Possible Insight

While my goal is to eventually apply the Market Space model to large enterprises, I’m going to begin with startups. Obviously, my work at RSCM makes startup close to my heart. And most large enterprises were new entrants at some point, so analyzing the birth of firms seems like it should lay some crucial groundwork. (For previous posts in this series, see here: one, two, three, four.)

Stage

Market Space really cleared up some issues I have with precisely determining company stage. After looking at hundreds to thousands of startups, I’ve found that time in existence, number of employees, product quality, investment amount, and even revenues all fail to capture crucial nuances.

But I can make much better sense of stage by focusing on where a startup is in its search of the Feature Subspace fitness landscape:

1. Wandering the Casino. Roll of quarters in hand trying to figure out which bandits to play.
2. Watching the Payouts. Has played some bandits but doesn’t have enough data to know whether any of them are systematically paying off.
3. On a Roll. Has found at least one bandit whose payoff seems consistently promising.
4. Climbing the Hill. Has played enough bandits enough times to accurately estimate both their individual payoffs and the landscape’s local gradient.
5. Scaling the Summit. Is climbing a very steep face towards a peak and its estimated altitude is relatively high by historical standards.

Of course, these stages aren’t always linear. A startup can advance to stage (2) or even stage (3), then discover the bandits it has been playing actually have poor payoffs. So it loops back to (1) or (2), i.e., it pivots. Also, many (most?) startups that reach stage (4) discover that the gradient isn’t steep enough that they could ever reach stage (5).

In my opinion, traditional early stage VCs really want startups at (4), with a reasonable certainty that (5) is at least a possibility. They are willing to take a chance on ones at (3), if they see signs of likely progression to (4). In contrast, RSCM wants to invest at stage (2). We are willing to take a chance on (1), if we see signs of likely progression to (2).

Of course, the signs that a startup will progress to (4) are all about whether a good market is developing ,while the signs that a startup will progress to (2) are all about whether the founders are likely to deliver some sort of initial offering. So the predictor variables that VCs and RSCM care about are quite different.

Stakes

The Market Space model also clarifies how RSCM and traditional VCs fundamentally play for different stakes. We really want startups that will establish themselves at penny, nickel, or dime value slots–at least initially. VCs only care about ones playing, quarter, dollar or ten-dollar slots, the ones that could pay off more than $100M.

Interestingly, target stakes can affect the target stage. VCs might back companies playing higher value slots at an earlier stage (e.g., pharmaceuticals, cleantech, semiconductors, etc.). They know that, if that company makes it to stage (4), there’s almost certainly a stage (5). Conversely, RSCM might back companies playing lower value slots at a later stage (e.g., Web, apps, cloud, etc.). We know that such companies have lower risk, but probably aren’t as attractive to VCs who would otherwise bid up the valuation.

Essentially, RSCM believes the fitness landscape has a lot of small hills.  VCs believe that it has a fair number of sharp, high peaks. Note that these are not mutually exclusive. There may be some areas of hills with no high peaks, some areas of peaks with no surrounding hills, and some areas with the two mixed together.

Now let’s apply Market Space to a less obvious case, Lean Startup. From my perspective, the Lean Startup methodology focuses on regions of the fitness landscape where smalls hills occur close to high peaks. Or, using the other analogy, situations where nickel slots are near dollar slots with correlated payoffs. So founders play the nickel slots with a “minimum viable product”, iterating to learn the shape of the local terrain. When they get traction, they climb the steepest gradient and play the dollar slots there.

Now, Lean Startup makes stronger assumptions about the terrain than either RSCM or VCs. It requires hills, peaks, and that they are close together. For example, it will fail to detect isolated high stakes, high-payoff peaks that VCs would love to find. Moreover, Lean Startup appears to at least imply that following its precepts can help companies achieve large exits.  But that path will fail if there are lots of places with decent hills, but no high peaks nearby. So I would roughly describe the relationship as RSCM’s target terrain in Market Space being a superset of Lean Startup, whereas traditional VC’s terrain partially overlaps Lean Startup’s.

Risk

Startups at different stages face different risks.  For convenience, let’s call ones at stages (1) through (3) “Searching” and ones at stage (4) and (5) “Optimizing”. By simply thinking about how startups at these two phases move through Market Space, we can see the sharp differences in the kinds of risk they face.

In the Searching phase, a startup faces three primary risks. First, the area it has chosen to explore may be completely devoid of hills or peaks. Second, it may be unable to explore much territory. Third, it might fail to find any hills before it runs out of money. I believe the first risk is completely unpredictable and uncontrollable. The second two might be manageable by screening for founders who have proven the ability to deliver new products quickly and can manage a limited budget.

In the Optimizing phase, a startup also faces three primary risks, but different ones. First, the terrain may shift so that the peak it currently occupies is no longer a peak. Second, it may have overestimated how high the peak actually is. Third, it may be unable to climb the peak. The first risk is fairly unpredictable and uncontrollable. But the second may be manageable through detailed market research and the third by screening for an executive team with a history of rapidly scaling operations.

Obviously, applying the wrong risk mitigation strategy in either phase will have bad consequences. Research on a nonexistent market and spending huge dollars scaling an unproven business won’t pay off. And we don’t want the executives at an Optimizing startup to focus on releasing a constant stream of low-budget new products.

Skills

The different types of risks when Searching and Optimizing require strengths in different kinds of decision making. When Searching, a startup needs to maintain option value and preserve resources to pursue alternatives (unless it has happened into an incredibly good patch of Market Space, which should be very rare). When Optimizing, delay has large opportunity costs and undeployed resources aren’t generating value. So those startups need to spend money decisively and quickly (of course, they also need to spend efficiently).

But it’s not just management decision styles where skills differ. Searching Startups and Optimizing Startups also need personnel with a different concentration of skills. When Searching, generalists are more valuable because they can adapt when the startup moves to different terrain. When Optimizing, specialists are more valuable because they can cover specific terrain faster.

So engineers with experience in a wide variety of tools and sales people who have targeted a wide variety of industries are more valuable earlier (holding total experience and competence constant of course). Once a startup is Optimizing, it wants engineers and sales people with experience focused in the precise tools and industry of that startup.

Note that this conclusion does not mean startups want to replace their early personnel with more specialized ones. The generalists who helped develop a particular startup’s products and business will have accumulated skills and knowledge highly specific to that company that an outside specialist probably can’t match. But an Optimizing Startup will tend to want its marginal employee to come from a more focused background.

Predictions

I think it’s good practice when developing a hypothesis to make at least a few predictions that someone could potentially test. This discipline ensures that I’m not telling “just so” stories and revising the narrative retrospectively. It also forces me to explicitly identify times when I’m revising my model.

With that in mind, here are few startup predictions:

• If you select a random startup at stage (1) or (2) with completely inexperienced founders, take their idea, and replace them with founders experienced only at stage (4) and (5), it won’t make much difference. In other words, founders with only (4) and (5) experience will not have a greater chance of taking a new startup from (1) to (3).
• Market size and revenue projections at stage (1) or (2) will not correlate (at least positively) with ultimate success. Startups at this stage can’t know enough about the local fitness landscape to predict how large the eventual market will be.
• Too much money at stage (1) or (2) can be harmful.  With lots of money, a startup may be tempted to start “scaling”, i.e., hill climbing, before it has enough data about the local landscape to accurately estimate the gradient.
• When the terrain is relatively stable, pivoting should pay off better because you can count on the gradient remaining constant.  When the terrain is relatively unstable, staying the course could pay off better because the chances are just as good that the current location will spontaneously improve.  There might be two proxies for terrain stability. Nearby large companies will have stable growth when the terrain is stable.  Recessions will lead to more instability.
• Successful startups at stage (1) and (2) will have defered big product direction decisions more often than successful startups at stage (4) and (5). At stage (1) and (2), delay preserves optionality, which may counterbalance the lower effectiveness of committing fewer resources.

So that’s my first pass at a Market Space analysis of startups. In the next post, I’ll see if I can produce similar types of insights for large enterprises.