Cost-effectiveness of sending personal messages

By Vasco Grilo @ 2022-03-29T14:38 (+8)

I have just published an article about effective donations on the online newspaper of my former university. I have already shared it in some groups, but was wondering about how cost-effective would it be to promote it via personal messages. As I could not find much information about this topic, I decided to share some thought here.

Cost-effectiveness considerations:

• Writing personalised messages tends to increase the impact per message, but increases the cost (time) per message.
• Thinking about the people to which sending the messages could be more impactful increases the impact per message, but also increases the cost per message.
  • Intuitively, I would expect most of the impact to come from a minority of people who would not be easy to identify a priori.
  • Consequently, cost-effectively selecting the targets seems hard.
• What does a good standardised message look like?
  • If the same standardised message is sent to most people, writing a longer message does not significantly increase the cost.
• It seems possible that the cost-effectiveness is maximised for some degree of personalisation.
  • Organisations looking for applications often send emails which contain the name of the recipient.

Example of impact required per message:

• I estimated (via testing it) that I could send standardised Facebook messages at a speed of about 10 per minute, i.e. 600 per hour.
• This means that, if I value my time at 6 € of effective donations per hour, the expected impact per message should be at least 0.01 € for sending messages to be effective.
• An expected impact of 0.01 € per message could be thought of as the recipients of the personal messages having:
  • Probability (p(V)) of 10 % of viewing the article.
  • Probability (p(P|V)) of 1 % of being persuaded to make effective donations given that they view the article.
  • Expected donation (D) of 10 € given that they are persuaded to make effective donations.
  • No expected counterfactual donation (CD) if they had not received the personal message.
  • (A Monte Carlo simulation might be useful to model the effect of these factors.)
• It is unclear to me whether this is feasible.
  • I published an article about the book "The Life You Can Save" on the same online newspaper about 1 year ago. Based on the results of a linked questionnaire, from the people who viewed the article:
    • 2 % (17/800) said they would make a donation to an organisation recommended by The Life You Can Save or GiveWell.
    • 2 % (19/800) said they would establish a monthly donation to an organisation recommended by The Life You Can Save or GiveWell.
  • I expect these results to overestimate the impact of the article.
    • I think people tend to overestimate their future altruistic intentions.
  • I guess p(P|V) and D (see above) would be smaller for the recipients of my personal messages than for the people who viewed the article about TLYCS.
    • People who viewed the article about TLYCS had some interest in it.

Any comments would be welcome!