5 ways to use AI in Growth
How to use AI as a sparring partner to become better
👋 Hi there, it’s Pierre-Jean. Welcome to this new edition of The Growth Mind!
In today’s post, we’re going to deep dive together into 5 ways you can use AI in growth.
Before we start, a few words on my vision of AI:
AI tools are incredibly powerful to help you achieve things you already have knowledge of.
But when using AI, you should be able to challenge the results of your prompts and have a clear vision of your objectives.
AI is powerful, but will not replace all your work. You still are the one who knows what you want to do and how you’ll do it.
AI tools will make you better at some tasks for sure. But will not change everything. So use AI smartly.
1) Fill your experiments backlog with fresh ideas
🎯 Objective: Get experiment ideas for your company/product, based on your current objectives and which part of the funnel you want to optimize. For each idea, get inspiration from companies who already implemented this type of experiment.
⚙️ Tool: Bard
We use Bard here instead of Chat GPT-4 because we want real-life examples for our experiments. As Bard is connected in real-time to the web, while GPT-4 is not, it makes sense to use it here to have fresh data.
✍️ Prompt
I want you to act a as Growth Advisor, with a strong experience in experimentation. I'm currently working as a growth manager at BlaBlaCar (website: https://www.blablacar.fr), a carpooling platform. On our platform, drivers publish trips, passengers book those trips.
My main objective this quarter is to increase the number of new successful drivers (new successful driver = driver who successfully made his first carpool). To do that, I need to acquire drivers, make them publish a trip, and then accept the request of passengers looking to book those trips.
I'm looking for 10 new experiments. We can imagine playing on top of the funnel channels, as well as integrating new product features or proposing a more attractive pricing/offer. Could you provide me, in a table format, a list of 10 experiments I could try? For each experiment, add:
A description.
The objective of each experiment.
A real-life example of a company which has done it. I want to get inspiration from other companies.
In this prompt, I used BlaBlaCar, the company where I work, as an example. To replicate this prompt:
Change the name of the company, the description of the product, and the website URL. It’s important to give some context about the company to get high-quality answers.
Change what is your main objective of the moment, to help the AI understand what is your current challenge.
💬 Result
2) Create content for your marketing campaigns
🎯 Objective: Create and iterate on copies for your marketing campaigns (CRM content, Ads…). Adapt messages to different channel formats.
⚙️ Tool: GPT-4 with WebPilot plugin (to scrap your website & landing pages content)
✍️ Prompt
I want you to act as a senior marketer expert in copywriting and powerful creative campaigns.
I'm currently working at Airbnb. We are planning to make a campaign to attract new hosts on the platform. Here is our landing page for Hosts: https://www.airbnb.com/host/homes Please scan it and learn from our value proposition.
We want this summer to base our campaign on 2 main benefits of hosting on Airbnb:
The savings you can do by renting your apartment or house
The coverages we provide in case something happen to your apartment/house
Can you write:
3 examples of Facebook Ads
1 example of email and 1 example of push notification, sent by CRM to our existing host base to activate them
3 example of Google Ads copies
In this prompt, I used Airbnb as an example. To replicate this prompt:
Change the name of the company, the description of the product, and the website URL.
Adapt the benefits/messages you want for your marketing campaign.
Indicate the channels you want to leverage and that you need content for.
💬 Result
3) Calculate A/B test duration & results
🎯 Objectives:
Calculate how much time you need to run your A/B test based on the uplift you want to generate.
Understand the results of your A/B tests. While it can look easy to see which version of a test was the winner, A/B test results are in reality difficult to estimate. You need to make sure your results are statistically significant, based on the MDE (minimum detectable effect) you defined.
⚙️ Tool: GPT-4
✍️ Prompt
Our prompt is divided into 2:
The 1st one is to calculate the duration needed for our test, before launching it.
The 2nd one, once we have the results of the test, to see if our new version is a winner or not.
1st prompt: Calculate the duration needed for the test
I want you to act as an experimented Data Analyst, specializing in A/B testing.
I need to run an A/B test on a product screen to test a new value proposition.
I want an MDE (minimum detectable effect) of 10%.
Each day, 2,000 people visit this screen. The current conversion rate is 11.11%.
50% of users will see version A (the original), 50% the version B (the tested one).
For how much time should I run my A/B test to get significant results? Feel free to take realistic hypotheses to perform your calculations.
Give me a clear text answer with the number of days needed to run the A/B test.
2nd prompt: Analyze results of the A/B test (use it in the same conversation as the 1st prompt)
Here are the results of the A/B test:
In my version A (the original), we had:
12,452 visits
1,384 conversions
A conversion rate of 11.11%
On version B (the new screen), we had:
12,948 visits
1,658 conversions
A conversions rate of 12.81%
Remember: I want an MDE (minimum detectable effect) of at least 10%.
Based on the results above, could you tell me if I can consider my test as a success?
Be very precise by giving GPT the exact number of visits and conversions you had, as well as the uplift you’re expecting to create to consider your test a success.
💬 Results
Calculation of the A/B test duration needed
Analysis of the A/B test results
4) Get an analysis of a data reporting
🎯 Objective: Get a primary analysis of data reporting (acquisition report in our example). Understand the major trends of your reports and get personalized recommendations to improve your metrics.
⚙️ Tool : GPT-4 with Access Google Sheet plugin (to read your reporting)
✍️ Prompt
I want you to act as a senior Data Analyst, with a strong expertise in acquisition.
Here is our report of acquisition for 2023: [LINK OF THE REPORT]
We look at 3 important metrics:
Website visits
Sign-ups (which is our final objective)
Visit to sign-up conversion
Can you tell me how my acquisition evolved over time?
Do we have months where something particular happened?
Should we focus should more on bringing more volumes or increasing conversion?
Be very precise in your analysis, we want to get the full picture with all important details here.
For this prompt, be clear about your objectives and ask for recommendations. If you don’t ask for a lot of details and don’t say which metric you want to optimize, the prompt answer will be way too generic.
💬 Result
5) Present the results of your experiment
🎯 Objective: Present in a concise but powerful way your experiment results internally. Make C-levels and key stakeholders quickly understand the wins and the learnings of your recent experiments.
⚙️ Tool: GPT-4
✍️ Prompt
I want you to act as a senior Growth Manager, with a strong expertise in communicating with C-levels and internal stakeholders.
I'm leading the growth team in my company and I need to present the results of the experiments of the past month.
We ran 3 experiments last month
2 of them are considered successful, 1 is unsuccessful
Experiment 1:
Description: New landing page
Results: +6pp of conversion (16% CVR vs 10% before) -> Successful
Learnings: New value prop tested works better than the old one
Experiment 2:
Description: Higher price for high demand period
Results: +13% of revenue thanks to the higher pricing, but we need to be careful as we lost some demand (-7%). Considered as successful
Learnings: Increasing pricing reduce demand, but as far as we know it was a revenue positive move. Need to check when this move will not be anymore positive with new experiments.
Experiment 3:
Description: New onboarding flow with an integrated product tour
Results: Activation rate increased by 3pp (from 30% to 33%)
Learnings: Educational content directly integrated into our product looks powerful to better activate. We should find new ways to leverage this type of content to continue improving activation.
Please prepare an email to present those results. Be concise, as we are talking to c-levels, but outline our wins and learnings with figures.
Don't keep the same structure than me, start with a global paragraph presenting the main results (what we bring to the company last month) and then outline our learnings & next steps.
As you can see, it’s really important to give details about your experiments, with the main results and learnings of each one. Here GPT-4 helps us to present our experiment results in a concise and powerful way, easily readable and understandable by our audience, but you still need to do the preliminary analysis job.
💬 Result
That’s a wrap for today! 👋
As you saw, I’m always using GPT-4. It’s only 20$/month and the results are really better than GPT-3.5, so in my opinion it’s worth it to pay for.
I hope you liked this article and that it gives you ideas to leverage AI tools smartly. Feel free to share any feedback by answering to this email or commenting on Substack.
See you in 2 weeks for the next article and have a nice holiday 🏝️😎
It's super useful for Google ads and the character limit per ad that's sometimes hard to respect !
Thanks PJ for this article :)