modelling approach could possibly have saved him (modal logic and probability theory would probably be candidates).  However, it is fair to say that quantitative methods do offer benefits in the areas of transforming  unknown unknowns into known unknowns, known unknowns into more precisely known unknowns, and generally awkward thinking into clear and precise thinking.  With more accurate models and sharper thinking come better decisions, and possibly Mr. Rumsfeld could have avoided his decision-making blunders.

Have a good day!


Big Data

We are what we search, and how dominating sales paradigms vary over time and across cultures

OLYMPUS DIGITAL CAMERAWe are what we search.  This is the underlying assumption of this blog post.  More precisely, we search for a specific term because we want to know more about this term, and we want to know more about this term because we are somehow new to a specific product, service, field, area, or discipline (hereinafter “field”).  If including also the effect of professional updating by people already in the field, search term frequency can be seen as a function of the inflow of new people to a specific field, plus of the current number of people in the same field.

We will in this blog post apply the above insight to the issue of studying how dominating sales paradigms vary over time and across cultures.  We will specifically look at four generic sales channel paradigms: proactive direct sales, reactive direct sales, indirect sales, and sales through digital channels.  To each of these channel paradigms we will identify a set of search terms that somehow are linked to the channel paradigm (but not to the others), and we will use Google Trends to measure search term frequency over time and across cultures.  To check the robustness of the approach, I will also include some search terms with assumed general validity across sales paradigms.

The cultural specificity of a search term was calculated as sum(fi^2)/sum(fi)^2, where fi is search term frequency for country i.  This index is always between 0.0 (highly fragmented) to 1.0 (all searches from one country).  This index is related to the so-called Herfindahl index for measuring market concentration.

For studying trends, we used the time period January 2004-April 2015, unless otherwise specified (see rightmost column in table below).  For studying cultural differences, we used the time period May 2012-April 2015.

Here is an example of typical output from Google Trends for search trends (blue = ‘inbound marketing’, red = ‘solution selling’, yellow = ‘spin selling’):


Here is an example of typical output from Google Trends for geographical distributions of search terms:


Here is what we found, for the above four sales paradigms and with 2-5 search terms per sales paradigm (X = start growth rate calculation, see right column):

Sales channel paradigm Search term Annual growth rate X-2015 Cultural specificity In case specific, territory Start growth calculation
Proactive sales Cold calling -5,0% 0,19 British Empire Jan 2004
Spin selling -6,6% 0,34 US, UK, India Jan 2008
Relationship selling -9,3% 1,00 US Oct 2008
Sales training -12,6% 0,13 Jan 2004
Solution selling -13,3% 1,00 US Aug 2010
Reactive sales Inside sales 3,6% 0,30 Jan 2006
Account management -2,0% 0,10 Jan 2004
Indirect channels Distributor -6,4% 0,05 Jan 2004
Reseller -9,7% 0,06 Jan 2004
Value added reseller -15,2% 1,00 US Jul 2004
Software reseller -17,2% 1,00 US Mar 2007
Digital channels Inbound marketing 34,3% 0,15 Oct 2009
Social media marketing 27,0% 0,07 Jun 2008
SEO 10,3% 0,02 Jan 2004
Adwords 11,2% 0,02 Jan 2004
Paradigm-neutral terms Interpersonal skills 3,8% 0,09 Jan 2004
Objection handling 0,2% 1,00 UK Mar 2008
Negotiation skills -4,5% 0,26 Jul 2006

Here are key conclusions:

  • The importance of proactive sales channels is globally decreasing with 5-15% per year, and hunter-style selling is becoming more and more a US and UK skill set.
  • Selling through indirect channels, like distributors and resellers, is also decreasing by 5-20% per year. However, if one looks beyond the numbers presented above, one will see that indirect sales is actually on the increase in a number of Asian countries.  This is consistent with lower barriers to trade in these countries, and a need to build distributions systems in previously locally served or under-served markets.
  • Manual, reactive selling is generally stable, with -5% to 5% annual frequency increase. This is somewhat strange, as one would assume that many reactive sales activities are being migrated to lower cost, better service digital channels, including pure self-service portals.  It could have to do with the professionalization of inside sales / account management.
  • Digital channels are rapidly becoming the dominant sales paradigm, with a 10-35% annual increase in search frequency. As expected, there is no particular search concentration, and Spain is for example on top for search term ‘inbound marketing’.  However, some digital channels, say inbound marketing, require significant technology sophistication, which is mostly found in mature economies.
  • Search term frequency for assumed paradigm-neutral search terms is generally stable over time, as expected. There is however significant cultural specificity, which is just a reflection of the fact that most such terms tend to be language specific.

There are a number of issues with the above methodology:

  • The selected search terms are either abbreviations or in English language. It means that similar searches in other languages have not been included in the analysis.  Example: ‘cold calling’, which translates into a roughly similar term in Spanish, ‘venta por teléfono’.  The effect is to over-weigh Anglophone countries this study.
  • Search term frequency is an unknown function of both inflow and level. Even with zero inflow of new practitioners, a channel paradigm could still live on for many years (until all the practitioners have left the profession, retired, or deceased).  In a way, why would a 40-year old successful spin selling expert Google ‘spin selling’?  Search term frequency could therefore be seen as a leading indicator, but not an accurate indicator.
  • Google Trends has a number of issues, including a normalization (i.e., absolute numbers are unavailable), temporal gaps (when search term frequency falls below a Google-defined threshold), and sensitivity to specific search terms.
  • It could be argued that conclusions regarding a specific sales paradigm depend on the specific search terms chosen for inclusion. This appears generally not to be the case.
  • When we are using Google Trends, we are looking at the world descriptively, not how it prescriptively should be, based on KPIs like win probability, revenue growth, RoI, or NPV.

But in general, no big surprises, and conclusions resonate reasonably well with most anecdotal evidence.

The interesting part of my story starts when we dig into some of the underlying issues that my analysis brings up.  Just some examples:

  • Could we use Google Trends as a general tool for market predictions and / or market sizing exercises? Hypothesis: Yes, and my guess, just as an example, is that a 12,6% annual reduction in the interest in sales training translates directly into a roughly 10-15% annual reduction in the sales training market.
  • What is the accuracy of Google Trends analyses compared with say case studies and various survey instruments? Hypothesis 1: Wrong question, conducting such interviews or surveys in a large number of countries over an 11-year time period would have been extremely costly and beyond the reach of most or all organizations.  Hypothesis 2: Competing approaches would have methodological issues of their own.
  • Is the sales community moving into a scenario in which proactive sales initiatives with excellent RoI or NPV will not be undertaken because shortage of hard-core sales talent and / or us being deluded by all this digital channels hype? Hypothesis: Yes.
  • Where to go if one really wants to learn the tools of the trade of hard-core sales, the hunter way (or to source the same type of sales talent)? Hypothesis: Stay away from Continental Europe (they have never been good at it anyway), and look to UK, Australia, Canada, US, India, and South Africa.  If in Europe, go the Netherlands or Germany.

But whether you are a management consultant endeavoring to analyse trends in a specific market, or a senior sales executive trying to identify, qualify, and hunt down massive deals, my message is this: We are what we search, and you ignore search as a most fundamental activity in any sales or purchase process at your peril.


Big Data

Why big decisions are about small data, and why big data is mostly about monetization (or many small decisions)

OLYMPUS DIGITAL CAMERAI read an interesting article some days ago in Information Management, by Navin Sivanandam and about “From big data to big decisions” (source: I then contrasted Sivanandam’s world with my own, and specifically an ongoing G2M campaign for a client in B2B space.  The contrast was stark, Sivanandam’s world is about randomized trials, statistical significance, and sample sizes of thousands or more.  My world is about various natural experiments, learning, and sample sizes of 1-10, but big decisions.

I decided to explore the apparent contrast between Sivanandam’s big data / big decision narrative and my own experience in the form of this blog, and specifically do some demything regarding big data.

Myth 1: Big data is about big decisions.  Big decisions are arguably about major investment projects, like entering a new market, launching a new product line, or acquiring a competitor.  The decision material for such Board-level decisions tends to be NPV estimates based on 10-100 parametric assumptions and complemented with a strong narrative.  Indeed, it is hard to imagine a CEO justifying a key decision to his Board with an output of a statistical algorithm based large amounts of data, rather than the NPV calculation and the narrative.  It is similarly hard to understand what role big data could have in this picture.

Myth 2: Big data is about decisions.  No, in my opinion big data is about monetization of information rather than about decisions, and to the extent about decisions, typically about operational decisions.  Let us look at some typical and frequently cited examples: Rolls Royce with its real-time system for monitoring of jet engine turbines, Google’s auctioning algorithms for Adwords real estate, and many credit-scoring algorithms used by credit card companies and banks.  One could of course say that this is about deciding to dispatch a repair crew with spare parts, deciding where to put what ad when, or deciding to accept or reject a credit card application.  However, the whole point is effective monetization of large amounts of data.

Myth 3: Big data is like a well-designed experiment in physics, based on a linear sequence of observations, hypotheses, experimental design, experiment, analysis, and conclusions.  No, in my opinion it is about an iterative process, in which one iteratively explores hypotheses, looks for patterns, draws preliminary conclusions, does further analyses, and goes back to the exploration of hypotheses.

Myth 4: Since statistical significance is a reasonable standard for any analysis based on big data, say an A / B test of a new web page (and the data often support claims of statistical significance), such requirement extends to other realms of business decision making.  A related myth is that you cannot do meaningful inferences in business without large sample sizes.  Let me shed some light on this issue based on some work for a former employer: I once (I think this was early 2001-2002) executed one G2M strategy in India and succeeded, we executed another one in Russia and succeeded less well.  Therefore, for me, it followed that the one we followed in India (a single distributor with sales force + direct channel) was good, the one in Russia (catalogue-based distributor / reseller) was less good.  Was conclusion correct and generalizable (in 2002): probably.  Was conclusion based on statistically significant evidence: probably not (as based on sample sizes of 1).

I made indeed an effort to gather information as to what constitutes valid basis for a conclusion, in a number of domains (specifically, venture community, business, law, engineering, science, hard science, and creative professions).  The standards differ widely, from 5 sigma in particle physics, 2-3 sigma in medical science, to beyond reasonable doubt in criminal law, to > 50% in civil law, to in accordance with standards in engineering, to around 60-90% probability of correct a priori NPV in business, to 1-2 successes out of 10 possibilities in the VC community.  However, one is led to respect that statistical significance is just one out multiple standards.

Myth 5: Number crunching, as in big data, is somehow a superior approach to gathering evidence for business decisions.  I would argue otherwise, and that you may want to invest as much in small-scale business experiments as you do in your massive Hadoop cluster with an R stats package.  For more information, I recommend Thomas H. Davenport: “How to design smart business experiments” (see  His conclusion is, for reference: i) understand when testing makes sense; ii) establish a process of testing (create hypotheses; design tests; execute tests; iterate); iii) build a test capability; and iv) create a testing mind-set.

So far, we have primarily had a descriptive perspective.  One could try to extend the analysis using a normative perspective.  I am not sure it would add additional insight.

In fact, and since we started out with Sivanandam’s article about “From big data to big decisions”, I would argue that if you want to explore the nature of big decisions, you should rather read the classics, like for example Ghemawat’s “Commitment—The dynamic of strategy”.  Ghemawat provides a clear perspective on what constitutes big decisions (those that require irreversible investments, in products, markets, or production facilities), how to make them right (by investing in durable, specialized, and untraded factors that are scarce and such that scarcity value can be appropriated), and the value of flexibility (or reversibility and optionality).  Think Apple and its launches of iMac (1998), iPod (2001), iTunes (2003), iPhone (2007), unt so weiter.  It is hard to see what role big data could have had in Apple’s big decisions and its subsequent and persistent success.

And to my readers: I would welcome any examples you have of big data having contributed to big decisions.  (There quite probably have been some, I am just not aware of them.)



The bimodality imperative in sales and business development

OLYMPUS DIGITAL CAMERAThere are currently heated discussions in the IT community about the need for bimodal organizations, one mode focused on predictability, performance, and efficiency (Mode 1); and one focused on agility, flexibility, and ability to operate in areas of great uncertainty (Mode 2).  Gartner, the IT industry analysts, is clear on this issue (source: “You need to become bimodal because one mode can’t answer the complex needs of the organization. It’s not nice to have. Gartner believes you must have both modes.”  The topic for today is whether this imperative also applies to sales organizations, end whether there is more to Mode 2 in a sales context than old-fashioned business development.

I have tried to track down the origin of the concept of bimodality, and it appears to have its roots in the sociology of science, and specifically the study of knowledge production.  It seems specifically to have originated in a 1994 book by Michael Gibbons, Camille Limoges, Helga Nowotny, Simon Schwartzman, Peter Scott and Martin Trow: “The new production of knowledge: the dynamics of science and research in contemporary societies.”  The authors’ thesis was that there is an alternative way of producing knowledge, based on “multidisciplinary teams [that] are brought together for short periods of time to work on specific problems in the real world for knowledge production” (source:  I am a layman in this field, and would appreciate if a reader with more scholarly background could shed light on the issue of origin.

Before we proceed, I must admit that I last year wrote a review of J. P. Kotter: “Accelerate”, a book about the need for dual operating systems in business organizations.  See  My review was for a long period voted the most useful critical review of this book on  However, the book was an easy target for a number of reasons, including that it was a rehash of Kotter’s earlier thinking and that Kotter seemed less willing to refer to previous thinking in related areas.  It is now increasingly clear to me that I was wrong in debunking Kotter’s concept of dual operating systems in business organization.

Here is my adjusted thesis: i) Bimodal capabilities are needed in sales organizations; ii) there is a need for a balance between Mode 1 and Mode 2 capabilities; iii) the concept of dual operating systems in organizations (in Kotter’s sense) is related to the concept of bimodality; and iv) Mode 2 in a sales context has limited to do with old-fashioned business development.

Regarding (i), here is why: I guess we have all observed them in real life, highly successful people who bring in large and strategically interesting deals through sheer agility, flexibility, and personal connectedness.  My favourite example is Mr. Zaoui, former head of European investment banking  at Goldman Sachs, who is believed to have worked on deals worth around USD 300b during his stint at the bank (source:

On the other hand and with reference to (ii), too much Mode 2 may not be effective.  Indeed, I have for many years worked with various early-phase technology companies and in some of them there were clearly too much Mode 2 (typically in the form of smart and charismatic, but also big ego and somewhat erratic founders), too little Mode 1.  In such organizations, everything revolved around agility, flexibility, and uncertainty, when it could be argued that what these businesses needed, were high-performing and predictable sales machines (which is how highly successful organizations like Aker Solutions, IBM, Oracle, Microsoft, and Accenture organize their sales activities).

A related side issue is whether there is an optimum mix of Mode 1 and Mode 2 for any organization.  My take on this issue is that no, there are a range of feasible mixes, and different companies in apparently same industry tend to select different mixes depending on for example previous organizational choices, available resources and current capabilities, seemingly without discernible relationships between exact mix and financial performance.

Some readers may argue that the need for bimodality in sales organization depends on the maturity of the business or the industry.  Yes, I agree, in many mature industries the focus should be on building market share and generating margin from current customers, which is the stuff that Mode 1 is made of.  On the other hand, Eastman Kodak was fairly Mode 1 (as I understand it), and it is well known how that story ended.

So let us get specific about Mode 2 in the context of a sales organization: It is by definition about agility, flexibility, and ability to operate in areas of great uncertainty.  Where in the organization does one build these capabilities, and does share share, co-locate or split Mode 1 and Mode 2 resources in the organization? McKinsey’s 7S framework (structure, strategy, systems, skills, style, staff and shared values) may be used as a guiding framework for such endeavour, but really it is about making hard and explicit mode-specific design choices in areas like reporting lines, sales reporting, incentive systems, and recruitment.  At a more fundamental level, it is (with reference to shared values in the 7S model) about instilling an acute awareness of strengths, weaknesses, opportunities, and threats at all levels in the Mode 2 part of the sales organization, and a corresponding eagerness to exploit strengths and opportunities / deal with weaknesses and threats.

If you believe that the bimodality imperative applies to your organization, here are four issues that you may want to consider:

  1. Do you have the required Mode 2 capabilities? What is the strategic rationale for building Mode 2 capabilities and how to calculate the NPV of agility, flexibility, and ability to operate in areas of great uncertainty?
  2. Do you have the right balance between Mode 1 and Mode 2 capabilities? Do you know your revenue split between Mode 1 and Mode 2?
  3. Where are you going to build Mode 2 capabilities in your sales organization? Born or bred?  Who is your Mode 2 champion, the CEO, the VP Business Development, the business development manager, the product managers, or someone else?
  4. Do you believe Mode 1 and Mode 2 should be based on shared, co-located, or split resources?

But conclusion is clear, your sales organization need to become bimodal, and Mode 2 goes significantly beyond old-fashioned business development.

Have an excellent Easter!