Inspiration
Every working day, employees commute, use office electricity, print documents and, increasingly, use AI tools. Individually these seem small, but across an organisation they add up.
We noticed that managers have plenty of data about how their teams work, but almost no visibility into the trade-off between sustainability, cost and productivity. Cutting emissions sounds great until it starts affecting performance or budgets. We also realised that one growing source of emissions is almost never tracked: AI usage itself. Companies pay for AI capacity, but allocated capacity isn't the same as capacity actually used.
That gap inspired EcoBalance AI: a tool that helps businesses grow sustainably without sacrificing performance.
What it does
EcoBalance AI is a sustainability intelligence platform that breaks an organisation's operational carbon footprint into four areas: commuting, office electricity, printing and AI usage, then turns that data into decisions.
Our approach: Measure → Understand → Optimise → Reward.
- Carbon Calculator: converts activity data into estimated emissions:
$$\text{CO}_2\text{e} = \text{Activity} \times \text{Emission Factor}$$
$$\text{Total CO}_2\text{e} = \text{Commute} + \text{Electricity} + \text{Printing} + \text{AI}$$
- Employee Dashboard: shows each employee where their own footprint comes from, plus their AI credit allocation and utilisation.
- Recommendation Engine: identifies the biggest driver of each employee's footprint and suggests practical actions. For example, if commuting is more than half their emissions, it highlights lower-carbon commute options; if printing is unusually high, it suggests duplex or reduced printing.
- AI Credit Optimiser: calculates allocated, used and unused credits, utilisation, and the financial value of unused allocation:
$$\text{Utilisation} = \frac{\text{Credits used}}{\text{Credits allocated}} \times 100\%$$
It then recommends a future allocation based on actual usage plus a safety buffer, rather than simply cutting unused credits.
- Incentive Model: compares each employee against their own Month 1 baseline, not against colleagues, so someone living 15 miles away isn't penalised compared with someone living 1 mile away. Improvements are classified into Green, Silver and Gold tiers.
- Manager Dashboard: aggregates team emissions by source, AI utilisation and cost, trends, incentive eligibility and recommendations. It uses metrics like CO₂e per employee so teams of different sizes are compared fairly.
How we built it
We modelled a prototype organisation of 70 employees across three managers (teams of 10, 20 and 40) on a hybrid model of three office days and two work-from-home days. Each employee receives 3,000 AI credits per month ($120).
Data sources were designed around how real companies could collect them: transport mode and distance for commutes (including multi-modal journeys like bus + tram), desk-level energy meters for electricity, print-management systems such as PaperCut for printing, and AI credit logs for AI usage.
- Frontend: React and Tailwind CSS, with interactive dashboards for employees and managers [edit]
- Backend: Python and FastAPI, with Pandas for data processing [edit]
- Database: PostgreSQL [edit]
- Logic: modular calculators for each emission source, plus a rule-based recommendation and incentive engine
We kept every calculator and rule modular, so more sophisticated forecasting and optimisation models can be added later.
Challenges we ran into
- Estimating AI emissions: there's no standard emission factor for AI credits, so we had to define clear modelling assumptions and keep them transparent.
- Fairness: rewarding the lowest footprint would unfairly favour people who live close to the office. Comparing employees against their own baseline solved this.
- Comparing teams of different sizes: total emissions made the 40-person team look worst by default, so we normalised with CO₂e per employee.
- Avoiding blunt recommendations: simply telling someone to "cut 1,000 credits" risks hurting productivity, so we added a safety buffer to allocation recommendations.
- Scope: with limited time, we focused on four measurable areas rather than trying to model everything.
Accomplishments that we're proud of
- Building a decision layer on top of carbon data, not just another carbon dashboard.
- Treating AI usage as an emissions and cost source, which most sustainability tools ignore.
- A fair incentive model that rewards personal improvement rather than circumstance.
- Connecting the full chain: employee activity → carbon impact → resource utilisation → cost → recommendations → incentives.
- Designing for two audiences (employees and managers) with views tailored to what each actually needs.
What we learned
- Sustainability data is only useful when it leads to a decision someone can act on.
- Fairness matters: how you measure people shapes whether they engage or disengage.
- Resource efficiency and sustainability are closely linked. Unused AI capacity is both wasted money and unnecessary emissions.
- Being transparent about modelling assumptions builds more trust than presenting uncertain numbers as precise.
What's next for EcoBalance AI
- Replacing prototype assumptions with authoritative emission factors and validated organisational data.
- Using real historical data to model how sustainability interventions affect productivity.
- Forecasting future emissions and detecting unusual consumption patterns.
- Optimising resources across entire teams, not just individuals.
- Expanding beyond our four areas to sources like business travel, cloud infrastructure and supply chains.
Our goal is to make sustainability measurable, actionable and rewarding: one employee, one team, and one decision at a time.
