Elevator pitches and reinforcement learning
Startup events and parties serve one good purpose - you can hone your elevator pitch through "reinforcement learning"
Last night, Manu and I were at a party organised by a Venture Capital firm. This was possibly our first “startup-related social event” since we started Babbage. I had a good time, got to meet (and reconnect with) lots of people, and my voice is hoarse today from all the talking last night.
It was a large party, held across a large lawn, and all guests had been given stickers, to write our names and company names and stick on ourselves. The last was an important detail because the moment you approached someone to talk to them, you would immediately know some priors about them, and if you knew something about them or their company, there was a readymade conversation starter.
The thing with running a three month old company which hasn’t yet released built its product is that people have no clue of what you are doing, and in a setting like last night, they will just ask you. And so I found myself, many times over, having to deliver an “elevator pitch”.
I remember my first such conversation from last night - I said something about what our company is doing, and immediately figured out that it sounded rather unimpressive. Now, we’ve spent a large part of the last month raising our seed funding, and talking to potential investors for the same. This has largely been by way of pre-arranged formal meetings (lasting 45-60 mins) which has given us enough time to say our spiel.
Now, suddenly being repeatedly put on the spot in a social situation, I struggled. Each time I said something about what we do, I would find that unimpressive. So the next person I met, I would describe our company in a different way. And would find that unimpressive as well.
Soon I figured out that parties like this are the best way to hone your “elevator pitches” because you can effectively do "reinforcement learning” on your pitches. You choose one random pitch, inflict it on the counterparty, and then see how they react. If they react well, then you use that pitch more often. If they are unimpressed, you use it less often. Classic reinforcement learning.
The pertinent point is that while you meet a lot of potentially important people (potential investors and customers) at such parties, they are highly unlikely to remember your elevator pitch. The purpose that is served is that they have put a name to your face (and vice versa), and so the next time you meet them (either in a formal or informal setting), there is a prior to that conversation.
And because (or so I think) people don’t really remember your elevator pitches, events like these are ideal to perform reinforcement learning on your pitch. You are pitching to different (and diverse) people, and fairly often. That you are face to face tells you their feedback immediately, and so you can learn. Done cleverly, by the end of the evening, you’ll have a great elevator pitch.
Except that I don’t think I did that great a job last night. As the night went on, I tired. I started seeking comfort in people who I already knew (or had met), to save myself the trouble of pitching once again. I started having longer conversations with people.
If you were to meet me now and ask me what our company is doing, it is highly unlikely that my answer would have taken into account all of last night’s learnings! At the next such party I’ll go to, I’ll probably start my RL all over again.
The only difference is that since I know that startup parties are a great place to perform RL on the elevator pitch, I’ll make a more conscious effort. Maybe if you meet me three parties from now, I’ll have a coherent pitch for you.
Actually - we already have one, in writing (this is the “one pager” we’re sending investors before our introductory conversations). Just that I find it harder to communicate the same verbally.
PS: The other thing that happened last night was not guessing correctly the other person’s domain knowledge. I remember at least one conversation where the other person got disinterested the moment I mentioned “AI”. And a subsequent conversation where I had grossly underestimated the technical and domain knowledge of the person I was talking to. The feedback you get from reinforcement learning (in real life) is not always perfect.