Madmarch

In 2026 AI is disrupting how organisations design and deliver services and products. Here’s a quick selection of questions we can ask when we implement AI and data platforms. They’re based on my experience of helping government organisations make technology choices.

For more details, see the case studies Service design for Government AI and Reusing technology. 

I’ve used Sarah Drummond’s awesome Full stack service design (opens in a new tab) framework to structure these questions. 

I didn’t use any AI to write this.

Service

AI can affect your end users through:

• the front stage – for example, using chat bots or consuming AI-generated content; or
• the back stage – using AI to help triage applications

In each case you need to make sure you are protecting your users’ data as well as making sure humans are still making the most important decisions. At the front stage, your website visitors should be aware if they’re interacting with an agent, and still have access to a human being who can help them rather than a computer.

The back stage is the part of the service your users don’t always see. In government services, that’s often made up of civil servants making decisions on applications for funding. An AI could help with the decision making as it can sort applications and emails and assign them into different queues for humans to work on. 

At the same time, you need to treat decision making by LLMs with great care. If you design an algorithm to decide whether someone’s entitled to a government grant, it could exclude some people by mistake. Most government decisions, in my experience, are still made by real civil servants who can account for how they act day to day. 

Now let’s consider web design and online marketing. If you use AI to create content, branding and imagery, you need to check how your users feel about it, using research. There’s now a glut of instantly-recognisable AI pictures and design on the intranet. At best it looks generic, at worst it sends a signal that you’re using AI. Not everyone will be happy with AI imagery. They might ignore your service or assume it’s run by bots rather than real people.

You also need to consider the fact your users are now using AI as well. They may have better access to AI tools than a lot of developers, as they’re not constrained by organisational policies over which tools they’re allowed to use. They might use ChatGPT to find out about your services without visiting your website. Over time they might expect to be able to interact with you through chat rather than online forms, such as with the pilot GOV.UK Chat (opens in a new tab) service.

If users send in documents or selfies to prove their identity, they may already be using AI to filter or even generate their images. You need to use online content to tell them what’s acceptable. Your frontline staff need to be able to identify AI-generated content and reject it, if appropriate.

Questions to ask about your Service:

1. How will AI affect your users?
2. How do your users react to AI-generated content, like images?
3. How do they use AI themselves?
4. Should you offer new contact methods?

Infrastructure

Human beings leave a trail of data behind when we interact with a service, including:

• analytics – when we view a website or when we log in
• survey responses and social media activity
• documents and images – we upload documents to a cloud and may use our image for ID purposes
• personal data – including our names, addresses, bank details, email addresses and passwords
• financial data – records of financial transactions, as well as our bank details

You can use the trail of data to run a better service, such as finding where people abandon online journeys or struggle with password resets. On the other hand, your service needs to protect people’s personal data and minimise the risk of it being breached. 

AI tools run on good quality data

To run AI tools like LLMs, and to offer interactive dashboards, you need to have access to data which is reliable, accurate and up to date. For example, every local authority and council in the UK helps provide adult social care. Local authorities have their own staff and liaise with people who run care homes and care services in the community. It’s useful to track how many people are receiving care in a council home, versus people who are looked after at home. Analysts can look for patterns in the data, and decision makers can be better informed about what to spend time and money on. 

But for that to happen, you need reliable data. You can either rely on data which is created automatically in computer systems, such as people joining or leaving a care home. Or, you rely on people like care home managers manually updating your system. Both have pros and cons as an approach, and service teams need to decide what will work to keep the system up to date.

If you’ve worked on Big Data in a government setting you’ll understand that it takes time to make data usable. Datasets can be out of date or incomplete, or full of duplicates – for example, a person can be entered on a system multiple times. The data may need to pass between multiple organisations. Staff from different organisations may need to update it, which is the case for the Share Homes For Ukraine Data service – it’s shared between MHCLG, the Home Office and staff from Local Authorities across the UK.

Data scientists may need to move the data from one environment to another, or ‘ingest’ data to make it accessible through a user interface. They may need to consider building pipelines, through Databricks, as well as hosting services.

So, service designers need to understand complex systems and help organisations come up with a strategy for how to handle it. We don’t need to understand the fine details of the technology (for example, picking AWS versus Azure as a hosting platform) but we need to bring people together to co-design the best technical solution for that service. We’ll usually include people who work in:

• delivery teams – developers, business analysts, researchers and designers
• policy teams – civil servants or legal staff, who understand government policy and new initiatives
• operations teams – people who work directly with the general public, often by answering emails or taking phone calls

People enjoy being experts

Finally, to run infrastructure, you need staff who can understand and use the software you provide. Over the past couple of years we’ve assumed that people will become more comfortable interacting with tools like Copilot to complete routine work tasks like manipulating data. However, that’s a dangerous assumption.

Office workers are famously reluctant to give up using Microsoft Excel. They use it for everything from tracking work to analyse columns of data. Excel is relatively easy to pick up but it takes time to build up ‘ninja skills’ and learn advanced functions, like formulas, pivot tables and macros.  People enjoy learning skills over time. Interacting with an AI agent is a new skill for them.

You also need to ask whether it can replace their existing tools. For example, a lot of people want hands-on control over a spreadsheet rather than prompting an agent to create an output. If you’ve worked on offering an online data platform, you may have found that expert analysts and academic users want to extract the data themselves. They can then clean it and manipulate it with their own software, which could include Excel, R or Python.

Run research with your staff and users to find out their tasks, habits and ‘mental models’ – how they describe the work they do. This will give you a balanced view of AI adoption and understand how it might co-exist with their existing tools and knowledge.  

Some questions to consider:

• What activity can you track in your service?
• How is the data behind your service organised? 
• What AI tools will you bring in, to improve delivery while protecting people’s personal data?
• How do you train people to use AI tools, and will it replace what they already use?

Organisation

Your organisation needs to have an up to date digital strategy, which covers data management and AI tools.

The government mandates how organisations buy and use technology, using tools like:

1. The Treasury Green book, for procurement and spending rules
2. A Technology Code of Practice, for building sustainable infrastructure
3. The government Service Manual, for delivering services based on user needs

Those are the starting points for government organisations developing new services. AI tools need to work with the organisation’s tech choices.

Usually technical architects or lead developers pick the tech stack for their teams, while product owners have to understand the implications of those choices. Teams will often write business cases to fund new developments. They’ll also go through ‘governance’ to justify their decisions, including:

• a Technical Design Authority, to review their technical choices and make sure they’re consistent with the whole organisation’s strategy
• service assessments, based on the government Service Standard, to check the service is based on research with real people, and is usable and accessible
• a front door where teams can request help and advice on building services – DBT has an AI Lab which meets teams on a regular basis to advise them on AI tools

Service assessment can be carried out by people within the same organisation or by a central government body.

Finally, there’s the question of Big Tech corporations, including Microsoft, Amazon and Palantir, who now provide Big Data and AI services for governments. It can be expensive to buy licences for AI assistants or tokens. In 2026 people have also have concerns about private corporations owning public data, as well as American corporations managing data belonging to British citizens. And that’s not even mentioning the impact of data centres on the environment.

So, governance documents and internal bodies help organisations regulate their infrastructure. However, service designers still have challenges to resolve:

1. How do teams align with each other, reuse service patterns and use an effective tech stack?
2. AI tools aren’t free: can the organisation afford enough licences and tokens to run AI-powered services?
3. Are you working with safe, ethical companies to deliver your AI systems?

Intent

An organisation needs purpose. Lou Downe talks about this in some detail in her excellent new book, Bad Services (opens in a new tab). Intent describes what services an organisation provides, and why. 

When an organisation switches to big data and AI, they may need to revisit their purpose. In theory the intent and outcomes will stay the same, but the organisation might need to revisit their business model. 

There are new opportunities for leveraging people’s personal data. For example, if someone contacts their local council to update their address, the council could potentially make changes to other datasets they manage with that address change: council tax but also adult social care, or the address held by school for a child.

AI is built on data and has to follow an organisation’s rules. If it’s unsafe to allow data sharing in a service, you need to build guardrails around data in a system. For example, most people need to keep their address private, especially from ex-partners.

So, organisations need to keep the intent of their service in focus, and ‘do no harm’ when they allow AI agents or third parties access to private data. 

Questions to ask are:

1. What outcome are you aiming for?
2. How do you measure the impact of AI?
3. What guardrails do you need around data sharing and AI?

Culture

The last category is Culture. If you belong to an ethical organisation who cares about their staff, you need to consider the human impact of AI. At the time of writing government departments in the UK are under pressure to reduce their spending and headcount. At the same time, they have to offer safe, useful services to citizens. Creating anxious, unhappy staff puts those services at risk. Good staff could threaten to leave. De-motivated staff will be less engaged and productive.

Most organisations I work with offer AI training to upskill staff and roll out pilots for tools like Copilot. In this way they can encourage gradual adoption and discussion while creating minimal anxiety.

One final point about Culture. Designers have confided in me that they’re now under pressure to use prompts to deliver designs, rather than working on screens and user flows by hand. Generative AI can actually take them longer than working on deliverables in a way they were used to. They can also feel a loss of joy and satisfaction when they used prompts. 

This feels like madness to me. Designers solve problems for real people and they make things. Without offering them creativity and a craft to learn, you won’t get the best out of them. They will lose their sense of purpose and become an input to an impersonal algorithm. That’s what we want to avoid if we implement AI in an organisation. So, run research on your staff and your users to keep your culture human-focused.

Questions to ask about Culture:

1. How do you manage change at your organisation?
2. Do your staff actually want to use AI? Will it make them miserable?
3. How do you put people in control over AI, so that they adopt it at their own pace – if at all?

Conclusion

These questions are based on my experiences of working at large organisations over the past two years. I’ve found AI has huge potential for increased inefficiency, but it can only exist with a culture and an infrastructure that can use it effectively. 

If you hire a service designer they can:

• run research on your staff and users, to find out how they think and their level of tech maturity
• check your staff’s level of comfort with new technology, and what gives them a sense of purpose at their work
• run co-design workshops to come up with ideas and strategies for better services
• map your services and their pain points
• agree strategies and frameworks for improving your services, including designing operating models and writing business cases

Contact Madmarch to find out how we can work together, or just for a friendly chat about AI and data.