Howdy wizards,
I keep improving my free GenAI use case tracker week by week. You can now see a preview of the top performing case studies within each use case! Truly appreciate those of you who share it with your friends — it helps me out a lot 🤝
Here’s what’s brewing in AI this week.
Which AI use case shows the best business results?
I’ve analysed 2,843 case studies from the last 12 months. When looking at all industries combined, there’s one particular use case that consistently shows outsized returns.
It’s lead qualification.
Source: Context Windows
Why this use case is golden:
Around 30-40% of web inquiries and leads arrive after regular business hours. Using AI to respond to these immediately, when they’re the hottest, is proving effective.
Sales teams have limited capacity, and often waste time on low-quality leads. The AI agent acts like a filter for their sales pipeline – it prioritises and routes high-value leads to the human sales reps.
It’s relatively easy to implement. There’s several ready-made AI tools for this exact use case and the main integration point is the CRM.
Examples from real implementations:
Lennar, a property company, was struggling to respond quickly to leads that arrived after business hours and also to prioritise a high volume of leads. They implemented AI that autonomously answers specific property questions from web visitors, books ≈12,000 home tours per year, and prioritises leads for the sales team.
Culligan, a water solutions company, drove $650K revenue in 4 weeks with an AI agent that handles website inquiries from dealers, communicates with them via SMS and schedules appointments autonomously — then updates their CRM after.
Asana boosted their pipeline 22% after implementing an agent that qualifies leads and hands off only high-value leads to their sales reps.
This use case has a ton of flavours and the above are just a few examples.
It’s one of those easy wins for companies looking to prove internally that AI works, e.g. before investing in more longer-term, complex initiatives.
Speaking of growth opportunities, this week I’ve teamed up with Galactic Fed, a marketing agency with a track record of helping over 600 notable brands succeed:
IN PARTNERSHIP WITH GALACTIC FED
The hidden team behind unicorn growth for Suno, Jeeves, and Picsart.
We’ve hired agencies. Most sucked. So we built one for founders like us.
Free growth plan for What's Brewing in AI readers. We’ll show you the levers to scale fast in your niche over 6 months.
NEWS NEWS NEWS NEWS ❦ NEWS NEWS NEWS NEWS
Hot tip: Claude just launched Routines
Routines in Claude Code lets you save a task and run it on a schedule — and you can trigger it via an API call or in response to an event.
A task here means a prompt, a repo, and a set of connectors.
Your laptop doesn't need to be on for it to work, as it executes on Anthropic's cloud infrastructure.
The Routines feature is available in research preview now to anyone with a paid Claude plan.
Why it matters Routines makes real automation accessible to more people.
Until now, automating Claude tasks meant either leaving your computer on or building your own server with cron jobs. I’ve been doing the latter (explained in more detail below), but there’s some technical complexities to it, and here’s where Routines comes in and makes it easy.
Because Routines have their own API and webhook triggers, it means Claude now does something similar to no-code tools like n8n and Zapier, except you create tasks by prompting instead of drag-and-drop nodes (easier if you ask me).
Bonus: How I’m currently automating Claude tasks for myself (Claude Code on VPS)
I’m writing about this to show you:
1) An example of a real task I’ve automated to give you inspiration
2) That there are flexible and cheap alternatives to new platform features like this
Here’s what I’m doing: I have a Hetzner VPS (an always-on computer in the cloud) with cron jobs (scheduled tasks) and Webhooks that invoke Claude Code (CC must also be installed on the VPS).
I use it to do daily tasks and tasks that trigger on specific events, and then send me an email after with the results.
An example of a Claude task I run on my VPS (a data quality check)
Each case study featured on Context Windows needs to have a verifiable company that implemented the AI solution. In 99% of cases, my automated pipeline is able to do the job.
In the edge cases where, for some reason, the pipeline can’t verify the company, it gets flagged for review (could be access restrictions, their website is down, company doesn’t exist, changed their name, etc.).
When that happens, my Claude Code agent on my VPS gets invoked, gets all the relevant details and proceeds to search the web and track down the company.
Then sends me a weekly summary of how many it was able to actually verify.
The good and the bad of doing it this way
The big benefit of this is that it’s platform and LLM agnostic, ie I can easily switch my automations to another coding agent if I want. It’s also infinitely flexible, I can connect it to any workflows, models, or apps — of course I have to set it all up by myself (with help of Claude).
The downsides to this method is that it can be more brittle. Occasionally things break and I have to troubleshoot a bit. For example, this week I suddenly stopped getting my daily emails from Claude about the results from its tasks. Turns out the authentication to Resend (the email service I use) had expired. It took me a couple of days to notice I wasn’t getting the emails, and then around 5-10 minutes of back and forth with Claude to fix it.
If you want the easy way, just use Claude Routines
If you want the easiest way to create scheduled automations, just use Claude Routines.
It’s the least technical fuss with the setup, and Claude’s built-in connectors (Gmail, Calendar, Slack, Notion, etc) come ready to use in your automations.
However — If you enjoy tinkering with AI just for fun, or you want to use more bare metal when building with AI, then go VPS. You’ll have full customisation possibilities and avoid vendor lock-in. Plus, a VPS—which you can use for a ton of things in this age of AI—only costs ≈$6/month.
❦
Hope that was helpful to someone! Let me know if these breakdowns get too technical, or if you want me to go deeper and tell you more about my own setups — just hit reply.
You are a delight.
Dario
(ps don’t be a stranger — connect with me on LinkedIn)
What's your verdict on today's email?
Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.