What My Solar Atlas Taught Us
A post-mortem on Annaven's first venture: https://mysolaratlas.com
A note on how this was written: Octavian drafted the original over roughly three hours, and it ran around 5,500 words. An LLM was used to condense it to what you see here. The voice and the content are unchanged; nothing was added, softened, or reinterpreted. Only the length was cut.
My Solar Atlas was the first software venture Annaven LLC kicked off. It started during the early half of 2025 and active management ceased around November 2025. Aside from reaching out to a few new users in February 2026, nothing further was built or marketed. Minor maintenance took place during August 2026.
The venture succeeded at sustained action. It failed at product-market fit, at scaling, and at revenue. Writing this is uncomfortable, which is most of the reason we are publishing it. We pay the cost of this failure either way.
Publishing it is an approach where others benefit from what the lessons cost us. The write up also serves as a mile marker which inspires us to build capabilities that ensure we do not repeat the same mistakes.
What we set out to do
The engagement was a three-month MVP (minimum viable product): a catalog of solar suppliers and installers across the most active U.S. markets. It was initiated with Octavian Andrei as Director of Operations covering ideation through development, administration, and marketing. The project was formalized through a service contract and a member loan agreement meant to recognize Octavian’s invested time (which would repaid upon profitability).
Two things were wrong at the starting line. The honorary rate was set far too low, which undervalued every hour that followed. And the plan assumed most of the time would go into building, when the strongest available signal of profitability would have come from a smoke test, a fake door, or a concierge MVP.
The central mistake
Most of the early months went into a data pipeline that ranked the top solar installers in a given area. Only afterward, with little thought given to the presentation layer, did we build a front end around it. By the time the web app was marketed the pipeline had consumed the bulk of our efforts.
Then the real problem surfaced: homeowners assumed we were an intermediary that would sell their information to installers. It had not occurred to us that the exact problem we set out to solve, an industry with no trust left in it, would be applied to us as well. We were treated as the thing we were built to replace.
In August (of 2025) we brought on a marketing freelancer for Reddit, an engagement constrained from the start because linking to the web app risked a permanent ban from the largest solar subreddit. Roughly 80% of that work went into answering homeowners' questions and offering guidance. That was genuinely good work and it helped real people, but it could not function as distribution. We were building goodwill in a room where we were not allowed to say what we had built, which meant the effort could not compound into anything the venture could measure.
By September a user asked for something we had though about but not prioritized: the capability to put quotes side by side. Octavian built the MVP, showed it to that user and others, got good feedback, and handed it to our developer. Then we sunset the capability to find installers.
The pivot itself was correct. The timing was the problem. It came too late, after the spend, rather than up front where it would not have required a pivot at all. The difference matters: we had already invested most of our hours before we learned what people wanted, when we could have found the demand first and built to meet it.
What went wrong
Rather than list every gap, here are the ones that share a root.
We could not see our own user flow. PostHog and Matomo were both used. Neither was configured to answer the three questions that mattered:
of the people who visit:
how many sign up
of those, how many use the app
of those, how many did we help.
That is a handful of SQL queries and a recurring report, and it went unbuilt for the venture's entire life.
We had no way to hold a relationship. There was no CRM (client relationship management) in any meaningful sense. The objects in our backend: user, project, and quote were all awkward to reach from the admin UI, so engaging someone meant a query or a lot of clicking to find an email, then a manual task in our tracker.
Nothing showed us that a person signed up on a date, created a project in a city, and never added a quote. We installed a support chat and never wove it into the experience; no one was told to use it, no one ever did, and we never noticed when it stopped working later on.
Later we realized Helpscout chat stopped working due to the account being deleted after not logging in. The reason we didn’t log in was because requests did not come through the web app. We do not recommend Helpscout due to their practices - although the feature set is impressive.
The cost of all this was the same: when a homeowner stalled, we had no way to know it had happened and no way to reach them while it still mattered.
We removed the only honest scoreboard. There was a paid plan early on. It was removed during the pivot to quote comparison. A paid plan never came back which left us with no conversion tracking, no explanation of the product on the homepage, and nothing that could tell us whether we were building something worth anything to anyone.
It got worse than that. We left the installer request button up as a smoke test and never wired an alert to tell us when someone pressed it, so those requests went into a black hole. And when Octavian finally took a hard look at what remained in August 2026, the sign-up form had been failing on most browsers. Every feature we added past the pivot worked against us, because it consumed attention that belonged to the question of whether anyone would pay us anything at all.
We reacted instead of deciding. The 2025 tariffs, then the confirmation that the federal residential credit would expire at year's end, put us in a permanent fight-or-flight posture. We believed installs could fall 20–30% in 2026 and that if we did not capture interest immediately we never would. So we never ran the retrospective that would have asked the only question worth asking: continue the venture, put it on life support, or end it?
The three-month engagement concluded and yet it was never re-scoped. Out of fear of failing loudly, we chose to fail quietly by continuing with changing priorities and no way of measuring what worked.
Underneath all of it: Octavian did not need this venture to succeed financially, and yet his effort was the most expensive contribution made to My Solar Atlas. Had we honored a profit motive, we would have acted from business acumen rather than duty. Duty did not convert into impact for homeowners. It converted into burnout and fading into obscurity.
The quiet wins
Some things went well. They are worth recording precisely because nobody ever saw them; they produced no revenue, no user noticed them, and in a venture measured only by outcomes they would disappear without a trace.
Listening, then pivoting. Moving from finding installers to comparing quotes came directly from a user, and it put real analytical power in homeowners' hands; including an honest view of what financing does to a return. Over a dozen homeowners were thrilled with the tool or the help we gave them on Reddit. People still use it today, and the sign-up issue has been fixed.
Making infrastructure decisions deliberately. We researched variable sealing early. With a team of two we did not implement it, but we kept secrets off our machines and mirrored QA variables from our provider rather than storing them locally. Our developer preferred Vercel and recommended migrating; we stayed put. By April 2026, Vercel's own infrastructure had been compromised through unsealed variables and a terminated employee's access token.
Let that sit for a moment. A two-person engineering team with zero revenue engaged their provider's support engineers about how to seal environment secrets, understood the risk of the API tokens being compromised, and had a plan ready to implement them for when the team grew.
On the other hand, one of the largest infrastructure companies in the world, running a platform that offered sealing as a feature they themselves engineered, did not use it. In addition, Vercel did not promptly revoke access for a terminated employee. The one freelancer we terminated had their credentials revoked to email, task management, and analytics minutes before they were informed.
Choosing an open-source alternative to GitHub was the same kind of decision. It cost us conveniences like Actions and a CI/CD pipeline that we never built. It also kept us clear of the early-2026 incident that GitHub had where code was silently reverted during merges.
This thought process also foresaw that eventually Microsoft would want to train AI on our codebase and we avoided that altogether. A few months later, Microsoft informed GitHub users that their codebase would be opted into training AI by default. We might have been fine either way. What matters is that both decisions were made the right way: risks reviewed, tradeoff accepted, consequences owned.
Building a team. This was Octavian's first time managing freelancers outside an employer's structure (where he managed part-time / full-time contractors), and it worked. Both engagements were mutually beneficial and played to real strengths, and a third freelancer’s engagement was ended within three weeks because it was not the right fit.
Without Lu and Hernán this venture would have died early and never become a quote comparison tool worth using.
The trouble with quiet wins is the system that housed them. Rather than an ecosystem geared for success, we built engineering maturity that outpaced our ability to acquire a single paying customer. Octavian acquired an understanding of this market that maybe 5% of the people spending $20,000–$80,000 on a system ever reach, and we never converted it into guidance which homeowners paid us for.
A concierge product would have beaten a good app with no users. We could have helped the same amount of home owners with no login, no pipeline, and nothing but a landing page if we had committed to design thinking and not building out of the gate.
What we carry forward
Market before building. Validate with a smoke test, a fake door, or a concierge MVP. Write code after someone says yes.
Point resources at paying customers. Free is not a strategy; it is a deferral of the only feedback that counts.
Own the user flow before the feature set. Keep a working CRM and a funnel you can read in one view.
Hold the cadences. Growth and engineering planning and retrospectives, on schedule, especially when things move fast.
Time-box every phase and honor the box. Go or no-go on pre-revenue ventures, documented while the decision is gets made. Nothing gets decided by default.
Keep it simple. Complexity beat simplicity here, and it did so predictably. Building was the comfortable place to hide from marketing; every hour spent on technical depth was an hour of anxiety soothed, and it felt like progress the entire time it was costing us the venture. That trade has been made by founders many times before, and it ends the same way each time (running out of steam before reaching the destination).
Manage expenses at least biweekly. We bought a year of tooling for three people, then worked with one freelancer for 2.5 months and the other for five. Annual pricing is only cheaper if you last the year. Don’t let your confidence overshadow the reality of a $0 product extrapolated over 12 months.
Value everyone's time at the industry rate. Octavian's was set so low it is embarrassing to print; a realistic floor is would be 10x the honorary rate he set for himself. Under-pricing an owner's time makes the whole engagement look cheaper than it is, including to the people deciding its future.
Find the community before you need it. Ours turned out to be our loudest detractors. Installers saw a threat; homeowners saw another lead broker. We spent our energy defending intentions we never had (we never explored affiliate revenue and never planned to), instead of building where we were welcome.
In closing
Ventures are like climbing grades. In bouldering, problems are rated on a V scale, and the gap between a V1 and a V3 is not a matter of trying harder it is a different set of skills you either have or you don't. We walked into the gym expecting to send (complete) a V3 when our actual capability was a V1.
Empowering homeowners in a market where a purchase happens once or twice in a lifetime, where the incentives have paid sellers $200–500 per project for over a decade, and where the rules changed under our feet mid-build, was always going to be a harder route than we had the technique for. Especially since we never committed to being an intermediary that could compete with the $200-500 incentive other platforms are happy to bank on.
We entered a contentious market at its most volatile moment and kept solving for the solution rather than the problem, where each solution reliably spawned the next list of things to do.
There were easier problems on the wall (so to say). We could have sold software on a monthly basis to people who needed something solved every week, or sold high-ticket solutions (software implementations for small businesses) where a single engagement would pay for itself. Either path would have built recurring relationships with people who had an ongoing reason to talk to us. That is what we should have built first. Not because the solar problem was unworthy, but because we had not yet earned the capability to attempt it.
We support more accountability in the solar industry and would be glad to see someone crack the code. We have to accept that we were not the team to do it in 2025 and we probably won’t achieve it this year. Never the less, we are grateful for the opportunity to learn lessons that we will carry with us forever.
If you want to talk
We are still building at Annaven, and we would rather learn + continuously improve than repeat any of this.
If you want to hear more about what happened with My Solar Atlas or if you see a place where our work and yours might fit together, we would like to hear from you. Leave a comment below or message us directly through Substack. We’re happy to connect!
