Aug 13, 2024
Advertising AI Governance on LinkedIn, Twitter and Politico
💡 Also published on the BlueDot blog.
The goal of our paid marketing is to reach people who are outside of our existing network and increase the number of exceptional applicants to our course. For the 2024 August iteration of the AI Governance course, we received >1,100 applications, where ~15% of accepted applicants were generated from paid marketing.
This was our first time running paid marketing campaigns for the AI Governance course. In the past, we relied solely on organic and referral-based marketing. We applied our learnings about targeting and testing ad adjustments from running paid marketing campaigns for the AI Alignment (2024 March), Pandemics (2024 May) and AI Alignment (2024 June) courses.
Based on the cost per accepted applicant, these cumulative improvements has made this the second best performing LinkedIn campaign (out of 4) and the best Twitter campaign (out of 2). The CPA (accepted) dropped by ~11% for Twitter.
| Platform | CPA | CPA (accepted) |
|---|---|---|
| £18.47 | £154.21 | |
| Twitter / X | £114.75 | £306.6 |
| Politico | £3432.49 | £3432.49 |
#1 Create mockups of your ads
We tried running ads with Politico for the first time and were surprised by how poorly it performed.
In retrospect, when we made mockups of our ads, we think they did not perform well because:
- The text was too small to read
- The ad looks very busy and cluttered
- It is not clear what the ad was about / why people should click on it
Some evidence we have for this is that half the ads we ran where the size of the ads was much bigger performed significantly better than those with smaller placements. Generating a 5x better click-through rate and performing 1.5x better than Politico's benchmark for ads in this placement.
We are running another ad campaign with the Bulletin for the Pandemics (2024 Sept) course, using two sets of creatives. One obviously calls out the course, and the other flashier ad has less information. Hopefully, this will help us learn what ads work best for media platforms like Politico.
#2 On attracting applicants with certificates
On LinkedIn, we tried out a creative that promoted getting a certificate as a benefit of our course. The hypothesis was that our target audience values certificates and would be more likely to apply to our course.
The creative generated the cheapest engagement in cost-per-click (CPC) and cost per application (CPA) but had among the worst acceptance rates.
| Creative | CPC | CPA | CPA (accepted) | Acceptance rate |
|---|---|---|---|---|
| Earn Certificate | £0.41 | £10.83 | £162.44 | 6.67% |
| Are you an ML engineer | £0.64 | £19.94 | £49.86 | 40.00% |
| Don't Get Left Behind | £1.06 | £38.34 | £125.25 | 30.61% |
| AI powered | £1.10 | £42.45 | £382.07 | 11.11% |
| AI risks | £1.09 | £44.97 | £764.46 | 5.88% |
| Headlines | £1.02 | £49.20 | £295.20 | 16.67% |
| Stochastic Gradient Descent | £1.47 | £79.11 | - | - |
| AI risks & benefits | £1.07 | £82.72 | £268.83 | 30.77% |
We’ll need to adjust the targeting to improve the acceptance rate, but given that the value add seems to resonate with many of our applicants, it seems worth it for us to continue running a version of this creative.
#3 Test creatives on cheaper platforms
We spent much less time making adjustments and experimenting with creatives in this campaign than we have with all the other campaigns thus far. Only 3 out of the 8 concepts were new, compared to the>25 different creatives tested on the Alignment course.
This meant that when we ran an ad campaign with Politico, we had a limited number of choices for creatives and not as high confidence that they would work well. Given the other priorities we had, I don’t think we would have invested more time, but it is good to flag that this was a missed opportunity to gain more information on the creatives on a cheaper platform like LinkedIn and Twitter before investing in a more expensive platform like Politico.
We’re excited to continue exploring other avenues for promotion! If you have any suggestions on where we can find excellent people who are keen to shape the positive development of future AI systems, drop us an email at team@bluedot.org.