Michael Idinopulos suggests on a SocialText post that the old-skool method of implementing and evaluating systems – small pilots – isn’t so appropriate for tools that depend on a network effect.

When companies pilot Enterprise 2.0 tools with small subsets of their organization, they’re not testing collaboration with a representative sample. An artificially constrained pilot is always a poor representation of post-pilot collaboration, because the range of potential interactions is so limited. The value delivered to each individual participant is exponentially smaller than it would be at full scale, and the ways that people will use the tool are different.

If a pilot is struggling, his recommendation is to make the pilot bigger rather than shutting it down.