Echo V2
Speech recognition
- 100% renewable energy (Kao Data UK)
- 0 kg of waste to landfill
- Target: carbon neutral by 2030
- HVO biodiesel backup (–90% CO₂)
Environmental impact of AI voice documentation in healthcare
One complete 30-minute Wellcom report is estimated to emit less CO₂ than sending an email, backed by public sources. Discover how we bring sustainability and healthcare innovation together.
Aligned with CSRD / ESRS E1, Climate change · Version May 2026
What does AI use mean for our environment?
One complete Wellcom report (30 min) is estimated to emit less CO₂ than sending an email.
- CO₂ per session (30 minutes)
- ~0.07 g
- Energy per session (less than 2 min of Netflix)
- ~1.48 Wh
- Renewable energy at Echo V2 and DeepL
- ≥ 98%
Methodology & sources
The figures shown are best-effort estimates based on public data from our suppliers and independent studies. The actual impact varies by session, model and energy mix.
Sources: Epoch AI, IEA, Energy and AI (2025), Carbon Trust / University of Bristol (2021), Google Cloud (2025), Patterson et al. (2021), Kao Data.
Three AI services work together for one report.
From spoken word to structured care report, each component with a minimal footprint.
-
Speech recognition
Echo V2
The healthcare worker dictates the report. Echo V2, our fine-tuned speech model, converts speech to text: compact, fast and cloud-based.
Energy: ~0.75 Wh Emissions: ~0.056 g CO₂e
-
Report generation
Azure AI Foundry
The transcript is turned into a structured care report. API call via Microsoft Azure.
Energy: ~0.65 Wh Emissions: ~0.049 g CO₂e
-
Translation
DeepL
Optional: the report is translated into the language you need. Compact translation model, 98% renewable energy.
Energy: ~0.08 Wh Emissions: ~0.006 g CO₂e
Total per report (30 min): ~1.48 Wh | ~0.07 g CO₂e
What is 0.07 grams of CO₂?
Compared with everyday digital activities.
- Wellcom report (30 min) 0.07 g
- Google search 0.2 g
- Watching Netflix for 1 min 0.9 g
- Email without attachment 4.0 g
- Streaming Netflix for 30 min 27.5 g
- Email with attachment 50 g
An email with an attachment is estimated to emit hundreds of times more CO₂ than a complete Wellcom report.
Annual emissions for a healthcare organisation.
Calculation for a healthcare organisation with 100 employees.
100 employees × 3 reports per day × 220 working days × 0.07 grams of CO₂ per report = approximately 4.6 kilograms of CO₂e per year.
4.6 kg CO₂e
per year, total portal use
350 kg CO₂e
1 return flight Amsterdam–Barcelona
8,100 kg CO₂e
1 Dutch person, per year
The annual AI use of 100 employees emits less than a single plane ticket.
All three partners invest in renewable energy.
Microsoft Azure
Report generation
- 34 GW of renewable energy contracted
- PUE < 1.12 (industry: 1.4)
- Carbon negative by 2030 (Microsoft)
- A100/H100 GPUs: 5× more efficient
DeepL
Translation
- 98% renewable energy
- GHG inventory for Scope 1/2/3 in progress
- EcoDataCenter partner
- Compact, efficient translation model
Sources: Kao Data (2022) · Microsoft Environmental Sustainability Report 2025 · DeepL
Training vs. use.
News about AI emissions is almost always about training models, not about everyday use.
Model training
One-off, enormous
- Training a large language model (GPT-3): ~1,287 MWh
- Equivalent: 552 tonnes of CO₂
- Comparable to more than 100 cars driving for a whole year
- Happens once per model generation
- Cost shared across billions of calls
Inference (use)
Daily, minimal
- Language model per query: ~0.34 Wh
- Wellcom session (30 min): ~1.48 Wh
- Equivalent: less than 2 minutes of Netflix
- Suppliers running on renewable energy
- Efficiency is improving fast (33× in one year)*
* Google reported in 2025 that the energy use per Gemini prompt fell 33× in one year, thanks to hardware and software optimisation.
CSRD and double materiality.
Aligned with ESRS E1, E5 and S1.
| ESRS | Topic | Scope | Quantification | Classification |
|---|---|---|---|---|
| ESRS E1 | Climate change | Scope 3, Cat. 1 | < 0.005 tonnes/year | Not material |
| ESRS E5 | Resource use | Water (suppliers) | Qualitative (supplier data) | Low materiality |
| ESRS S1 | Own workforce | Less admin burden | Positive social impact | Positively material |
| Double mat. | Impact + Financial | Both perspectives | Not material on either axis | No discl. required |
Recommendation: report AI use as non-material Scope 3. Refer to supplier reports for substantiation. Highlight the positive ESRS S1 impact: less administrative burden for healthcare workers.
Questions about sustainability.
How much CO₂ does one Wellcom report emit?
One complete 30-minute report uses about 1.48 Wh of energy and emits around 0.07 grams of CO₂e. That is less than sending one email without an attachment (around 4 grams).
How does that compare with other digital activities?
A Google search accounts for around 0.2 grams of CO₂, watching one minute of Netflix for around 0.9 grams and an email with an attachment for around 50 grams. An email with an attachment is therefore estimated to emit some 700× more CO₂ than a complete Wellcom report.
Do Wellcom’s AI suppliers run on green energy?
Mostly. Echo V2 (speech recognition) runs on 100% renewable energy and DeepL (translation) on 98%, according to the sustainability information of the data centre and of DeepL. Microsoft Azure (report generation) does not give a percentage, but has contracted 34 GW of renewable energy and aims to be carbon negative by 2030.
So why do you read that AI uses so much energy?
Reports about AI emissions are almost always about training models: a one-off investment per model generation (a large language model such as GPT-3: around 1,287 MWh, around 552 tonnes of CO₂). Everyday use, inference, is minimal: a 30-minute Wellcom session uses ~1.48 Wh, less than watching 2 minutes of Netflix.
Does our organisation need to include Wellcom’s AI use in its CSRD reporting?
The emissions fall under Scope 3 and, at less than 0.005 tonnes per year, remain well below any materiality threshold. Our recommendation: report AI use as non-material Scope 3, refer to the supplier reports for substantiation and highlight the positive ESRS S1 impact: less administrative burden for healthcare workers.
Sustainable documentation, without compromise.
The key findings at a glance.
- 0.07 g CO₂e per report, comparable to 5 seconds of watching TV
- Echo V2 and DeepL run on ≥ 98% renewable energy
- 100 employees, 3 reports/day → ~4.6 kg CO₂e/year in total
- Less than one return flight Amsterdam–Barcelona
- CSRD classification: non-material Scope 3, positive S1 impact
- Voice-driven documentation = less admin, better quality of care