What We Built in April (and What We Learned Testing It)
April was the first month we put Recon in front of real businesses outside our own testing. Here's what actually happened.
What Shipped
The core pipeline has been running for a while, but April is when we started treating it like a product instead of a prototype. A few things landed that changed how the tool feels to use.
Stripe checkout and subscription gating. For the first time, someone can go from landing page to paid subscription without us touching anything manually. Scout ($49/mo), Operator ($149/mo), Basecamp ($349/mo). The webhook handler verifies signatures, handles duplicate deliveries, and updates the customer record without human intervention. That felt like a threshold.
Agency dashboard. We've had early conversations with a handful of digital agencies who want to run Recon reports on behalf of their clients. We shipped a first version of a dashboard for that use case. It's minimal on purpose. The waitlist is open. We're charging early agencies $249/mo while we figure out what the full multi-tenant version actually needs to do.
Demand Intelligence. One of the things that bothered me about the original report was the keyword section. It was technically correct but based on educated guesses. We replaced it with a real-data pipeline: we scrape the actual headings and metadata from the customer's site and their competitors' sites, pull 12-month trend curves from Google, mine autocomplete suggestions, and check search rankings via Tavily. The result is keyword analysis that's grounded in what's actually happening, not what the AI thinks is plausible. Each row now has a trend direction, who's ranking, whether the customer is ranking, and a plain-English explanation of why it matters.
Email authentication scoring. SPF, DKIM, DMARC. We added this as a scored category because it shows up constantly in our data. A surprising number of legitimate small businesses are sending email that looks like spam at the DNS level, and they have no idea. It's a fixable problem that hurts deliverability for every email they send, from invoices to newsletters.
What Broke
The agency landing page. We shipped a page for agencies before the underlying infrastructure was ready. The copy implied features that weren't built yet. That was a mistake. We rewrote the page mid-sprint to be honest: here's what exists today, here's what's projected for Q3, here's the waitlist price. It's a better page now. The lesson was obvious in retrospect: don't describe a product you haven't built yet, even in soft launch language.
A Stripe webhook bug. Stripe delivers the same webhook event multiple times, sometimes seconds apart. We weren't handling that. The result was that checkout completions could fire twice, creating two subscription records for one payment. We fixed it with an idempotency table that catches duplicate event IDs before any side effects run. Not glamorous, but if this fires on a real customer's checkout it's a very bad day. We treated it as a P0.
Lambda timeout on long reports. The AI analysis pipeline has three stages that use the larger model for strategic reasoning. That combination can run close to the Lambda limit. We removed one of those stages from the larger model, which cut about a minute off the pipeline time without meaningfully changing output quality. Still something we'll need to revisit as we add more data sources.
What Early Users Said
We ran reports for a few businesses in the Spokane area as demos. Two of them were willing to look at the output with us and give feedback.
The consistent theme: the report feels like something you'd expect from a consultant, not a tool. That's the goal. But it also means the length is a lot to digest in one sitting. The most useful part for both of them was the section that compared their Google Business Profile against specific competitors. That was concrete in a way that abstract scores aren't.
The least useful part was anything that required a follow-up question. A score without context is just a number. We're working on making each finding self-contained, so you don't need to cross-reference multiple sections to understand what to do about a specific problem.
One person asked if they could get the report updated automatically each month without having to request it. That's on the roadmap. It's actually why the subscription tiers exist.
What's Next
A few things in the near term. We need to get out of SES sandbox mode so reports can go to any email address, not just verified ones. That's a request we're putting in with AWS. We're also working through the agency onboarding experience: right now it's partly manual, and a real agency partner needs it to be less so.
The bigger thing is refining what the report actually says. The pipeline can pull a lot of data. The question is which data changes how a business owner thinks about their situation, and which data just adds bulk. We'll keep running demos and asking that question directly.
If you want to see what the report looks like, the sample report reflects the current output. If you want to get one for your own business, the free version is at gonerecon.com.