Reflections on start-up life: Week Four

10… 9… 8… 7… 6… 5… 4… 3… 2… abort, abort — Houston, we have a problem.

That’s pretty much how close we got to launching “Twendly” our automated Twitter list bot, this last week before pulling the plug.

We decided to divert our attention into this tool for a few reasons you might recall.

1. Proves the engine in a production like environment.
2. Provides a hands on demo customers can actually see it working.
3. Lets us access the taxonomies that people build to see how the actually use the Engine.
4. Done right, should lead people back to our site — it’s broadly targetted, but we know a lot of E2.0 types are on Twitter and may be interested in HiveMind.

These were (and still are) good reasons for doing it, but as we got closer to launch we realised that we had become fully diverted from problems to do with expertise location and instead were becoming fully focussed on engineering problems that had everything to do with Twitter. After a long hard discussion, we decided that there was nothing to be gained by solving the Twitter problems, because we could not deliver a tool that effectively show-cased what we were doing because of these limitations. At about 80+% done, we terminated the project and pushed it aside to get back and focus.

For those that are interested, the fundamental problem is that to do effective expertise location we need a complete set of experts, otherwise we report incorrectly — we either had to “swallow Twitter whole” or resort to sampling with it’s attendant statistical problems to really show we could solve the problem. Our solution was to limit the input based on selection criteria, but in effect we were becoming “you show us the experts and we’ll tell you if they’re an expert” — not what the core engine is about at all.

Still, it’s not a complete waste, although the timing was probably off! The biggest lessons came from trying to productionise the engine. Over the course of the proceeding three weeks we’d managed to create an environment that was too complex to actually replicate easily and needs some simplification / reengineering efforts. We’d also confused the core engine and the implementation — in our rush to get Twendly up, we’d hacked in Twitter code which we need to take out and properly modularise. So the act of doing (trying to shift the site into production) taught us a lot on where we actually are.

We also let our integration efforts slip which caused problems as I began to get roadblocked because I couldn’t test the site effectively and it no longer ran on Windows (rectifiable but we didn’t want to bother in the heat of the moment) and we had no integration environment.

We’ve also gained a lot more clarity on the problem of expertise location. By driving an actual implementation of the engine, the theoretical niceties were exposed and the core of what it really does is actually a lot simpler than some of the “fluff” we started adding around the outside.

Actually we are more excited than ever about the core engine and we’ve already got our next iteration scheduled and planned (we spent Friday afteroon doing this instead of waiting until Monday). So the Alphington release was a bust, but the Fairfield release is already underway and brings us back on track in the week before Christmas to having a real tool we can show people and seek their feedback — this time we think it’s a more realistic proposition that may put us on a path to revenue shorter than we might otherwise of found.

Highlights?

• Continued feedback and sign-ups for the beta — now at near enough to 100.
• Finally resolving the OAuth problems.
• Sense of excitement in seeing a real implementation of our engine come together.
• Renewing some old acquaintences and validating the approach for the Fairfield release.
• Business relationship continues to go well, we’ve had some tough discussions and always come through together with flying colours.

Lowlights?

• Server problems — frustrations that were largely outside of our control.
• Nagging feeling all week that Twendly wasn’t going to work, but not being able to nail it.
• Finally putting Twendly down on Friday and the nagging feeling we’d just wasted a chunk of time (actually this has diminished as I started the Fairfield sprint over the weekend — we’ve learnt a LOT and the next cycle is screaming along already).

If I could change on thing / goal this week?

• Back to getting something we can demo live to customers ASAP.