Artificial Intelligence

What the sessions actually say

1. Write the one-sentence version of your product before anything else

   Gael's first step is a piece of paper: what does the idea do, who is the audience, what is the target market, what is the business model. Then compress it to a single sentence, for example "a solution for business owners and entrepreneurs that lets them network and meet each other". That sentence becomes the test every feature has to pass.

   Gael DonnayHow to build an MVP without blowing your budget →

2. Work backwards from your feature wish list to find the MVP

   Take every feature you have imagined and remove them one at a time. For each, ask whether the solution still does what it needs to do and still delivers the value it has to deliver. What survives is the lean MVP. Gael quotes Steve Jobs: simplicity is the ultimate sophistication.

   Gael DonnayHow to build an MVP without blowing your budget →

3. Sort features with the MoSCoW matrix and only build the must-haves

   When scope gets hard and budget is fixed, split every feature into must have, should have, could have and won't have. Only the must-haves belong in the MVP. Could-haves come later if budget allows, and Gael's view is that they generally should not. Be honest in the sorting, and expect your tech partner to push back.

   Gael DonnayHow to build an MVP without blowing your budget →

4. Stop single-shot prompting. Ground the model with project files.

   Cabral says a single prompt gives generic output that everyone can spot. Upload files into ChatGPT (or Claude) and use the projects feature so the same reference material is pulled into every response going forward. Add custom instructions for tone, target audience and Australian English spelling, because American Zs in your copy are an obvious tell that it came from ChatGPT.

   Jeremy CabralJeremy Cabral: AI Agents, SEO and Growth Engines →

5. Ask the tool to write your prompt template first

   Instead of guessing at structure, tell the model your goal: you want a project file that informs all future responses, so give me a prompt template to fill out. Fill it in, load it back into the project, and the LLM has structured knowledge it can reference easily. Unstructured files take more effort for the tool to use.

   Jeremy CabralJeremy Cabral: AI Agents, SEO and Growth Engines →

6. Shop between models for each task

   Cabral used Gemini to write the prompt for Leonardo AI (the image tool Canva acquired), then ran that prompt through ChatGPT because the result was better. Gemini produced the best prompt structure and the most repeatable image outputs. Models change constantly, so test the dropdown rather than defaulting to one tool.

   Jeremy CabralJeremy Cabral: AI Agents, SEO and Growth Engines →

7. Rescue a dying email list by mailing only your engaged segment

   One brand Kelly works with slid from a 30% open rate to 15-16% over six months and assumed email was dead. It was not, the platforms had started flagging them and their quality score was degrading. She cut the send list back to engaged subscribers only, sent about three campaigns to that segment, hit a 40% open rate, then slowly reintroduced the unengaged. They are back at 32%.

   Kelly SlessorKelly Slessor: the ecommerce fixes that actually move revenue →

8. Aim for a 40% open rate, treat 60% as the stretch

   Kelly's benchmark is simple. Thirty per cent is what most brands sit at, 40% is a good, healthy place to be, and she has one brand running at 50 to 60%. If you can get to 40%, your list health and content alignment are working.

   Kelly SlessorKelly Slessor: the ecommerce fixes that actually move revenue →

9. Kill the double opt-in on your welcome flow

   Double opt-in is the default setting in a lot of email platforms and Kelly says it is awful. Someone has just given you their email, which is the digital equivalent of agreeing to a first date, and you reply by asking them to confirm they really meant it. Turn it off and make the welcome frictionless.

   Kelly SlessorKelly Slessor: the ecommerce fixes that actually move revenue →