Capacity planning for PDC
There's an interesting discussion about PDC over at Yahoo Groups' Windows Technical: Off Topic forum. Kevin Hegg writes:
The last three PDC's were overcrowded, with the last PDC being the worst. The solution seems pretty simple to me. Stop selling more admission tickets than you have available seats. Microsoft has had enough events at the LA Convention Center to know what it's capacity is. There really shouldn't be any surprises any more.
All of us involved with PDC 2003 remember that the sold out conference meant we had overfilled sessions. And if we didn't remember, all it takes is a quick browse through the written comments in the session evaluations to remind us ;) As Content Owner for PDC 2005, I’ve spent a lot of time looking over the 2003 attendance data. And my conclusion is the exact opposite of Kevin's: there is no way to avoid surprises. The best we can do is expect surprises and have a plan in place to deal with them.
It's tempting to imagine that the capacity planning math for PDC looks like this:
# available seats = # rooms per timeslot * average seats per room
ideal # attendees = # available seats * .x
where x is some number to account for the fact that not every session will be equally popular, and thus you assume a percentage of open seats will be unused in every timeslot.
If it were this easy to model capacity at the event, then we could do some more simple math:
max # 2005 attendees = # 2003 attendees * percentage 2003 overflow
For example, if data showed that the average session in 2003 was y% overcrowded, just reduce the max number of attendees in 2005 by y% vs. 2003, and voila, problem solved.
And if it were this easy, then Kevin’s right, there would be no excuse for surprises.
The data for 2003, however, suggest that this approach is over simplifying things. I’m convinced that our problem wasn't a systemic capacity shortage (too many people, too few seats) – our problem was mismatched room scheduling. Here are a few interesting data points I've found:
- If you look at the top 20 most crowded sessions, you find that 12 of them were scheduled for rooms that held less than 260 people.
- The most overcrowded session was booked into a room for 254 people. During that same timeslot, for comparison, there was one session in a room of 258 that was 30% empty, and another session in a room of 1200 that was 60% empty.
- The single timeslot with the most overcrowding occurred had an average session attendance of little more than half of total capacity
- The single timeslot with the least overcrowding occurred had an average session attendance of more than two thirds of total capacity. The timeslot with more people had less overcrowding.
I’ve been thinking about this data for a few weeks, now, (I posted some thoughts last week, in fact) and the only real conclusion I can draw is that it's just hard to predict what sessions will be most popular. We could have cut our attendance in half in 2003, but we probably still would have had overcrowding. It doesn’t make a difference whether 800 people or 400 people are interested in a session, if the room only holds 250.
We're working on ways to improve our capacity planning for 2005. Some of things I’ve been pondering (with the help of the rest of the planning team):
- Don’t schedule any rooms that hold less than 300 people. It’s too easy to get surprised and have 350 folks show up. Instead, keep those smaller rooms for repeat sessions, or even video overflow if needed.
- Do a better job predicting attendee plans. Help attendees plan their schedule more accurately by suggesting a few different core curriculum – "if you're developing web commerce sites, we suggest sessions x, y and z, if you’re developing Win32 client apps, we suggest a, b and c". The more accurate the session preference survey is, the better we can anticipate capacity and schedule accordingly.
- Be ready to be wrong. If a session is overcrowded, be ready to immediately set up an overflow room or schedule a repeat.
- As Scoble hinted, consider allowing fewer attendees
So, Kevin, we hear ya, and we're working on it. We're not going to be perfect in 2005, but when we're off base, we'll be quick to make a correction.
Comments
- Anonymous
March 24, 2005
Just put the sessions up on a web site a month before PDC and get people to rank each timeslot. The put the most popular in the big rooms.
If people don't bother going online to vote then it becomes their problem when a session is full - Anonymous
March 24, 2005
Andy, we do this each PDC, we call it the session preference survey. And it turns out to be a poor predicter for a couple reasons:
1) not many attendees actually do it. If I could find some way to give priority seating to those who pre-scheduled the session, I would, but we haven't got the infrastructure for that.
2) attendees are doing it blind -- imagine trying pick the right sessions for yourself in 2003 when no information about Avalon, Indigo or WinFS had been made public yet
3) speakers sometimes suggest follow-up sessions on the fly. If you're in a 1200 person room, and a question comes up where the speaker says "oh yeah, we'll cover that in session CLI350", but CLI350 is in a 250 person room, you have a problem - Anonymous
March 24, 2005
Of course, I understand that conference scheduling is hard and imperfect. My comment was meant to be simplistic and provocative. I'm glad that it got the reaction that it has. My point was that instead of trying to maximize the attendance Microsoft should try to maximize the enjoyment of the attendees and that can partially be solved by having fewer attendees than in 2003. It is nice too see that Microsoft understands the problem and is trying to solve it. I look forward to attending the PDC this year. - Anonymous
March 24, 2005
Wouldn't it make sense to require people to atleast make a partial itinerary, and include the itinerary file as an attatchment through the ticket purchase form? This way each ticket purchased would have a corresponding itinerary off all the sessions that the purchaser intends to go to.
Or perhaps require people to register an itinerary of sessions they plan on attending. Doing so will generate them a unique ID number which they must provide when purchasing a ticket, and you will be able to know how many people want to attend the sessions.
If something is voluntary, but could potentially help make the PDC experience that much better by reducing overcrowded sessions, then make it mandatory. In this case... a list of sessions people intend to sit in on. - Anonymous
March 24, 2005
Maybe I'm a contrarian. I attended PDC 2003 (my first PDC!), and I had a grand time. I attended the overcrowded sessions, but by careful planning, I always had a seat.
I would be opposed to having to schedule ahead of time what sessions you attend. I've been to conferences before where you are bound to a particular 'track' and can't visit other sessions. It's really not as fun, and not as dynamic of environment. - Anonymous
March 24, 2005
And speaking of overcrowding, is ther any chance that network coverage will be better? The WLAN was so oversubscribed last time that you were fortunate if you could get on. And the LAN would run out of DHCP addresses, so you had to continually ipconfig /renew, hoping that you'd get lucky. - Anonymous
March 25, 2005
I'm guessing that there is a video feed available for any overcrowded session. Can y'all just allocate some "flex rooms" with projection systems that are used for the overflow? If a popular room fills up, post a sign outside directing folks to the right flex room.
Even better, provide a remote control in each flex room. Let the people there determine who gets the remote and what session they watch. And you can televise the fights to another room where you charge for admission... :-) - Anonymous
March 25, 2005
If you attended PDC 2003, you probably recall that there were some big issues with overcrowded sessions. Jeremy Mazner, content owner for PDC 2005, has written about some of these challenges... - Anonymous
March 28, 2005
Keizer's blog - Anonymous
March 29, 2005
My hypothesis is that you can gain some insight into expected room attendance by doing a little correlation/regression analysis of PDC 2003 based on variables like these: speaker(s), track (architecture, client, mobile, etc), session title keywords (app security, transaction support, xml), time of session, attendance at previous time slot, neighboring session. Then you could map these to this year's conference.
But, to be honest, I frequently just followed the crowds in 2003, trying to attend the popular sessions. - Anonymous
April 04, 2005
The comment has been removed - Anonymous
July 13, 2005
PDC 2003 has screens outside the overcrowded rooms, but only 20-30 people could actually see anything, so Walter Lounsbery's idea is pretty good.
You could also stream the overcrowded sessions live. If the wireless network works this time, people could simply sit anywhere while watching a streaming video of the session they're interested in.