Inspiration
New AI models are coming out every week, every company is trying to capture value and marketshare with them. However, the question is, “does this AI actually generate value?” Separately, CEO’s are tasking devs to consume tokens equivalent to half their annual salary. “Are these tokens worth it?”
Right now, the market place can’t answer this simple question; it can’t answer the ROI of token usage. In a world where SaaS demand for tokens is at all an time high and foundation models are transitioning to a pay per usage model, this yields extensive cost exposure.
Introducing Paretokens.
What it does
Our company aims to map token consumption to internal KPIs to inform enterprise CFOs and CTOs of the real cost of AI integration and opportunity.
Our MVP targets SaaS dev teams, isolating JIRA data points to token consumption to then API costs (OpenAI, Anthropic).
This employee level data is then fed into a dashboard for key customers to understand the health of their company. However, reading a graph mapping JIRA story points vs. token usage may not be digestible for a CFO, in turn, we smooth the interpretation with OpenAI integrated recommendation engine. This allows the CFO to understand why a certain employee may be consuming tokens at the rate they are, answers if it is reasonable or even something to celebrate, then offers a recommendation on next steps such as: “Dev team used Opus 4.7 for light tasks, the story points against token usage was low, recommend guardrails to only offer Haiku 4.6 for email writing. Accept Recommendation? Escalate to CTO? Defer?”
How we built it
We had a team of 3 developers, 1 designer, and 1 business perspective. After 6 hours of deliberation, we finalized our idea. Hitting the ground running, we developed a wire frame including front and back end outlines.
Leveraging Loveable, Python, and OpenAI, the project was split across MVP data sources, front end, and back end development. The process started with defining our core data set, finalizing an initial mockup then optimizing tool token consumption to ensure our MVP was viable and efficient.
Challenges we ran into
Open source data was difficult to access for JIRA projects which served as the core integration step and moat of our product. In turn, a significant amount of time was finding and curating a strong JIRA candidate for our MVP.
Accomplishments that we're proud of
The product is cleanly coded and a first of its kind. It's addressing a real industry problem that will face every enterprise looking to utilize AI. We're particularly proud with how we acknowledged the real difficulty of the problem, came up with a solution, and created real value for potential customers.
It is a strong idea, a strong product, and a strong opportunity.
What we learned
We learned about the marketplace and the gaps a tool like this could fill. We learned the risks of passive vibe coding and the horrors it can have on the back end. Finally, we learned that planning is the most important part of product development.
What's next for Paretokens
We are excited to continue fleshing out the product to maintain a first mover moat. This looks like expanding recommendations beyond dev and SaaS teams to intensive knowledge clients such as lawyers and consultants. Then we'd look to actually commercialize the product through our haircut pricing model of 1% margin on top of your AI spend.
Log in or sign up for Devpost to join the conversation.