I built an AI product to a rough prototype, then killed it before writing the real version — and it was the right call. The idea: an agent that simulates a parallel version of you, pursuing the goals you abandoned and emailing you a daily update. What killed it wasn't the concept; it was three numbers — roughly RM3–9 per user per month in LLM costs against a viable ~RM29 price, a self-improvement retention curve that collapses to 5–15% by day 30, and an emotional-support liability a solo founder with six months of runway can't carry. Here's the full teardown, and the one signal that would reopen the file.
Three weeks ago I had a product idea that felt genuinely original.
An AI agent that runs a parallel simulation of you.
You feed it your unfinished goals. The ones you wrote down at the start of the year and quietly abandoned. It wakes up every morning and decides what your other self would do today. Then it emails you a daily update.
Your other self woke up at 6am, drafted the first chapter of your book, and sent a cold email to three potential clients. You have 247 days left in the year.
I built a rough version in a weekend. I wrote a validation plan. I drafted the landing page copy. I started mapping out a 7-day free email trial.
Then I killed it.
Here is exactly why.
The Math Broke Before I Even Launched
The idea lives or dies on daily AI-generated narrative. That means one API call per user per day at minimum.
At current Claude Sonnet pricing, generating a 300-word personalised daily update costs roughly RM0.10 to RM0.30 per call depending on context window usage. Across a 30-day month, that is RM3 to RM9 per user in raw COGS before hosting, email delivery, or any other infrastructure.
If I priced this at RM29 per month, a realistic entry point for a self-improvement tool in Malaysia, my gross margin ceiling is somewhere between 70 and 90 percent at best. That sounds fine until you account for churn, support load, and the fact that this category has notoriously bad LTV.
If I priced higher to protect margin, I would kill conversion before I even had users to churn.
There is no pricing band where this feels like a good business at early stage. Not for a solo founder with under 6 months of runway. I am bullish on AI when it sits in the critical path of a workflow and genuinely earns its cost; a daily narrative email is the opposite — pure recurring cost wrapped around fragile value.
Self-Improvement Products Have the Worst Retention Curve in SaaS
This is not my opinion. It is a documented pattern.
Apps in the habit tracking, journaling, reflection, and self-improvement category see median day-30 retention rates between 5 and 15 percent. The novelty of the experience drives initial engagement. Once the novelty fades, which it always does, there is no functional need pulling the user back.
The parallel self idea has a specific version of this problem.
For the first week, the updates feel eerie and motivating. By week two, you start skimming. By week three, the email lands in your inbox and you feel vaguely guilty, then annoyed. By week four, you unsubscribe.
What I would be selling is not a product. I would be selling a feeling. And feelings do not retain.
Email Signups Validate Curiosity, Not Willingness to Pay
My validation plan was a 7-day free email trial. No credit card. Sign up, give me your goals, get your first parallel self update.
The problem is that this format selects for curious people, not paying customers.
The conversion funnel for a free trial without a payment commitment looks like this: a large number of people sign up, a smaller number engage past day two, a very small number would ever consider paying, and an even smaller number actually would.
I have seen this play out before. High open rates, low conversion, and a founder who mistakes engagement for validation.
If I cannot get someone to put in a credit card for a 7-day trial of something this emotionally compelling, the product does not have a paying market. It has a curious audience.
There is a difference.
The Emotional Weight Would Create a Support Nightmare
This is the one nobody talks about when they pitch ideas like this.
What happens when someone feeds the agent a goal they have carried guilt about for years?
I was supposed to write that book. I told myself I would start the business. I promised myself I would be different this year.
Now they are getting daily reminders of the version of themselves they did not become.
For most users, that would be fine. For some, it would not be.
I am not equipped to handle that. I am one person building a product between client calls and business operations. I do not have a support team. I do not have a mental health framework baked into the onboarding. I do not have the capacity to triage what that inbox would look like after 1,000 users.
The liability is real. The reputational risk is real. And the ethical weight of running a product that could genuinely harm vulnerable users is not something I wanted to carry without proper infrastructure.
I Cannot Afford Another Silent Build
I have six months of runway.
I have one product already generating revenue. I have a content platform in early distribution. I have a team that depends on the decisions I make right now.
The parallel self idea would have taken six to eight weeks to build to a state worth validating properly. That is six to eight weeks of focus pulled from things that are already working. The discipline that keeps my client builds shipping in 12 weeks is the same one that tells me when not to build at all.
Founders romanticise the pivot. The scrappy midnight build. The bold bet.
What they talk about less is the quiet cost of building the wrong thing while the right things starve for attention.
I am done making that trade.
What Would Change My Mind
I am not saying this idea is dead forever.
If I received 10 or more unsolicited emails from people saying "I would pay for this, here is what I would pay," I would revisit every assumption above.
Not signups. Not likes. Not comments saying "this sounds cool."
Emails with a number in them.
That is the signal that would reopen this file.
The Honest Takeaway
The best product ideas often carry a specific kind of trap.
They are emotionally resonant. They solve a real feeling. They are technically buildable.
And then the unit economics collapse, the retention curve is brutal, and the support requirements are invisible until they are not.
I killed this one before it killed the runway. If you are staring at an idea of your own, the free audit I run for clients is essentially this teardown applied to your stack — economics, retention, and risk, before a line of production code.
If you would genuinely pay for a product like this, I mean actually pay, email me at saimon@agradeprestasi.com.
Tell me what you would pay. Tell me what would make it sticky past day 30.
If 10 people do that, I will build it properly.
If not, you now have a framework for killing your own ideas before they kill you.
Frequently asked
How much does it cost to run a daily AI agent per user?
At Claude Sonnet pricing, one roughly 300-word personalised update costs about RM0.10–0.30 per call, or RM3–9 per user per month in raw LLM costs before hosting and email. Against a realistic RM29 price that never left a margin I was comfortable with at early stage.
Why do self-improvement apps have such bad retention?
Habit-tracking, journaling, and self-improvement apps see median day-30 retention of about 5–15%. They sell a feeling rather than a functional need, so once the novelty fades nothing pulls the user back.
Do free email trials validate a startup idea?
No. A no-credit-card trial selects for curiosity, not willingness to pay. The real signal is unsolicited emails from people naming a price they would pay — engagement and signups get mistaken for validation all the time.
When should a solo founder kill an idea before building it?
When the unit economics, retention curve, and support or liability load all point the wrong way against your runway. I had six months of runway and a product already earning revenue, so a six-to-eight-week build on a fragile idea was a trade I could not justify.