1. Samir Mikati, MIT Engineering Systems Division ESD 71: Engineering Systems Analysis for Design Professor Richard de Neufville December 9 th, 2008 Using Flexible Business Development Plans to Raise the Value of High-Technology Startups 1

2. 2 Slideshow format:  Introduction  Development of an Application Portfolio  Comparison of Analysis methods: Decision Tree v. Lattice  Conclusions

3. 3  Introduction  Development of an Application Portfolio  Comparison of Analysis methods: Decision Tree v. Lattice  Conclusions

4. 4 Introduction: The Engineering System  Engineering Systems can take on different forms  A space station, computer motherboard, power plant, irrigation system, World Wide Web, complex organizational setups  System Analyzing: a high-technology startup  Central question: How does one model the development of such an uncertain system? 1. Identify system attributes/boundaries 2. Identify key uncertainties 3. Use a decision tree and/or lattice to model development of startup  Builds a business development “roadmap” that:  Proactively incorporates uncertainty recognition  Uses flexibility to mitigate/take advantage of uncertainty * Must recognize advantages/limitations of DT/lattice modeling methods to understand business development roadmap

5. 5 Introduction: Motivations  Schumpeter (1942): “Creative Destruction” process  Innovation as the seed for new value creation and the destroyer of obsolete products and services  Chesbrough (2002): Corporate Venture Capital  CVC investments based on: 1. Objective  Strategic (increase/migrate capabilities)  Financial (make money carrying out strategic objective) 2. Strength of operational ties (between opportunity and company, this is difference between CVC and Venture Capital)  CVC provides an essential window into new strategic opportunities (innovations)  The Problem: How to correctly model and hence value highly uncertain CVC investments?  Decision Tree and Lattice valuation  2 methods that proactively incorporate uncertainty into system valuation  Both have limitations, should understand them

6. 6 A Paradigm Shift  Faulkner (1996): Way in which we value technology developments has progressed from:  Least accurate, deterministic heavily discounted DCF method  Most accurate, uncertainty recognizing and flexibility incorporating appropriately discounted DCF method  Example of evolution of DCF methods (Faulkner 1996): Decision Tree Valuation Method Incorporates flexibility to better handle uncertainties

7. 7  Introduction  Development of an Application Portfolio  Comparison of Analysis methods: Decision Tree v. Lattice  Conclusions

8. 8 Application Portfolio 1: Describing the Engineering System What is the system, what does it include and what does it exclude?  ENI (an Italian state-owned petrochemical company) evaluating a specific technology startup company based on its strategic/financial value  Analyzing a solar-thermal technology startup’s potential future value What are its principal design levers or variables?  Amount, if any, of investment placed during a step in the development of the technology What are the benefits of this system?  The value of merging the expertise of a startup with the vast capabilities (financial and project based) of an investing company  Investing firm’s ability to provide startup with enough capital and other resources to rapidly launch product development and commercialization (ability to pursue “call option” on expansion of successful technology)

9. 9 Application Portfolio 2: Key Uncertainty Identification Critical Uncertainties incorporated into model: 1.Market size (use carbon tax rate as a proxy) 2. Success level of technology development (how good is product?) Future fuel shares (observe projected size of renewables market) Source: IEA 2006, pp. 73 Past and predicted future best lab cell efficiencies Source: Stanford E 104 Lecture, Benson

10. 10 Application Portfolio 3: Defining the System  The Players  Investing Firm (ENI, a large Italian energy company)  CSPond venture (a solar-thermal hi tech startup based in MIT)  Basic Description of Technology (Slocum 2008): CSPond’s salt-filled tank (Slocum 2008) CSPond illustration of basic concept (Slocum 2008)

11. 11 Application Portfolio 3: Defining the System The Fixed Design: A fixed business development plan Analyze the uncertainties relevant to startup’s successful development, and then create a fixed “optimal” business plan A Flexible Design: A dynamic business development plan A flexible, dynamic model gives management the ability to decide on the level of investment in the technology after more information is learnt. The following figure is a part of the decision tree which illustrates management’s ability to decide on investment strategy given uncertainty in the carbon tax: Fixed investment decision Flexible investment decision

12. 12 Application Portfolio 4: Building a Decision Tree  The Method 1. Build a basic sequence of stages in the development of the system being modeled 2.Insert the first critical uncertainty and decision node pair i.Uncertainty node: turns a linkage between two development stages into an uncertainty node with several different outcomes ii.Decision node: placed after the uncertainty node (reflects a decision management can make that will minimize the loss in system performance associated with unfavorable outcomes, and improve system performance by taking advantage of situations where the outcome is favorable) 3.Iterate (2.) n-times (according to desired complexity) 4.Analyze Decision Tree  Use “fold back” method to prune least attractive decisions, leaving one optimal decision for every decision node (now have a business roadmap)

13. 13 Application Portfolio 4: Building a Decision Tree The Tree  Inputs: an intelligently linked model  Critical parameters (i.e. cost of the flexible/inflexible contract, discount rate, length of company/product operation, and probabilities associated with the carbon tax and technology development uncertainties) defined as variables for easy manipulation  Structure  2 Decision-Chance pairs: Decision 1Chance Event 1Decision 2Chance Event 2 Flexible: Buy right to develop technology and produce product in 5 years (for 15 years). Can decide how heavily to invest depending on new information obtained @ year 5. Carbon Tax is: High Low Nonexistent *this chance node is a surrogate for “market” uncertainty identified in AP 3 Decide to either invest: High Low No Investment Technology is: Very Successful Successful A failure * this chance node represents “technology” uncertainty identified in AP 3 Inflexible: Buy technology and commit to producing products for 20 years (given investment strategy) Cannot decide, must choose an investment strategy @ year 0. Here, we have decided in the “Low” investment strategy.

14. 14 Application Portfolio 4: Building a Decision Tree Visual of Decision Tree  Red lines indicate best decision for each decision node  This gives a “roadmap” of optimal decision for any uncertainty outcome

15. 15 Application Portfolio 4: Building a Decision Tree Outcome distributions and VARG

16. 16 Application Portfolio 4: Building a Decision Tree Key Values Q: Which business development plan (flexible v. inflexible) is better? A: It depends on which value one is most interested in: ($, millions) Design Which is better? FlexibleInflexible ENPV 182.074609 8 152.75405 9Flexible Minimum NPV-8-34Flexible Maximum NPV587557Flexible Initial CAPEX6050Inflexible NPV/CAPEX3.03453.0550Inflexible

17. 17 Application Portfolio 5&6: Creating a Lattice model The Method 1. Choosing a “Start value”  Look at the revenues that would be accrued from a very small time after the start of operations of the company 2.Choosing “up” “down” and “p” values  P: likelihood current state will increase in value in (present+1) state  Up/down: magnitude of increase/decrease in (present+1) state  Initially choose an arbitrary set of values for the up, down and p values simply to allow the lattice to be created  Once all necessary lattices created, we can manipulate these values to create a model that has similar properties to the decision tree 3.Dynamic Programming  “Looking forward method of analyzing a lattice”  Can create a “flexible” lattice by a “stop operations” put option 4.Choosing 1 decision  Lattice method works best when incorporating 1 “flexibility” decision  Our decision: “whether to continue company operations (contracting CSPond projects) or halt them” (put option)

18. 18 Application Portfolio 5&6: Creating a Lattice model The Method: To build our model, must build: 1. “Profits” lattice  Small profits in beginning, grow according to up, down, and p values 2.“Probabilities” lattice  Depends on p and (1-p) values 3.“Cashflow” lattice  Values from “profits” lattice – yearly operating costs (fixed+variable) 4.Baseline “Yearly contribution to ENPV” lattice  Multiplies yearly discounted cashflows by appropriate p, sum all cell multiplications to get ENPV 5.“Dynamic programming-based inflexible” lattice  “looking into the future” method; ENPV should be the same as in (4) 6.“Dynamic programming-based inflexible” lattice  Same as above, but now insert “put option” flexibility to stop operations  When stop operations, incur only fixed costs  Similar to flexibility option in decision tree: amount invested 7. “Continue or stop” company operations” lattice

19. 19 Application Portfolio 5&6: Creating a Lattice model Key Results:  Note: this VARG only models first 6 years of company operations Downside risks mitigated Take advantage of upside

20. 20  Introduction  Development of an Application Portfolio  Comparison of Analysis methods: Decision Tree v. Lattice  Conclusions

21. 21 Advantages/limitations of “decision tree analysis” modeling  Structure:  Uses a “tree” approach that iterates an uncertainty-decision node sequence Strength- clearly illustrates a visual roadmap of the many different paths that the business development could take in the future Strength- the shape (i.e. number of decision options and uncertainty outcomes) is not constrained by any limitations (in the lattice model, we are constrained to a binomial decision process)  No regularity constraints: While the binomial model is limited by “regularity” (it assumes that the diffusion process is “stationary” in that the probability of the next state remains constant throughout the periods considered), the decision tree is not limited to this constraint  No outcome constraints: A major limitation of the lattice is that it can only generate lattices with purely positive or negative state values. the decision tree does not require additional analysis: a user simply inputs the relevant outcome values in the end stage, and conducts a folding-back analysis to evaluate the tree  Conclusion:  Decisions analysis approach is the more flexible approach because of its lack of regularity and outcome constraints, as well as its “unlimited” uncertainty/decision node outcome possibilities. Decision trees are more suitable when we have complex (i.e. more than one outcome) and irregular (requires changing of probabilities) processes.

22. 22 Advantages/limitations of “lattice ” modeling  Structure: Similar to a decision tree analysis in that it uses a “tree” approach to model uncertainties and decisions, thus developing multiple paths.  Key limitation: in order to keep the number of states increasing linearly with the number of stages, we must assume path independence (since in this case all paths to a state have the same result)  In our system, barring the extreme cases (such as no projects acquired, or a truly explosive growth of work that causes relocation to a much bigger office) we can assume that this system is a relatively path independent process. That is, the order in which our company grows (i.e. slow growth then faster growth, or vice versa) will not affect the current state since we can adjust (to a certain degree) the number of employees working; hence our system can respond to changes without being fundamentally altered.  Limitation: the evolution of one state can only be into two future states.  Solution: if we want to model several outcomes (states), we can do this by introducing several stages (which will progressively double the number of states).  Limitation: “curse” of regularity.  In the lattice model, the diffusion process is necessarily stationary: the probability of “up” or “down” states does not change with time.  Limitation: only positive or negative values generated  Solution: Define our initial lattice to have only positive values. Then transform these necessarily positive values into potentially negative values by subtracting the relevant operational costs in each state

23. 23 Advantages/limitations of “lattice ” modeling  Powerful advantage: lattice approach models single decisions over many stages very effectively  It is thus useful in its ability to assess, at any given year and state, whether the option to stop investment should be pursued (decision tree analysis cannot do this as effectively, would have to create a very complicated tree)  How relevant is thus advantage to our solution? If we can simplify our model to an “invest” or “stop investment” scenario (currently it has 3 levels of investment), then this advantage could be used to gain detailed (yearly) information about when to continue/stop investment

24. 24  Introduction  Development of an Application Portfolio  Comparison of Analysis methods: Decision Tree v. Lattice  Conclusions

25. 25 1. There are of course advantages and limitations for the use of either modeling method. The key to successful use of these models is in understanding the mechanism behind which they operate  Models are not black boxes that somehow magically transform input data into perfectly correct output results 2.It is a modeler/user’s duty to understand exactly what the strengths and limitations are of the modeling technique in order to truly gain values from the results. 3. In the decision tree model, based on our extensive financial analyses (36 cashflow statements for each decision tree outcome) and model setup:  The flexible and inflexible business development plans predict an expected value of ~$185 M and $153 M, respectfully. This correlates to an improved system performance (NPV of the CSPond-based startup) of ~ 21% 4.How flexibility was incorporated to raise system value:  By giving management the flexibility to vary the amount of investment they make into the development of the technology at any given stage, they can use new information learned to make better decisions Conclusions

26. 26 Conclusions 5.What is the “best” design: Depends on user  The optimal choice may change depending on the circumstances of the user (i.e. if cash-strapped then minimum initial CAPEX very important)  The only criteria in which the inflexible model performs better than the flexible model is in the CAPEX required (because the cost of a flexible contract versus a similar fixed contract is at a 20% premium) 6. “Systems thinking”  The interrelatedness of so many different factors that must be synthesized into a coherent model that appropriately uses all the relevant information requires two equally important things: 1. A solid grasp of the “big picture” of the system being modeled 2. It is insufficient to only be aware of the overall picture and not have a firm command of pumping out the “nitty gritty” calculations. The exercise of going through “nitty gritty” financial and other data/variable manipulation/generation provided “hands-on” exposure to the entire modeling process

27. 27 References Chesbrough, H.W. Making sense of corporate venture capital. Harvard Business Review, March, 2002. Christensen, Clayton M., Raynor, Michael E. (2003). “The Innovator’s Solution” Harvard Business School Press. Faulkner, T. W. (1996) “Applying 'options thinking' to R&D valuation”, Research- Technology Management, vol. 39(3): 50-56. “World Energy Outlook 2006” International Energy Agency. Head of Publications Service, 9 rue de la Federation, 75739 Paris Cedex 15, France. Accessed: October 12 th, 2008. Schumpeter, Joseph A. (1942) “Capitalism, Socialism and Democracy” Harper Perennial. Stanford University, E 104, lecture Slide #11, Lecture #16 Slocum et al. (2008). “Concentrated Solar Power on demand: CSPond” (non-published technology description of CSPond technology)