The Shadow Workforce: Managing the AI Your Employees Are Already Using with Alison Wright, Microsoft UK

John Emmerson: Hello and welcome. I’m joined today by Ali Wright from Microsoft UK. Thank you so much for joining us today, Ali.

Alison Wright: Hi John, nice to see you.

John Emmerson: You are the SMB Director at Microsoft UK, which puts you in a position where you’re talking to small and medium-sized businesses right across the country about AI — probably more than most other people in the country. From those conversations, what’s the vibe at the moment? Where are most of those businesses at right now when it comes to AI adoption?

Alison Wright: It varies hugely. Some are curious and in an experimental phase, but others have completely transformed how their business runs based on AI and are thinking about an AI-first approach. What we typically see is most businesses going through a journey. They start with AI as an assistant — whether that’s writing emails, transcribing calls, that type of thing. Then, as they get more confident, they bring in agents. Often pre-built agents that can do research for them, facilitate a meeting, and really support humans doing the work in the flow.

And then finally, as their competence grows and the technology changes rapidly, we’re starting to see what we call autonomous agents — where agents take on end-to-end business processes with human oversight and sign-off. It could be onboarding new team members, managing a supply chain process. So typically we see businesses going through this journey, and they could be at any one phase at any one time. But most of them are excited and are starting that journey.

John Emmerson: Do you have any data on the levels of adoption across those different phases?

Alison Wright: Yes, there’s been some research — and Microsoft has done research in this area as well. It varies across the size of organisations. Where businesses have IT departments that take a lead on AI adoption, we see it in the 20-plus per cent mark. In smaller businesses, it’s around the 10 to 20 per cent range, but that’s changing rapidly.

What we’re seeing more and more is that there’s a real appetite and desire to get involved with AI. Workers are bringing AI into their organisations, particularly where they don’t have a prescribed tool or guidance from their employers. We’re seeing around 71 per cent of UK workers saying they’re bringing consumer-grade AI into the workplace. And 51 per cent continue to use it weekly. The appetite and the adoption is there. But the biggest gap is where organisations think they can bring AI in and deliver value without any behavioural changes. It’s not necessarily about the technology or the platform you choose. It’s much more about the human and the change management — how people and processes and technology come together to drive value. That’s the biggest thing I’d recommend organisations think about.

John Emmerson: This comes up in every single conversation we have. It’s always about the human piece — the skills, the knowledge, the culture. That really seems to be the key.

Alison Wright: Definitely. AI literacy from a skilling perspective is super important — learning to work with the tools. But then there are also those human skills, the soft skills that we don’t always focus on, that are being elevated by AI. Critical thinking, bias detection, emotional intelligence, judgement. If you’re creating new content with generative AI, you’ve got to have some view on whether that output is good, bad or indifferent. Those human skills are even more important in the age of AI, and that’s an area I’d recommend any business think about alongside their AI literacy.

John Emmerson: I find it fascinating how many people will use an LLM to produce a piece of work and then just send it without any editing or checking. You wouldn’t do that if a human gave you a piece of work — you’d check it. Why treat AI any differently?

Alison Wright: Yes — it’s about getting you to the first draft, helping you avoid blank page syndrome. Think of it as a helpful internet. We all know AI can make mistakes. So how do you choose AI tools that give you referenceability, so you can go and double-check the sources? How do you read through the output to make sure it matches your business tone and the brand you want to portray? And from a governance perspective, every business needs to think about where humans need to be the final sign-off and final arbiter. Certainly if content or systems are going to engage with customers, your supply chain, finance, or your people — humans need to be the final decision-makers. That’s why Microsoft uses the term co-pilot. We’re not taking humans out of the loop. It’s about augmenting human ingenuity, capability and creativity.

John Emmerson: Do you think there is a gap between what SMB leaders think AI will do for them and what it actually does?

Alison Wright: Yes. It comes back to the point that AI can only create value when you’re really clear on how you’re going to use it. Often businesses say they want AI without thinking through the steps before that — what problems are they trying to solve?

We always try to bring customers back to a framework with four pillars. First: how are you going to engage your employees better? How do you give people back the joy of work by taking away the things they don’t enjoy in their day-to-day? Second: how can you engage and delight your customers through increased, personalised services — whether through chatbots or other tools that speed up their experience? Third: business process re-engineering — where have you got processes that are duplicative, time-consuming, and where could you take cost and time out by redesigning? And fourth: how can you innovate faster? Maybe bringing new products to market more quickly, or entering new markets through translation and multilingual capability.

Those four categories — your employees, your customers, your processes, your innovation — bring everything back to the fundamental question: what am I trying to achieve with AI, rather than treating AI as the output in itself.

John Emmerson: The language point is something I keep hearing. I was chatting to an SMB owner last week who switched on the ability to talk to customers in any language using AI. Their NPS scores skyrocketed and they started getting more international customers — just from being able to talk to people in their native tongue.

Alison Wright: Exactly. We had a customer — 50 employees, geographically spread across three continents. They were recording calls, transcribing them, sending them for translation, then distributing the notes. Microsoft Translator within Teams has helped them save significant time and money, and be much more collaborative. That’s allowing them to enter new markets for the first time. Think about the challenges first and AI will help you there.

John Emmerson: You mentioned shadow AI — employees using tools that haven’t been approved by the organisation. Do you see this as a problem for SMBs? What should business owners do about it?

Alison Wright: Around 71 per cent of UK employees are using AI tools their employer has not approved. And research has shown that nearly one in four are putting finance-related data into those tools, which is worrying. Employees do this mostly because it’s what they’re familiar with from their personal lives, and often because there’s no approved option or guidance from their employer.

The steps I’d recommend: first, assume it’s already happening. We’ve worked with customers where, once we gave them visibility into what was going on in their organisation, they found multiple unapproved tools they didn’t know about. So assume it’s happening — and it’s not about banning usage or telling employees off. Getting visibility is the first step. Then choose a safe alternative, like enterprise AI, that gives employees the functionality they want wrapped in the privacy and security that organisations need. And then set a simple policy. It doesn’t have to be long. I sat on the board of a 50-person charity and we wrote a document that was a page and a half: these are the tools we’ve chosen, this is when we can use them, this is when humans need to sign off, these are the cases where you definitely can’t use it. Clarity and working with your employees is the best guidance.

John Emmerson: Very good advice. You’ve been running what you call agentathons — where business owners prototype an AI agent in a day. How are those going and what’s happening after the sessions?

Alison Wright: They’re very new and we’ve been piloting them. They’ve been hugely popular — oversubscribed, which shows the demand is there. What we wanted to achieve was to take customers from “I’m interested but AI use cases are still quite abstract for me” to a concrete use case and outcome they can walk away with. We deliver them with our partner channel and teach attendees using simple no-code, low-code tools. They don’t have to be technologists. With LLMs, we can all work in natural language.

We work with business owners on their core business challenges. The types of use cases that come up: a chatbot on their website to guide customers through purchasing, automating how they triage customer enquiries, automating responses to tenders and bids to make them more consistent, sales follow-up and qualification, finance tasks like invoicing and categorisation. Real business problems, real prototyping. And then the next step is: I’ve learned how to do this for one use case — I can probably go away and replicate that process with another idea we have. The goal is that they become confident and proficient enough to do it themselves.

John Emmerson: Has anything unexpected come out of those sessions?

Alison Wright: The interesting thing is that people don’t actually have flashy ideas. It’s typically the boring things. Everyone has a bit of their role that they procrastinate about — the work that drains energy. It’s typically there that AI delivers value fastest, more than the big flashy ideas.

A few examples: we’ve done chatbots on websites that help guide customers through purchasing. We helped a small family distillery help customers choose the right gin for a gift. Five employees. And what was lovely about it was it didn’t just recommend products — it got involved in cocktail recipes and cross-sold into distillery tours. A family-run business of five people, still taking advantage of the power of agents to augment the customer experience online.

John Emmerson: Outside of structured events like that, business owners are being pitched AI solutions every few seconds. How should they decide what’s worth paying attention to?

Alison Wright: Come back to those four pillars before you go out to market looking for tools. Start with the problem or challenge you’re trying to solve. And think about what we call the blank sheet question: if you had a blank piece of paper, would you design your organisation the same way if you were starting from scratch today with AI? Probably not.

From a sales process perspective at Microsoft, we augmented our sales team with digital sellers. Instead of our human salespeople spending time on prospecting that could be a goldmine or a waste of time, we use digital salespeople to do that initial prospecting, prioritisation, outreach and engagement. As soon as a customer starts to engage, we hand it over to a human. It’s thinking through: where could I deliver real change for my organisation if I just took off the shackles of legacy thinking? Sometimes the best way to do that is to give it as a side project to a small team.

John Emmerson: I’ve spoken to organisations that have gone down that route — completely ripped up their organisational model and asked what it would look like if they were starting again. In every case, what I hear is they’re not looking to remove people. They want to make the organisation more efficient and use the people they have to create the growth and scale they want, without needing to hire more heads. Then on the other side, there are leaders who say they have no idea where to start, they’ve got legacy data and legacy systems, they’re tied into contracts.

Alison Wright: Yes, and if you don’t know where to start — particularly if you’re not a technology company — then find a great partner who can help you on that journey. Show you what’s possible and take you through it in a way that feels comfortable. Sorting out your data, getting it into the right state, making sure it’s protected and secure, and then taking you forward with AI from there.

John Emmerson: Are there any misconceptions about AI that you hear from the SMB audience?

Alison Wright: It’s less misconceptions and more that many people don’t understand what’s actually possible — they need help to think it through. They understand what AI can do in their personal lives, but applying it to their specific business and industry is another step. That’s why we’re trying to use agentic technology to provide one-to-one coaching at scale.

On a recent roadshow, we were talking with a range of businesses and one said: I’m a pizza company in Liverpool — how would I use it? They want specific guidance for their use case. None of us can scale to talk to every single small business in the UK. So what we’ve created is an agent coaching experience where a business owner can go through a one-to-one conversation about generative AI for their specific business. If they say: I want to sell more pizza to students in the Liverpool city region — the agent takes them through building a social media campaign, walks through the prompts with them, and they leave with an action plan. That’s how we’re getting to scale.

John Emmerson: Final question. What’s your big prediction for what will change for SMBs in the UK in the next 12 months?

Alison Wright: I think it’s the agentic AI piece. We’re familiar with AI as an assistant. But it’s moving towards agents that can completely own a business process with human oversight. If I’m a retailer selling home décor and I’m out of stock of white paint — the whole system of noticing that, placing the order, the whole process, can be automated and taken care of by agents. With appropriate thresholds, so if the order gets over a certain value I want to sign off before it goes through. But it’s the mindset shift of giving work off to agents and the fact that they can completely run these processes.

And then you start thinking: I could have a digital employee working alongside my human employee. How do you manage digital employees? How do you govern them? How do you manage their lifecycle? People entering the workforce today — you may not have given them a people manager job yet, but you could give them an agent manager job easily. I think within 12 months, businesses will be thinking about a world where humans and agents work alongside each other as team members and co-collaborators to create value for customers.

What is shadow AI and how should SMBs respond to it?

Wright’s data point is stark: around 71 per cent of UK employees are using AI tools their employer has not approved, and nearly one in four are putting financial data into those tools. Her recommended response has three steps. First, assume it is already happening — organisations that have looked have consistently found unapproved tools they did not know about. Second, replace rather than ban: provide a sanctioned enterprise alternative that meets the same need, wrapped in appropriate security and privacy controls. Third, write a simple one-page policy covering approved tools, acceptable use cases, and when human sign-off is required. It does not need to be a formal governance programme to be effective.

What are the four pillars Microsoft uses to help SMBs identify where AI creates real value?

Before reaching for any tool or vendor, Wright recommends SMBs work through four questions. How can AI improve the experience of your employees — giving them back time and removing the work they find draining? How can it improve how you engage and delight your customers through more personalised, responsive services? Where can you take cost and time out of your business processes by redesigning how work gets done? And where could AI help you innovate faster or enter new markets — through translation, accelerated research, or new product development? These four pillars bring the conversation back to the problem being solved, rather than the technology being adopted.

What is an agentathon and what do SMBs learn from attending one?

An agentathon is a Microsoft-run session where business owners prototype a working AI agent in a single day, using no-code and low-code tools, guided by Microsoft partner organisations. The goal is to take participants from ‘I’m interested but AI use cases are abstract for me’ to a concrete, functioning output they can replicate independently. The consistent finding from these sessions is that the use cases that deliver value fastest are not the ambitious ones. They are the repetitive, time-consuming, low-energy tasks that employees procrastinate on. Once participants have built one agent, they typically leave with the confidence to apply the same process to other business problems.

What is the blank sheet question and how should mid-market leaders use it?

Wright’s blank sheet question is this: if you had a blank piece of paper today, would you design your organisation the same way you did before AI existed? For most businesses, the honest answer is no. The question is useful because it separates the conversation from legacy constraints and encourages leaders to think about where genuine redesign is possible, rather than incremental improvement on existing processes. Microsoft applied this to its own sales function, replacing human prospecting with digital sellers that handle initial outreach and qualification before handing over to humans once a customer engages. The constraint most businesses hit is not imagination — it is the difficulty of acting on the answer.

Why do the highest-value AI use cases tend to be the boring ones?

Wright’s observation from running agentathons across the UK is that when business owners are given open space to identify AI use cases, they rarely come up with ambitious, strategic ideas. They identify the tasks they procrastinate on — the repetitive, time-consuming, low-energy work that drains their day. These turn out to be the use cases where AI delivers measurable value fastest, because they are well-defined, frequent, and the improvement is immediately visible. The distillery she describes — five employees, a chatbot that recommends gin gifts, cross-sells cocktail recipes and tours — did not set out to build a sophisticated AI product. They removed a friction point and the results followed.

What does the shift to agentic AI mean for SMB organisational structures?

Wright’s 12-month prediction is that SMBs will move from AI as a productivity tool to AI as a co-worker. Autonomous agents running end-to-end business processes with human sign-off at defined thresholds are already available today. The organisational implication is significant: businesses will need to think about how they manage, govern and define the lifecycle of digital employees alongside human ones. Wright notes that people entering the workforce now could be given agent manager responsibilities before they are given people manager responsibilities. This reframes what early-career development looks like and what management means in an AI-native organisation.

How should SMBs handle the volume of AI vendor pitches they receive?

Wright’s advice is to start with the problem, not the vendor. Before engaging with any tool or supplier, return to the four pillars — employees, customers, processes, innovation — and get clear on what you are actually trying to achieve. The volume of AI products on the market is significant and growing daily. Without a defined problem to test solutions against, it is easy to be drawn towards tools that demo well but do not map to a real business need. She also recommends finding a trusted implementation partner for businesses that are not technology companies — someone who can help assess readiness, address data governance, and guide the adoption process in a way that feels manageable.