I’m hoping to get some feedback on my approach to solving a client’s problem using Glide and specifically my current workflow.
Goal: I can press a button and get a list of leads I can call.
Yes that sounds easier said than done 
Here is the detailed version:
Client sells non-gmo food ingredients.
- Specify an product category, e.g. energy drinks
- Provide a URL and search for similar companies using the keyword “energy drinks” (via exa.ai Find Similar API)
- Get specified number of websites that match query along with sub-pages from top-level domain based on “list company location, products and ingredients” as response. Result could be …com/our-products or …/store or it could be that each category of product has it’s own sub-page, e.g. …com/jelly-lollies, …/sauces…
- Click a button!
- Enrich the org data (using Apollo API)
- Use AI to check for the most likely place to start looking for products and their ingredients. Example: chobani.com.au/our-products, however depending on the site structure, this page may not be obvious or it might just be …com/
- Use Apify’s API to crawl the specified URL (that is most likely to contain product and ingredient info), call a Glide webhook and request data as JSON which is stored in a glide column.
- Use AI to read this response and identify a list of products and their ingredients.
- Take the non-gmo ingredient list my client sells and compare it to the crawl data/ingredients, and get a list of ingredient categories that a client uses, e.g. sweetners, starches, protein powder, as well as “[non-gmo ingredient] used in [client’s product, e.g. snack bar” so the sales rep knows what to discuss with lead. Using AI for this vs. logic, but I think I’d need to split data into arrays to get the lists I need.
- Find any people in specific roles at the lead company, e.g. R&D (using Apollo API)
- Find any news or other info about the company that could be a hook or opening discussion point, including any companies that the person worked at previously that my client has also worked with, based on organisation names in Pipedrive.
- Create a summary in Glide for each call using all the inputs above.
- Allow for call summary notes
- Generate follow-up email, edit and send using a Glide action.
And that’s it! Steps 4…12 are executed via a workflow that is triggered at the push of a button.
I’m most interested in feedback on the AI-powered steps especially:
6: Not all website structures are the same so my prompt needs to account for this and fall-back to the high-level domain if nothing suitable is found, maybe a confidence score is needed in prompt?
8: Not as hard as my prompt returns JSON,
9: Hardest and most crucial: I need to know which ingredients to discuss and which product the lead will use the ingredient in, e.g. sweetners (used in Jelly lollies) and to get output in a form I can use to display lists.
I’m thinking these prompts might need to be document to text or something other than generate text but open to suggestions.
Here is a screen showing the steps broken down so I can trigger and test each step in the workflow, but they are actually all triggered by a Research button one screen up from this one.
I guess the final state will be that once the research step has been completed and the summary generated, then the status and/or button will update or be hidden and replaced with “View call summary” or similar.
Yes there are a lot of steps, but each one on its own is reasonably straight forward. As with all AI-powered workflows, getting the right data in and outputs out is key which is what I’m hoping you can share your collective wisdom on. Thanks!!!