Product for the People — November 2026 — notes
Product for the People on Thursday 19 November 2026, at Hippo Digital in Leeds.
The 10th one, my fifth, and a full day on digital sustainability (the first time there's been a theme?). Four talks and a panel.
There are my notes. They might contain errors/misrepresentations. Apols - let me know if you spot any,
1. Steve and Ishmael (DWP): introduction to digital sustainability
Why it matters
- World Economic Forum risk survey has environmental risks dominating the long term list
- Deloitte survey of company executives in 2024, 70% expected climate change to hit strategy and operations within three years
- Heat this summer closed offices and sent people home
- Planetary boundaries are nine limits (fresh water, land use, biodiversity and so on) that keep the planet liveable. Seven of the nine are now crossed. They interact. Compared it to organs in a body, you wouldn't only check your heart
- We're about 1.5 degrees above pre industrial levels. Weather is a bad day, climate is the trend
- Average person in the UK is responsible for about 12.7 tonnes of CO2 a year
Scopes, in plain words
- Scope 1 is direct emissions. A diesel generator in the basement, a petrol fleet
- Scope 2 is emissions from energy you buy, like the electricity running the room
- Scope 3 is everything else. Teams, cloud, suppliers. Biggest slice for most organisations and someone else controls it
- Greenwashing mostly lives in scope 3. Net zero often means buying offsets rather than cutting anything
Digital's footprint
- Job is a bit like changing the engine of a car while driving it down the M1. We're asked to digitise more and reduce impact at the same time
- One email is tiny. The UK sends billions a day, plenty of them "thank you" replies
- Cloud has already caught up with the airline industry on emissions. Airlines get far more grief
- Most stored data is never opened again. The average organisation holds about 27 copies of its data
- Public records can need keeping for 25 years, but most data isn't a public record and nobody deletes it. Paper was expensive so it got managed. Digital is too easy. "A lot of this is process, not technology"
- The UN put e-waste at 62 million tonnes in 2022 and it hasn't got better
- The government design principles now have an 11th: minimise environmental impact
AI, briefly
- It generates patterns.
- Every question is worked out from scratch, not pulled from storage. So the same question from different people uses the energy again each time
- AI only knows your context if you give it. Copilot on a work laptop could know you work in - say - digital sustainability, a lot of tools don't, and that context isn't shared across an organisation. Don't assume AI knows anything
- Making one image uses about the energy of charging a phone
- Good information is hard to find. Claims swing between a data centre using as much water as flushing a toilet and draining rivers. Suppliers aren't necessarily on your side
- Trained people get a decent answer in 3 or 4 prompts. Untrained take 30 or 40
What digital sustainability is
- Using technology in a way that does less harm and more good
- Three buckets.
- How we use tech day to day
- How we design and build services
- And using digital to solve sustainability problems
- Co-benefits matter most. Leaders still assume sustainable means more cost and more time. Pitch it as helping what you're already doing, like cost or accessibility
- Most of a service's impact is set at design. Product managers have a say in how much data gets collected and kept
- Ask suppliers hard questions. The Government Digital Sustainability Alliance is a group of suppliers committed to being digitally sustainable. Ask what they can do for you in the thing you buy from them. If they do the work for you, your life gets easier
What to do
- Put sustainability in discovery, roadmaps and business cases, as a non functional requirement
- Work out what you can measure
- Get it into governance, so senior leaders talk about it at product sign off
- Talk to suppliers
- What you build now will probably still be running in 10 to 20 years
Q&A
- France and Germany are ahead. They act while most countries just measure
- Cloud isn't automatically greener. The biggest source in most organisations is monitors, print and network. One department found 160,000 monitors
- Supplier tender questions get brochure answers. Ask product level questions
2. Rachel (Capgemini): AI and sustainability
Framing
- Footprint runs the whole life cycle. Making chips and servers, training, using the tools, then e-waste at the end
- Treat it as an environmental challenge and a business one. The same drivers sit behind rising AI cost and rising impact
- Uncontrolled AI spend, people running out of tokens, prompts that need redoing. Fixing those cuts both cost and harm
- The fix isn't no AI. It's intentional AI
- Suppliers are getting asked about this by public sector customers now, mostly how they'll measure and investigate emissions from what they provide
Discussion
- Barriers: education, guardrails and people working on their own
- Without shared context, everyone prompts in their own way. Joined up context would be more sustainable
- Automation shows up cracks that were already there. Mostly a people, process and culture problem
- It's easy to reach for AI without a clear objective
- The barrier to entry is low, and sustainability rarely features in training
- Opportunities to bring sustainability in: from the start, in the business case, in policy and governance, alongside security and accessibility
Three layers of influence
- Technology infrastructure: power, data lifecycle, hosting
- Development: design principles early on
- Individual use: training and awareness
Putting it into practice
- Triage AI use cases on sustainability as part of the value case, not afterwards
- Actions differ by role, so work out what applies to yours
Case study
- A team looked at six levers in an AI system, including model size, and tested configurations for quality against cost
- The best option wasn't the cheapest. It was the best balance of quality and cost
- Sustainability doesn't mean always picking the lowest energy and water, it means deciding on purpose up front
Practical tips
- Ask if you need AI at all. A search might do
- Pick the right tool and model for the task
- Be clear about purpose and write a good prompt
- Know when to stop. Endless back and forth in one chat is waste
- Think twice before generating images and video
Q&A
- Measuring AI against doing it another way is hard, and provider numbers can be out by a long way
- We're good at measuring inputs and poor at measuring the value of outputs
- AI companies say AI will solve it, while putting gas turbines on site
3. Mark (Posetiv): are data centres suddenly the enemy?
Short version
- Not the enemy, but not helping themselves
- A year and a half ago nobody blocked data centres. Now planning is being blocked and politicians stay away from meetings
Why people have turned on them
- Energy bills. US studies say data centres aren't pushing bills up, but people in the US see their bills rise and blame them
- Water. The most emotive one, and mostly not based on facts. Though in some places, like desert towns in the US, they really have taken local water
- Jobs. Big promises that shrink. One UK site's job numbers shrank a long way between the council press release and the environmental statement
- Noise and pollution. Slough, and Loudoun County in Virginia where a data centre was built 70 feet from a home. One pollution calculation left out most of the site's generators
- Carbon. A site was sold as zero carbon by counting only the building shell and leaving out most of the impact
- Blocked by planners or regulators: Finland, Australia, a moratorium in a US county
Nobody has clean numbers
- The widely quoted per prompt energy figure came from one line in a blog post, with no method. It's now baked into estimation tools and could be way off
- Media makes it worse. An MP claimed water use of three major cities without saying which
- Operators say they use no water because of closed loop cooling. That cooling needs more energy, and generating energy uses water. It depends where you draw the boundary
- PUE (power usage effectiveness, total power over the power the computers use) is badly abused. It depends on how full the site is and the design, and planners get convinced by it
- Providers will rarely give water data. You can sometimes get energy data, but it's usually market based (counting renewable credits and power purchase agreements) rather than what came out of the plug. You need both
Accountability
- Of the sites he reviewed, almost none had any remedy if they missed their commitments
- He found no case anywhere of a data centre provider paying a serious financial penalty
- UK reporting rules exist but it's self declared spreadsheets, no verification, no consequences, and nothing on your share of the use
- Grid connection queues are long, with one developer told 2037. The delay is pushing operators to put gas turbines on site
- xAI's Colossus data centre in Memphis is running gas turbines it was told it couldn't have. Mark said emissions at a site like that are far higher than the same work would cause somewhere like Scotland
What works
- Heat reuse sounds easy and mostly isn't. A site promised heat for 7,000 homes but had no one nearby to take it, and the council had no budget to build the connection
- Deep Green puts GPU pods in swimming pools to heat them. A similar idea puts small servers on home hot water tanks, for work that isn't critical
- Blackpool Council wrote hard commitments into its data centre procurement: PUE, water, heat reuse, with financial penalties if missed. Every supplier accepted and none put their prices up. Now the council can show the truth when the press gets hold of it
- Without a contract term, operators won't hand over data. It's not in their interest
- A government working group is drafting a common set of supplier questions, answers picked from a list and scored automatically, so buyers don't need to be experts. Hoped for within a year
Cloud, budgets and process
- Public cloud is a really expensive way of consuming services if you don't change how you use it
- Biggest problem is budgeting. A bank had separate run and change budgets, no certainty of getting the next one, so it oversized everything up front and hoarded it. Nothing rewards handing it back. Their estate sat mostly unused
- What's needed is credit for returning resources, and sizing for the period you'll actually use
Sovereignty
- No hyperscaler (the very large cloud providers) offers a fully sovereign service. Layers of ownership, location, operation and security get in the way
- He scored operators against the EU sovereignty assessment. The big providers scored low on almost every band. That doesn't make them bad, it makes them not sovereign
- Banks are moving AI work back to their own facilities, partly for sovereignty and partly because they found it ridiculously expensive. One said they were paying a huge margin
- Alternatives: small models on your own GPUs, spare capacity on laptops and phones inside your boundary, community owned edge computing
- A participant from Northumberland said a site is going up on an old power station, where an aluminium smelter was closed years ago for using too much energy. The local MP is talking up jobs
4. Ned: embedding sustainability into your work
Ned is a service designer who spent the last couple of years as sustainable design lead on Defra's digital sustainability team. Defra coordinates digital sustainability strategy across government.
Where government is
- The roadmap for modern digital government came out in March this year and includes digital sustainability commitments
- The Technology Code of Practice gets a new sustainability point in the first week of December
The Defra approach
- Service assessments check teams against the 14 point service standard. Defra added a 15th, deliver a sustainable service. Assessed from 1 April this year
1. A library of actions
- Started as one slide, then 10 principles, a GitHub page, spreadsheets. Now going on gov.uk, hopefully next week, as digital sustainability actions
- For product people, the first theme: put sustainability in the business case, project planning and non functional requirements, and include climate risk in business models
2. An end to end process
- Before assessment: a sustainability statement, like an accessibility statement. Currently a Word template. Actions taken, actions for the next phase, trade offs, constraints
- Assessment gives red, amber or green
- Green is a fairly low bar at first. Recorded thinking and two or three things they're doing, even ones done for cost or accessibility reasons. Tightening over time
3. Measurement
- No whole service carbon figures expected. No tooling, little data, few people doing it
- Proxy measures instead. Data stored, data transferred, cost
- Suggests the Green Software Foundation's software carbon intensity, now an ISO standard. Carbon per unit of something useful, like per completed transaction or per user. Energy used times grid carbon intensity, plus the carbon of making the hardware, divided by that unit
What they've learned
- Teams mostly embraced it, with some pushback from very busy people. It lets them show things they were already doing
- Teams use the point to push back on short timelines. Rushing creates technical debt and rework
- One team realised vets and farmers use old devices and hadn't planned for them
Q&A
- Worry that an easy green means nothing changes. Ned said an early version risked so much pushback the point would have been removed
- Ambers in service assessments with a return in three months worked better than reds, which escalate to a director or minister
Panel and room
One practical action each
- Rachel: before using AI, stop for five seconds and ask what else could do the job
- Ned: look for the overlap between sustainability and what you're already asked to do. It isn't a default trade off
- Mark: product managers decide where people's time and energy goes. Early decisions like design principles and hosting are close to impossible to change once live. Training helps, like the Green Software Foundation practitioner course
Measurement and trust
- Directional data is enough. Plug your cloud bills into a tool that converts them to CO2, open source ones exist
- ESG targets (environmental, social and governance reporting) get set far out, like net zero by 2040. As dates approach, executives get nervous and the work fades
- Don't obsess over decimal places. Track at a high level and act
- Don't obsess over what you can't change. Take that decision to the person who can make it
AI arguments
- Mark's challenge: show him one business anywhere that has shifted revenue and profit at scale from AI
- Pricing is heavily subsidised and providers lose money per customer. What happens when the bill comes due?
- Investment is going into data centres and society is paying through graduate jobs and power
- Product people can quantify the cost of doing something and the cost of not doing it, including AI against a developer for a week
Coming up
- Unconference on digital sovereignty in Newcastle in February 2027, date to be confirmed
- More Product for the People events are planned next year, including Cardiff
Snippets of personal curiosity
Mostly a people and process problem
- Dark data nobody deletes. Oversized infrastructure nobody hands back. Budgets that reward hoarding. AI summarising AI with no one deciding anything
The numbers aren't good, so go directional
- Provider data is thin. Per prompt figures are close to made up. The press exaggerates the other way
- Get the rough size, make the likely better call, don't wait for the perfect number
Ask, then put it in the contract
- Suppliers give brochure answers unless you ask product level questions
- Write commitments into the contract, with penalties
Co-benefits get things through
- Cost, accessibility, speed, less rework. Came up in nearly every session
- Pitch it as a trade off and it stalls. Pitch it as another reason to do what you're already doing and it moves
Decide early, and put the decision where it belongs
- Design, hosting location and procurement are the cheap moments to change things
- Not your decision? Take it to whoever can make it
Intentional AI
- Do you need it? Then the right model, trained people, shared context
- "We need to be seen to use AI" with no goal behind it came up more than once
Low bar first, or greenwash?
- Defra started with an easy green so the point wouldn't be removed, then tightened
- Worry in the room: easy means nothing changes
Not the enemy, but trust has gone
- Promises about jobs, heat, water and carbon didn't survive planning. Local anger followed
- People want transparency
Curious next actions for myself
- Read Defra's digital sustainability actions once they're on gov.uk, hopefully next week. Which of the nine themes would a product lead actually own?
- Check what the new Technology Code of Practice sustainability point says when it lands in early December
- Try the sustainability statement idea on a service I work on. It's an accessibility statement with a different topic
- Add a sustainability line to my discovery and business case checklists, as a non functional requirement. See if anyone pushes back
- Work out what a functional unit would be for a service, like carbon per completed transaction
- Find out where a service I know is hosted, and look at how much the grid carbon intensity varies by place and time
- Ask a supplier product level questions next time. Ask how they measure it
- Look at how my own habits stack up. Old photos, spare devices, copies of files
- Try the five second pause before reaching for AI, and see how often a standard Google search does the job
- Look at the digital sovereignty unconference in Newcastle in February 2027, date to be confirmed
- Take the free Green Software Practitioner course from the Green Software Foundation. A few hours, self paced