Before we start…

Which model leads the next benchmark? Which AI lab ships the next major breakthrough?

Kalshi lets you trade on real-world AI and technology events as the industry moves. If you follow launches, model updates, and benchmarks closely, put that knowledge to work.

Trade $25, Get Up to $500

_{Bonus credit varies from $15 to $500. Terms apply.}

Let’s get into it.

Every commercial conversation at Bestapps lands in the same place. Gmail.

A sponsorship request sitting between a newsletter and an invoice. Partnership stuff. Cold pitches from brands, agencies and solopreneurs.

For a while that was fine. I'd open the inbox, work top to bottom, and everything got a reply the same day.

Then the volume moved. More commercial emails a week now, and my time to first reply went from same day to 2, 3, 4 days.

That number IS the business. There’s an inverse correlation -I’d say of around at least 0.9- between response time and odds of closing the deal.

A few things were doing it.

Volume. Every relevant email arrives mixed in with the ones that aren't. To find the four that mattered I had to read the sixty that didn't, and I was doing that scan two or three times a day.

Time zones. Inquiries come from all over the world: US, Europe, India, China. Sometimes their morning is my evening, so one round trip on a thread burns two days before anyone has said anything meaningful.

Nothing looked more important than anything else. A $1,000 sponsorship inquiry and a cold pitch for an affiliate deal are the same grey line in Gmail until you open them. Priority only existed in my head, and only for the threads I happened to remember.

I could see the messages but not the business.

I couldn't answer any of those without opening twenty threads and rebuilding the timeline in my head. Every single time.

So I built an internal app to answer that. I've been calling it “Bestapps Agentic CRM“, which sounds a lot greater than what it actually does, which is quite simple: Mail in, info organized and visualized with beautiful charts, autoresponse with all the required information.

It doesn't sit there refreshing my inbox. A message lands, Google Pub/Sub pushes a notification, and the app pulls that thread. Read-only, so it can't send or delete anything.

Junk gets thrown out first. Newsletters, receipts, no-reply addresses. None of it reaches the AI, which keeps the bill down.

The easy facts it works out on its own. Who replied last, how long a thread has been sitting there. That's a date and a comparison, so every conversation gets a status: waiting on me, or waiting on them. You don't need AI for that.

The AI only gets the judgment calls. Is this person interested, or being polite? Real sponsorship offer, or someone after a free link? It comes back with a category, a priority, and how sure it is.

Then it links the sender to their company and any open deal.

Replying is a setting with three positions. Off, where it just tells me. Draft, where the reply is waiting and I send it or rewrite it. Or send, where it handles the whole thing.

I started talking to ChatGPT. The conversion ended with me requesting a PRD (Products Requirement Document):

I saved it as PRD.md inside a folder in the project and I gave cursor this prompt, in Plan mode, running Grok 4.5 High and GPT 5.6 Sol Medium.

Once the plan looked right, the loop was the same every time:

Human review stayed between every major step. Especially around the Gmail sync and the CRM matching, where a quiet bug doesn't crash anything, it just poisons your data for two weeks before you notice.

The models did most of the typing. I kept the product decisions and the "is this actually done" call.

A big agentic build gets dramatically easier when the agent has a plan, one small task in front of it, and a finish line you can test. Take any of those away and you get a thousand lines of confident nonsense.

The frontend shows me the work waiting: replies due, overdue follow-ups, new opportunities, and the AI calls that need a human.

If you want the boring details: it's TypeScript top to bottom. React on the front, Tailwind for styling. Express API in Node. MongoDB for the data. Gmail comes in through read and write OAuth, and the sync runs with Google Pub/Sub. Classification and autoresponders are GPT 5.6 Luna with few-shot training.

It lives on MongoDB Atlas, Railway for the API, Vercel for the frontend. Costs about $20 month, which is less than the CRM seat we’re not paying for.

Say a closed deal is worth $2,000 to you, and 20 conversations worth having land in your inbox every month. You close a quarter of them. Five deals, $10,000.

Now you reply first instead of fourth, and you close 20% more. Six deals instead of five.

That's $2,000 extra a month. $24,000 a year, from answering faster.

The app runs on $25 a month.

That's the whole case for it.

If you have any question about any workflow and how to automate it, here to help.

Stefano Martell
Co-Founder, The AI Leverage by Bestapps.ai

ps: We’ve made Bestapps Academy Free for the first 100 (50 slots remain) https://www.skool.com/bestapps-academy-8162/about