Published using Google Docs
EAGxCambridge 2023 Retrospective
Updated automatically every 5 minutes

EAGxCambridge 2023 Retrospective


Written by the core organising team for EAGxCambridge, this retrospective evaluates the conference, gives feedback for the CEA events team, and makes recommendations for future organisers. It’s also a general update for the community about the event, and an exercise in transparency. We welcome your feedback and comments about how we can improve EA conferences in the future.

You can watch 19 of the talks on CEA’s YouTube channel, here.

Attendees’ photos are here, and professional photos are in a subfolder. We thank Ryan Fallon for the professional photos.





Basic stats


Why Cambridge?

Focussing on the UK and Ireland

Core team

Compressed timelines



Some admissions statistics



Main venue: Guildhall first floor

Secondary venue: ground floor

Tertiary venue: Lola Lo


Coordinating with venue staff on the day

Overall view



Attendee experience

More snacks

Better acoustics

Faster wifi

Food was “incredible” / “amazing” / “extremely good” / “really excellent”

Attendee Slack


Attendee favourites

“Were any sessions you attended particularly good? Which ones?"


Satellite events



Comms strategy

Comms tactics

Closer to the conference itself

Feedback survey results

Net Promoter Score


Resulting actions



We think EAGxCambridge went well.

The main metric CEA uses to evaluate their events is ‘number of connections’. We estimate around 4200 new connections[1] resulted, at an all-told[2] cost of around £53 per connection (=$67 at time of writing), which is a better cost-per-connection than many previous conferences. The low cost-per-connection is partly driven by the fact that the event was on the large side compared to the historical average (enabling economies of scale to kick in) and encompassed 3 days; it was also kept low by limiting travel grants.

Of these 4200 new connections, around 1700 were potentially ‘impactful’[3] as rated by attendees. (Pinch of salt: as a rule, people don’t know how impactful any given connection is.)

The likelihood-to-recommend scores were on a par with other EA conferences, which are usually very highly rated. (The average answer was 8.7 on a 1-to-10 scale.)

Besides making connections, we also wanted to encourage and inspire attendees to take action.

82% of survey respondents said they planned to take at least one of a list of actions (e.g. ‘change degree’) as a result of the conference, including 14.5% resolving to found an EA organisation and 30% resolving to work full-time for such an organisation or in a primary cause area. After applying a pinch of salt, those numbers suggest the conference inspired people to take significant action. We heard of several anecdotal cases where the conference triggered people to apply for particular jobs or funding, or resulted in internships or research collaborations.

A pie chart showing the percent answers to the feedback survey question "Which actions do you plan to take as a result of this event?". The figures are available in a table in the  in the ‘Feedback survey results > Resulting actions’ section.

Results from the feedback survey. Note that the percentages are calculated as a fraction of the actions reported, not as a proportion of people, since it was a multiple-choice question.

We're very thankful to everyone who made this happen: volunteers, attendees, session hosts, and many others.

Basic stats

We estimate that 497 people attended EAGxCambridge on March 17th-19th 2023.

There were 543 applicants and 417 approved applications. We directly invited around 100 others to host sessions or represent key organisations.

There were about 70 session hosts, hosting 25 talks, 20 meetups, 13 workshops, and 10 ‘office hours’.


These were our objectives, in order:

1: Connect the EA UK community.

2: Welcome and integrate less well-connected members of the community. Reduce the social distance within the UK EA community.

3: Inspire people to take action based on high-quality reasoning.

Why Cambridge?

Until 2023, the UK had played host to seven EA conferences: five in London, and two in Oxford. There had never been an EA conference in Cambridge, despite the city’s status as EA UK’s third hub, and being a short train journey from London.

The original instigators of the conference had hoped it would help establish Cambridge as a bigger and better location for EA in the UK, building links between it and the rest of the national community. In the event, there was a mixed picture on this objective (partly because the core organising team pivoted to targeting the UK in general). Most organisations we platformed were based either in London or Oxford. We were able to platform only two major Cambridge-based EA organisations: the Cambridge Alternative Proteins Society, and 5 academics at the Centre for the Study of Existential Risk (of whom 1 pulled out due to illness). A greater number of local organisations were not platformed, but did send representatives, in some cases because we or they thought it would be more appropriate to do so, and in others because we weren’t able to invite them in time. (Namely: BlueDot Impact, Cambridge AI Safety Hub, Cambridge Existential Risks Initiative, Existential Risks Alliance, and the Leverhulme Centre for the Future of Intelligence / CFI).

Focussing on the UK and Ireland

We decided to limit admissions (in almost all cases) to those based in the UK or Ireland, or planning to move there within the next year. The main reasons behind this were:

Some downsides of this strategy were:

An unforced error was that we should have emphasised the UK/Ireland focus in our communications and marketing more, and from the very start, to save the time of applicants whose location made them ineligible.

Should future EAGxs in EA hub countries such as the UK limit attendance in this way? We endorse having done so here, but think if our venue had a larger capacity, it would have made sense to open the gates to international applicants. Adding more international attendees mainly shouldn’t prevent the (supposedly) more valuable local connections. It would be possible to implement a more fuzzy rule where location was just taken into account to some degree, though this would make it more complicated to communicate our criteria to attendees.

Core team

The original instigators of the conference were Harriet Patterson and Ray Amjad, who set things in motion by booking the main venue and by recruiting the core team. The core team originally consisted of David Mears (team lead), Seli Ballo, Rebecca Eddington, and Juan Rocamonde; we later brought on Jordan Pieters and Amy Shaw, bringing the core team up to about 5 FTE by the week of the conference. We also benefited hugely from support by volunteers, particularly volunteer team leads (more on which later).

We worked remotely from the UK, South Africa, the US, and the Netherlands, with the exception of some site visits and food tastings (attended by 2 or 3), and 5 of us worked in person together in Cambridge in the final week. Remote working worked fine, but we found that it was hugely easier to collaborate in person, especially as we didn’t know each other beforehand. For those in Europe, it would have been worth the expense to meet each other in person earlier on, to bond the team; and, given that we didn’t, it probably would have been worth us staying in Cambridge for two weeks in advance of the conference rather than just one.

It was particularly valuable having someone with experience of organising a previous EA conference with us full-time in the last few weeks (Amy Shaw from EAGxRotterdam). Not only were we able to pre-empt more problems as a result, but also it allowed us to save time by copying several things (e.g. document templates, or contracting the same person to make the venue map).

We split up responsibilities as much as possible, and recommend doing so: delegating entire swathes of work (such as ‘comms’ or ‘volunteer management’) works well, as then the head of each department knows all the relevant context and autonomy to make decisions. It would have been good to have enough team capacity that the team lead was not doing as much ‘direct’ work, as this robs people of autonomy and relevant context, and frees them up to do more zoomed-out strategizing.

Compressed timelines

Throughout, the core team suffered from a lack of capacity, partly as a result of only being recruited in time to start work in earnest at the start of January, less than 3 months before the conference; and partly by having other commitments. (For future planning: we estimate the total time worked was something like 16–1700 hours, about 40% of these attributable to the team lead.)

As a result, two of us, including the team lead, temporarily burned out due to high workload and stress. Our desire to see a functional conference meant, for at least two of the team, working through most weekends for three months. David reports an unpleasant experience, working on the conference for an average of 70 tracked hours a week for the last four weeks. We think we would have made better decisions, and had a better time, if we were not working so hard.

Some other costs of compressed timelines included:

If we had been less lucky, and hadn’t been able to hire two more people to our core team, we might have cancelled or scaled back the conference, which would have been a big waste.

We’d advise future organisers to at least hire further in advance, and possibly ask for more hours per person (most of the team had other major commitments during January and February).

The team also felt we would have benefited from more communication between our team and CEA. Mainly, David took on the role of go-between for our team and Ollie on CEA’s Events Team. We’d encourage future organisers, and CEA staff, to err more on the side of reaching out proactively rather than strictly going through a single designated contact person.

Our productivity was somewhat limited by the use of a brand new platform to manage applications (in Salesforce) which was not editable by us. We couldn’t do our own retrieval of data from this, but had to go through CEA for all basic questions like ‘what are the attendees dietary requirements’ or ‘how many people said they were into animal welfare’ or ‘how many applicants are students’, delaying many of our decisions. It seems better to have open access to data, otherwise there’s less point in asking the questions in the application form.


The estimated overall spend comes to around £223,500, including the reimbursement of the core organisers, who were contracted to CEA. We provide an itemised breakdown below in the spirit of transparency.

Catering (meals)

£ 46,020.00

Core organiser payment (estimated)

£ 44,000.00

Audiovisual tech equipment and labour

£ 32,086.85

Primary venue hire (1st floor Guildhall)

£ 29,767.56

Travel and accommodation reimbursements

£ 28,665.00

Furniture hire & set-up

£ 9,093.70

Lola Lo venue hire

£ 7,625.00


£ 5,781.00

Ground floor venue hire

£ 5,222.40


£ 3,494.82

Signage printing

£ 2,220.48

Organisers accommodation (1 week in Cambridge)

£ 2,083.86

After party

£ 2,100.00


£ 1,500.00


£ 1,322.40


£ 1,300.00

Orlene's Kitchen (café)

£ 850.00

Speakers reception dinner

£ 717.00

Wired internet for venue (unused - booked for ill speaker to present remotely, who then pulled out due to illness)

£ 660.00

Art for walls

£ 582.68

Pub social the day before conference

£ 500.00

Staff travel (estimated)

£ 500.00

Probably other things that didn't make it onto the spendtracker amid the chaos of the weekend / final week (number plucked out of air)

£ 500.00

Graphic design

£ 479.32

Use of Sidney Street office as co-working space during conference

£ 400.00

Laptop hire

£ 264.00

Volunteer gifts

£ 177.45


£ 106.14

We would have been able to save a whole lot of money and time if we had been able to be told what inventory CEA has in Trajan House. Unfortunately, key people at CEA were on holiday and we couldn’t otherwise access the list. We suggest CEA should ensure that the list is kept up to date after each conference, so that purchases aren’t duplicated.

For information, these are some things we might have spent money on if we had more budget or made more savings:


As previously mentioned, the core team was recruited less than three months before the conference. This limited the amount of time available to work on the admissions strategy and process applications. Launching admissions was further delayed by CEA switching their admissions software from Zoho to Salesforce. Admissions were therefore only open for just over a month.

When writing the admissions strategy, a lot of thought was put into the demographics of the conference and setting targets around these. We decided to limit attendees to those who were based in the UK/Ireland or that had definite plans to move there in the next year. This decision is discussed in more detail in the Strategy section of this document. As the UK is a key EA hub, we decided to concentrate on integrating people in the community with the potential to become highly-engaged in EA (‘proto-HEAs’), as opposed to prioritising welcoming new people to the community. Our target was for 60% of attendees to be proto-HEAs and 40% to be more experienced. We had strategies for reaching underrepresented groups in the community, namely; people of colour; women & NB people; people whose closest city was non-Loxbridge; and people with lower socioeconomic status. We added a question to the application form to track socioeconomic diversity. In practice, demographics weren’t very well tracked or tailored during the admissions process due to a number of factors:

Just after launching admissions we had a calibration session, to make sure we were aligned in our decision-making based on the admissions strategy. We created a Slack channel to discuss applications we were uncertain about within the core team. If, after this we were still uncertain, we’d use ‘Second Opinion’ and have Ollie make the final decision. In the first couple of weeks, while applications were lower and a few people in the team had capacity to assess applications, this worked well. Ollie gave feedback that we’d been using the ‘Second Opinion’ option less than other conferences and in the final week, we used this a lot more, mainly due to only having the admissions lead processing applications as the rest of the team had other priorities.

We expected a big spike in applications as the deadline approached, and this certainly happened. Since the number of people coming affects lots of decisions – from catering, to number of direct invites, to the need for spill-over venues – and a significant proportion of all applications are submitted on the final day, it’s good to have the application deadline well in advance if timelines allow. Definitely make sure your team has capacity to process a lot of applications in the final days. We didn’t have enough time to implement early-bird discounts, but strategies like this, that encourage early applications, are worth using.

We recommend having a hard application deadline, and not making exceptions (since this would lead to an unfair two-track application process). We stuck to our initial deadline but discussed extending it, as we were surprised by the low number of applications around 2 weeks prior to the conference. After a surge of last minute applications we decided to keep the initial deadline. If we’d extended the deadline, some of the key outstanding tasks would have had to be carried out to a lesser standard, particularly the volunteer management as the admissions lead was also the volunteer lead. We would recommend closing admissions at least two weeks before, to concentrate on other priorities pre-conference. Having a hard application deadline also meant that there were no decisions to be made about late applications. We had a template to send to those who asked to apply late – there were lots!

Some admissions statistics

25% of attendees first got involved with EA in 2022.

Based on approved applications: a clear majority of attendees came from one of the three EA UK hubs (London, Cambridge, and Oxford), with Durham, Bristol and Edinburgh trailing behind.

We didn’t receive enough data on ethnicity at the point of application (more than 90% didn’t provide ethnicity information), but see Feedback survey results for the more substantial post-conference data. In terms of gender, among approved applications, 19.9% said they were Male, 14.3% Female, and 1.2% Nonbinary, Agender or Genderqueer. (Scaling that up would produce 40%, 56%, and 3%, showing a skew towards Male in the feedback survey relative to the approved applications.)


By ‘stewardship’ we mean something like ‘trying to do better than serendipity in helping people meet the people they want to meet’.

Feedback from past EAGx conferences (mainly drawn from retrospectives) and from CEA in the EAGx Handbook was that it’s difficult to beat the ‘market’ of the conference, and the stewardship programmes often didn’t facilitate more meaningful connections than would otherwise occur. We decided early on that we would therefore create spaces for people to organically connect rather than trying to match people for 1:1s or similar. We also knew that we probably wouldn’t have the capacity within the team to do a good job of matching people as it’s a time intensive task.

Two ideas which we decided on were having ‘cause area guides’ and creating spaces for peer-led discussions. We initially tagged people in the applications if we thought they’d be a good mentor (see above on why this tagging wasn’t very effective) but, in the end, reaching out to these people and creating a stewardship programme was deprioritised. We found a volunteer willing to help organise the stewardship programme around 3 weeks before the conference, but because of team capacity we ended up dropping most of our stewardship plans, which meant that the time we put into planning them in depth was wasted.

We did reach out to university group organisers and ask them to encourage promising students or students from underrepresented groups to apply, and also to provide a list of students they thought might get a lot of value from the conference. Ollie also stepped in the week of the conference to carry out the MVP stewardship tasks, as listed in the EAGx Handbook.

While the time we had to implement a stewardship programme was definitely limited, we would also recommend deciding on a stewardship plan early on and dedicating some time to work on it each week.


The main determining success factor for any conference is the venue. Luck was on our side here in various ways, which we’ll give detail on below.

Main venue: Guildhall first floor

Our main venue, who pencilled us in around 5 months before the conference, was the first floor (in American English: ‘second floor’) of the Guildhall, which is an old-fashioned council building on the marketplace in central Cambridge, opened in 1939. This was a relatively good deal, because it was managed by the council rather than by a ruthless for-profit business, and because we were able to stick to the 2022 quote and avoid a very large 2023 price hike. We paid for this public-sector discount in other ways, such as in dealing with a more sclerotic bureaucracy (painfully slow communication).

Communication with Guildhall’s venue manager was challenging, since they had a lot of people handling different things. We did not get an email reply within 1 month (from December to January) because she was on a long leave, only a few months before the mid-March conference, with our contract still unsigned. We became fairly desperate, and a number listed on the council website that we left some messages on was even silently taken down. For future organisers, it might be a good idea to ask in advance for a second and third person to reach out to, especially for production purposes: venue, catering, AV.

We were very lucky that an event scheduled at our venue on the previous day was cancelled due to a strike, which allowed us to book in enough time to set up in advance of the conference. Without this, the chances we would have been able to set up in time would have been slim, and stress-levels would have gone through the roof. We would have had to start the conference on Saturday in order to eliminate the risk of not being ready in time – partly because it was difficult to know how long to allow for the set-up when accounting for unknown unknowns (and indeed there were big, surprising hitches on the day). We learnt about this previous-day event only when we asked to book it in, after our contact person at the venue returned to work. At that point, we’d already advertised some conference dates. If the event on the previous day had gone ahead, the correct, though somewhat embarrassing, decision would have been to change the publicised dates and have a shorter conference.

Choosing a spacious and easy to navigate venue should be a top priority, and it would be nice-to-have for the location to be near the city centre. The Guildhall was located very centrally in Cambridge and was easily accessible from train and bus stations. However, the venue had limited 1-1 spaces and the floor layout was very confusing, which was only partially mitigated by the addition of plentiful signage.

Secondary venue: ground floor

We added two more venues to our arsenal as time progressed and the need for them became clear. To say we were lucky to find these would be an understatement.

Most importantly, we added the ground floor (in American English: first floor) of the Guildhall (managed by another company, Allia). We happened to spot a poster after the site visit to the primary venue in January, advertising that the ground floor could be booked from this other company. Booking this extra floor certainly added complexity, since the two venue providers had barely even met each other, despite inhabiting the ground and first floors of the same building, so it was incumbent on us to facilitate them in talking to each other. Despite this added complexity, it was very clearly worth it to have access to the ground floor rooms. Without these extra rooms, we would have had to pare back on many basic functions of the conference. To demonstrate, we used these extra rooms for:

As a result, annexing the ground floor gave us the freedom to designate some first-floor rooms to non-central functions, namely a ‘living room’ chill-out space, a room for storing furniture that we cleared out of the living room, a nap room, and extra space for the caterers to plate up meals. The counterfactual if we had not booked the ground floor would also have impacted the main content schedule: we would have not had the space to accommodate nearly as many sessions.

For this reason, it’s very important to start allocating your room uses, and nailing down the session schedule, early enough that you can still take action if you realise that you don’t have enough space.

Tertiary venue: Lola Lo

Photos taken during the conference.

Taken after day 1:

Around three weeks before the conference, we became increasingly worried about the lack of space for 1-1 conversations, coupled with the acoustics issue mentioned below which reduced effective capacity by a hard-to-predict amount. The team lead spent a significant amount of time calling local restaurants and pubs to find more space, in order to prevent the lowish-chance, high-impact possibility of the space getting maxed-out for 1-1s. In the end, we went with the nightclub Lola Lo, which had three floors. As this was the only option we found which was close enough to be usable (it was directly opposite the back of the main venue) and which had availability, this was another instance when luck was on our side.

Lola Lo was beautifully decorated, while not as professional an environment as would be expected at a conference, and good value for money, theoretically; but we struggled to advertise its existence, meaning that, in the event, it didn’t contribute that much value.

Taking all the venues together, we could have fit more people into the conference, perhaps 20% more if attendees made full use of all of our venues, which would have let us accept more attendees and create more connections without significant extra costs. But we were uncertain about this, and didn’t want to end up with a disastrously crammed venue (given we didn’t want to rely on good weather for sending people outdoors), so we played it somewhat safe. This is one reason in favour of conferences using the same venue repeatedly: to reduce uncertainty around the capacity.


This is a very important thing to consider in venue choice, and was the joint-most commented upon thing in free-form fields in the feedback survey, with 6 mentions.

Due to the boomy acoustics in the largest rooms, it was hard to hear others in discussions, the noise was overwhelming to some (adding unnecessary mental load to a potentially tiring weekend), and one person said they were losing their voice from trying to be heard.

Although the boomy acoustics were the most significant issue discovered during our January site visit, there was little we could do about it at that point, since the venue had been locked in around 5 months before the conference (we couldn’t have found any other affordable venue options at such short notice). We tried to mitigate the issue by

  1. Booking some carpets and sound absorption panels / dividers, but we didn’t have the funds to cover the whole space, so this had a negligible impact. In future, this should be considered for small rooms with hard surfaces, eg this room divider.
  2. Booking a spill-over venue for 1-1s (Lola Lo’s), with drier acoustics. Not enough people were aware of the existence of this spill-over venue, despite Swapcard notifications and a shout-out in the opening talk, and this will have been hampered by the fact it was not in the same building.

Coordinating with venue staff on the day

We had some difficulties with communication between us and the venue staff. Although it’s good to have the contact number for various venue staff available, do not encourage its use by too wide a group of people. One of the duty managers (we had two of these, who alternated shifts) was unable to cope with the influx of messages from us to them, some of which were apparently contradictory due to imperfect communication on the core team and volunteer teams. So we needed to limit the info flow to a single individual / pair of individuals, and through a single channel (i.e. only whatsapp, not texts & whatsapp & email).

Unfortunately, we often had to remind the venue to do things like disposing of waste, and repeatedly correcting their mistakes with putting the security staff on the wrong doors. We negotiated during the conference for them to hire more cleaning staff, but some of the toilets’ level of cleanliness was unsalvageable (permanent smell).

A hitch that could have become problematic if it hadn’t been solved in time was that the venue ran out of its water supply, and said that it had no time to find a supplier during the weekend. Luckily, we managed to find a supplier and fix the issue in time such that attendees never had to find out how close they were to dehydration.

Overall view

My personal overall view (David) is that the issues with the venue (acoustics, poor communication, poor on-the-day attention-to-detail, difficulty in navigating) were significant enough that any future proposed EA conference in Cambridge should put significant effort into finding an alternative venue.

We needed to book multiple venues to accommodate for the missing features of the original main venue, and this multiplied the existing difficulties with navigation.


Volunteers were the backbone of the conference. We asked a lot from our volunteers and they came through, making the lives of organisers and attendees easier by being proactive and friendly. The volunteer team leads deserve a special vote of thanks: Stefan Heimersheim and Tom Gardiner as logistics team leads, Chay Morris as speaker liaison team lead, Annabella Wheatley and Richard Annilo as registration team leads, and Tymoteusz Syrytczyk as room management team lead. Some scattered recommendations:


Attendee experience

The common themes in the feedback survey (n=135) were:

More snacks

Mentioned 6 times. Our mistake regarding snacks was (probably) not in ordering too few, but in failing to ensure that we had a process for getting them distributed and topped up. This was dropped amidst everything else going on. As far as I (David) recall, we forgot to designate a volunteer team to put out the snacks, until we got feedback from attendees that they needed snacks. I also seem to recall that we had some mutual misunderstanding within the team about whether the snacks we were ordering were for volunteers or for attendees, which might have contributed to this being omitted.

Better acoustics

Mentioned 6 times, a major issue. See discussion on this in the ‘venues > acoustics’ section.

Faster wifi

A common complaint, with 4 mentions. We had tried to mitigate this by (a) asking people to use their own data plan where possible in the attendee guide (though it seems this message didn’t reach most people, as most people will not read the guide in its entirety) and (b) designating the Sidney Street Office as a workspace (but this was under-utilised since it entailed a 3-minute walk from the venue and again we struggled to advertise its existence).

Food was “incredible” / “amazing” / “extremely good” / “really excellent”

There were 4 comments in this category of superlative compliments, 2 of which gave a shout-out to breakfast in particular. Remember that this was in answer to a free-form question – we didn’t prompt people to comment on the food here! One of these went on to say: “I really do want to highlight how much I enjoyed the food - I will be thinking of it for days”. Our caterer was Feed The Village, a vegan-specialist caterer based in London. We had them provide light bites and dinner on Friday, three meals on Saturday, and breakfast and lunch on Sunday.

Note that we also had critical comments about the food quality, as is to be expected; and (along a separate axis) one or two complaints about portion size. In particular, one meal option ran out on one of the days (the burritos), which is a risk entailed by providing options.

Overall, our attendees were satisfied[4] with the quality of the food, especially on taste and quality. We were lucky that we found Feed The Village within our three arranged food tastings, since they were so preferable to the other two leading contenders and we did not have enough time to continue scouting.

Our tips:

Attendee Slack

We used Slack as the group chat for attendees, to enable coordination in advance of the conference (on travel / accommodation) and during the conference (for socials and group meetups). The motivation was that it’s hard to have multi-person conversations in Swapcard (it’s possible, but users tend not to know how to do it).


You can watch 19 of the talks on CEA’s YouTube channel, here.

We took care to choose which topics to cover, and in which order to prioritise them, as we knew we wouldn’t be able to cover everything. Our prioritisation took 80,000 Hours’ top 8 as the starting point, then added in animals, global health and global poverty as these are favoured by rank-and-file EAs. Animals received a further boost partly because there is an active alternative proteins scene in Cambridge. We were also informed by the EA Survey results, scoped to the UK (we had to request this data specifically).

If we had had many months to organise the conference, and if the application form included a question about topics of interest, we could have been more responsive to the topic interests of the attendees themselves. In lieu of this, we received many good content suggestions that ended up being included, by creating a form for people to suggest content, which we linked from the ‘save the date’ post and the website.

Although we wanted to aim high in terms of cause diversity, we didn’t manage to cover all the topics that we aspired to include. The topics we hoped to cover, but didn’t, were: nuclear weapons, forecasting, improving institutional decision-making, climate change, and great power conflicts.

In the event, we had a preponderance of AI and meta-EA-related content, with 9 and 15 talks/workshops respectively; we had 4 talks or workshops on each of animals, global health & development, and biosecurity; and 6 on existential risks besides those focused on biosecurity and AI. (These numbers exclude meetups.) This was more lopsided than we had aimed for.

In the end, there are limits to what you can do to control the balance of the programme, as it depends on who responds. The most important tips are to start early and to keep actively tracking the balance. People within the EA movement and people who work on movement-building are more likely to respond.

On the demographic diversity side, quantitatively, we invited 54 male speakers, of whom 28 were interested, and 46 female speakers, of whom 15 were interested (non-binary speakers excluded from those summary statistics due to small sample size); of 16 invited speakers of colour, 7 were interested. If you aim for equal representation of genders among speakers, you will need to invite more female speakers than male, because of the response rate.

It might have been a good idea to send out speculative invitations to speakers, asking them if they would potentially be interested (which we generally did not do until later in the process), because keeping that flexibility until you know more about your options, audience, and existing balance of content allows you to make more informed decisions. This method does run the major risk of disappointing people who you later decide not to invite, if they feel they have been disinvited, so always communicate clearly.

In order to meet our objectives of facilitating useful connections and of inspiring people to take actions, we aimed to have many interactive sessions, and plenty of topic-based meetups. In conflict with those objectives, note that attendees historically tend to prefer networking over talks, and talks over workshops. This may just reflect the quality or effort put into each of these; our attendees were relatively keen on our workshops over the talks.

It was a pity that we had to abandon our hopes to platform domain experts and practitioners from outside EA, and only had one talk that could fairly be categorised as ‘criticism of EA’, due to team capacity (it takes much longer to discover names outside one’s network than inside it, and they are less likely to be excited to contribute). I (David) had hoped the conference could help engage EA with more outside perspectives.

One thing we should have done differently: start making a skeletal content schedule earlier in the process so you can be confident that you have enough space/time for all the sessions and have capacity for all the interdependencies. For example, different formats require different set-ups, and different room sizes. We were lucky to find a secondary venue, giving us one more room to accommodate our content!

Attendee favourites

“Were any sessions you attended particularly good? Which ones?"

Below are the tallies of shout-outs (for the top sessions), in answer to the above question in the feedback form. (Half points where it’s not certain which session was being referred to.)



Attendees (including repeat sessions)

‘Scout Mindset - Developing A Map Of The World Instead Of Fighting For Our Beliefs’ workshop by Friederike Grosse-Holz



‘Thinking Outside The EA Cause Area Box’ session



‘Building a Theory of Change for your Research’ workshop by Michael Aird



‘The What, Why, and How of Applying for Funding’ workshop by Michael Aird



‘Upside Bargains: Why And How To Get Way More Value In Your Life By Thinking In Bets’ by Peter Wallich



‘The Estimation Game’ by Adam Binks



Anders Sandberg (talk or interview)



‘Competition Policy And AI Governance’ by Haydn Belfield and Shin-Shin Hua



‘Asking 'Daft' Questions With Confidence’ by Yulia Ponomarenko



We also have data from Swapcard, which allows participants to rate sessions out of 5, but due to (we infer) small sample sizes and a small range of ratings given, it seems less useful than the answers to the feedback survey above (very noisy).


Merch is worth getting out of the way early, because you’ll inevitably need to make various changes to the design proofs, and allow for delivery of samples; and you don’t want to spend much time on it close to the conference when there are more important things to think about.

We wanted to make a break from the traditional EA habit of using various shades of blue, which is a little tiresome and also masculine-coded, and used a vibrant pink for volunteer t-shirts.

We only provided t-shirts, rather than also hoodies/sweatshirts, to save money.

Although we followed the guidance on proportions of different clothing sizes given in the EAGx handbook, we ended up with a lot of smalls left over, and did not have enough larges. Previous conferences tended to have the opposite problem.

With items besides clothes (e.g. notebooks), don't order as many. Expect that around two-thirds of people will take extra items on top of clothes.

Satellite events

Causing events to be held in the days before and after the conference was a key part of our strategy to maximise connections and welcomingness. The theory was that, by advertising an enticing variety of events in the run-up and aftermath of EAGxCambridge, we would encourage EAs from around the UK to spend more time in Cambridge, fostering more connections between EA in Cambridge and EA in the UK. If people know that they will be able to meet EAs outside the conference, it becomes more worth their while to stay for a longer time. Most communications mentioned satellite events for this reason. See this forum post for more reasons to focus on satellite events, and suggestions.

We ended up having around 13 satellite events organised by community members, including karaokes, picnics, and topic-specific socials. These organisers deserve praise and thanks for the added value they brought. I will particularly single out Stefan Heimersheim for running two socials on top of his already extensive commitment to the volunteer logistics team, and Jade Amos for masterminding the afterparty.

Despite this panoply, I (David) expect we didn’t manage to have the details of these locked in early enough for us to see a real effect on travel plans – most people would have already made travel plans and booked their accommodation before they heard about most of these events. We published the satellite events calendar only on March 8th, 9 days before the conference. (The feedback form did not track connections made at satellite events separately, so there isn’t any quantitative data on this.)

In practice, community members often step up to the plate to organise satellite events, but they don’t naturally do this far enough in advance for it to have an effect on attendees’ travel plans unless a devoted organiser makes a concerted effort to prompt them.

As such, it would be a good idea to have a specific person (perhaps outside the core team) to manage the satellite events programme. Unfortunately, since we didn’t find someone willing to take this on, it was relegated to a lower tier of priority and came under the team lead’s remit. The calendar came together in as last-minute a fashion as other aspects of the conference, meaning that they probably were less well-attended than they could have been.


The person working on comms should be a well-integrated member of the team so that they have the full context of what information is the most important. Needless to say, it’s ideal to have this person be near native-level in the target language, so that their writing is idiomatic for that language and requires less proofreading.


A logo is among the first things that should be prepared, because the logo design will be used in posters, merchandise and social-media posts. It took us roughly 2 weeks to get to a finished logo with a contracted designer due to the back-and-forth revision process. This doesn’t account for the time spent searching for a designer in the first place: we first rejected using high-quality professional contractors because we discovered they would charge several thousand pounds for a logo, and then we lost a few weeks waiting to hear back from some pro-am designers before we found Zian Bonoan. We advise you to ask for help or designer recommendation from previous conferences in the EAGx Organizers Slack workspace.

Recruiting at least one core team member who has basic design skill with photoshop or canva also helps to increase efficiency, especially given the time constraints in preparing for the conference.

As of 2023, AI image generation is not a good replacement for a logo designer except in terms of inspiration, since it can’t achieve perfect block colours (which are necessary for affordable printing).

Comms strategy

Co-creating a comms strategy document was helpful to guide our focus and follow the objectives of the conference. Every conference will have a different concept; it is essential to align the communications strategy (who should be reached out to, who do we prioritise, what does the timeline looks like) with the conference’s main purposes, for example if you want to target a particular under-represented group.

The under-represented groups we particularly wanted to reach came under:

For all except regional diversity, the best way to make headway on these seems to be asking for group organisers to pay special attention to it. We wrote to them:

> Additionally, we would also like to request that you take a moment to consider if there are any people in your network from groups under-represented in the EA UK community (women, people of colour, low socioeconomic background) who would benefit from more connections in the community. We think it would be really valuable if you could reach out individually to suggest they submit an application, as such groups can need more encouragement to apply.

Because we didn’t want the conference to be too student-heavy, this was reflected in the comms strategy. We sought out profession-based groups such as those listed by High Impact Professionals here so that mid-career professionals would be able to hear about the conference.

Split your comms timeline into several stages: the new information/headline gives people a reason to keep reading. E.g. “save the date”, “applications open”, “main speakers announced”, “early bird tickets deadline”, “application deadline”.

Comms tactics

Announce your conference in relevant places as soon as you know the dates. Even if you don’t have much to say about it at that point besides the dates and location, it’s very helpful for people to be able to pencil it into their diaries as soon as possible. This was especially important for us with our short timelines. Our initial EA Forum announcement post was very powerful for our comms, because it included a form where people could sign up to be notified when applications opened. Around 300 people signed up to this, and we had a good click-through rate (around 20%) from emailing these people. (Do allow people to unsubscribe.)

The most helpful comms channel is to ask for help from the organisers of local and other EA groups: they have much better access to their members than you do, and provide a lot of help by reaching out to individuals to encourage them to apply. Send them templates of copy that they can share in order to save them time.

Besides these, we also sought to be mentioned in major newsletters, such as the EA UK newsletter and 80,000 Hours’ newsletter.

You could consider asking for a mention on podcasts, but note that due to podcast production timelines this can require months of notice.

Here is the general approach that we used:

  1. Create Slack and Facebook accounts
  2. Start joining relevant EA Slack channels and Facebook pages
  3. Create a digital poster to include in posts. (Note that if you mention some speakers in the poster itself, you are potentially creating problems if the content changes.)
  4. Copywriting:
  1. Customized links are helpful to track click-through rates for different online venues, and learn about where our interested applicants come from. You can create around 50 free links. In our case, the top performing avenues were from posts shared by organisers of local and professional groups and from Slack channels, but that could easily reflect simply that we put the most effort into these.

It’s probably not worth paying for ads on social media as you can target people who are sufficiently engaged with EA much more efficiently through EA groups.

Closer to the conference itself

The other part of comms is communicating important information to applicants, registrants, and attendees.

There is a lot of information you want attendees to know, and they have limited attention, so make sure you highlight the most important things very clearly. Don’t be afraid to TL;DR yourself. An attendee guide is helpful but won’t be read in its entirety by almost anyone. Even if something is linked from the Swapcard homepage, and emailed to all attendees, people may not see it (some people were not aware that our attendee guide existed). This means there will be things that seem obvious to you but not to attendees (in our case, the existence of various facilities such as the quiet working space and the Lola Lo venue), so important information should be communicated in more than one way.

Although we tried to avoid sending too many emails in a single batch, and to vary the email content somewhat, we still ended up being caught by spam detection, probably as a result of sending 500 quite similar individual emails within a few hours of each other. It’s possible that using a mailing list management system, or just bcc, would work better than mail merge.

Feedback survey results

In no particular order, here is a synopsis of the feedback survey results. (Some other results are described elsewhere in this retrospective, including in the ‘Attendee experience’ section.)

A minority of connections were rated as ‘impactful’: 40%.

Net Promoter Score

% promoters


% detractors


Net Promoter Score









White or European origin


Black, Black British, African American, African Caribbean, or African origin


Asian, Asian American, or Asian British origin


Hispanic or Latin American origin


Middle Eastern, North African, or Arab origin


Resulting actions

82% of survey respondents said they planned to take at least one of a list of actions (e.g. ‘change degree’) as a result of the conference (multiple-choice question), including 14.5% resolving to found an EA organisation and 30% resolving to work full-time for such an organisation or in a primary cause area:

Which actions do you plan to take as a result of this event?

Not "none of the above"


Choose a thesis topic in a Primary EA Cause Area


Have an internship in a Primary EA Cause Area


Change major/degree to study in a Primary EA Cause Area


Choose job/degree in a Primary EA Cause Area


Take the Giving What We Can pledge


Donate more than 10 percent of income in the last year to Primary EA Cause Areas


Contract, volunteer, or intern at an EA Organization or in a Primary EA Cause Area


Receive a grant from an EA Organization


Work part time for an EA Organization or in a Primary EA Cause Area


Work full time for an EA Organization or in a Primary EA Cause Area


Found an EA Organization


None of the above







We wanted our conference to “welcome and integrate less well-connected members of the community”. In the feedback survey, the closest proxy for how well connected someone was is the number of EAG(x) conferences they had been to before.

In the event, people for whom this was their first EAG(x) conference made on average just as many connections as those who had attended once or twice before. 70% of first-timers left with at least 5 connections.

The average rating on the statement “EAGx is a place where I felt welcome” was 4.5 out of 5. This did not vary significantly by number of conferences attended (nor by race or gender).

[1] The feedback survey defined a connection as “a person you feel comfortable asking for a favour. This could be someone you met at the event for the first time or someone you’ve known for a while but didn’t feel comfortable reaching out to until now. A reasonable estimate is fine.”

[2] By ‘all-told’ cost I mean to include the payment of core organisers, which is often excluded from EAG(x) cost-per-connection figures. The figure that’s more comparable with other EAG(x) retrospectives (i.e. excluding the payment of core organisers) is £43.46 (55 USD at time of writing.) A reason not to take such figures quite at face value is that EA conferences benefit a lot from free labour, particularly from volunteers.

[3] The feedback survey defined impactful connections as “connections which might accelerate (or have accelerated) you on your path to impact (e.g. someone who might connect you to a job opportunity or a new collaborator on your work).”

[4] “On a scale of 1 to 10, how happy were you with the food provided on-site?” Average answer 8.3/10, and an overall NPS score of 43. There were 9 ratings <=6, and 49 ratings >=9, out of 92 food ratings overall.