Samantha Maeer Interview on Why AI Adoption Fails When Businesses Ignore the People Using It
Samantha Maeer is the founder and creator of contAIn™, the methodology layer that directs AI to the outcome you define. She built it after discovering that having the tools was not the same as getting reliable results. Answers changed from day to day, context had to be rebuilt repeatedly, and when something went wrong, it was not always clear who was responsible.
Samantha Maeer, Founder & Creator of contAIn™
After thirty years delivering complex projects and behavioural change, how do people actually experience new technology compared with what the rollout plan expects?
I’ve worked on a number of sophisticated technology programmes where there was an amazing disconnect between the technology and the people it was designed to help.
Planning a schedule, hitting milestones and budgets is all good on paper, but the human cost doesn't always show up in the reporting. A project plan can’t tell you whether someone who's been doing the job for 15 years is worried about looking incompetent if they don't understand the new system straight away, losing control of their work or even becoming redundant. They may also be expected to adapt to a process redesigned by somebody who's never done their job and hasn't thought to ask how it works.
What interests me is that these people aren't necessarily resisting technology. They're dealing with the personal consequences of a decision made elsewhere, and those concerns rarely fit neatly into a project report.
Across countries, cultures and industries, I've learnt that getting the technology live and getting people to accept it are two different things. One of the most expensive mistakes a business can make is confusing implementation with successful change. You can meet every milestone and still have people quietly finding ways around the system you've just spent time and money introducing.
What gets missed when leadership, IT and the people doing the job are having three different conversations about the same change?
The thing is, nobody is necessarily disagreeing. They’re just approaching the conversation from entirely different viewpoints. Leadership is discussing the business case and cost, IT is discussing how the system will work, and the last in the queue are often the people who are actually going to use it, wondering how it will affect their jobs.
You also need to know who the gatekeepers are: Sharon from HR might have the spreadsheet that technically runs the company. In an SME, the boss might be the one with all the passwords on the back of a napkin or in their head. In a corporate environment, the executive assistants understand how decisions move through an organisation, who needs speaking to before a meeting and which relationships make things happen. If you don’t find out who those people are early on, you miss a massive part of the process and don’t give the project the best chance of success.
Ultimately, I don’t think this is a communication problem; it’s a listening problem. If you don’t listen to the problem, you can’t find or implement a suitable, tailored solution.
You enjoy learning from specialists in very different fields; how have curiosity and critical thinking shaped the way you approach problems?
An old boss of mine used to say, ‘If you are the smartest person in the room, you are in the wrong room.’ I love being around people who know far more than I do, plus it doesn’t hurt to have a good old debate occasionally. That’s how you learn. I’ve worked across multiple industries, including aviation, banking, hospitality and manufacturing, and the first thing I did was find the expert in each process to ask them how the pieces connect. Why do we do it that way? Who relies on this information? What happens when that system stops working? Sometimes the obvious question is the one everyone internally has stopped asking.
I’m also stubborn as a mule and perfectly capable of disappearing down a rabbit hole because something doesn’t add up. One of the great things about AI is it allows us to examine terms and conditions that we usually skip and make a more informed decision about whether to accept them.
Critical thinking is about reviewing multiple angles and asking the specialists difficult questions before you commit time, energy, data, money, people or reputation to the project.
Your Mythbusters series tests AI’s bigger promises; where does the hype get ahead of the evidence, and what might businesses lose by believing it?
My favourite claims are the ones suggesting that anyone can now do practically anything perfectly with AI, regardless of experience. It’s like buying a piano and expecting to turn into Elton John.
I'm fascinated by AI and use it extensively, but hype sells, and the less glamorous reality of getting it to do what you actually want rarely gets the same attention. I recently tried to produce a simple explainer video. The brief was agreed, the images approved and I knew exactly what I wanted. The AI apparently had other ideas and kept changing things I'd specifically told it to leave alone. I wasn't asking it to put someone on the moon. I just wanted it to stop redesigning something we'd already agreed.
It's precisely these sorts of claims I examine through Mythbusters. Producing something impressive doesn't necessarily mean it's fit for purpose, and if businesses are being persuaded that AI removes the need for experience, expertise and judgement, they may find themselves spending more time checking and correcting the work than they expected to save.
Your two published papers examine failures in AI-generated governance; what first caught your attention, and what did your investigations reveal?
The em dash was such an obvious AI tell, and it was driving me to distraction. I kept telling AI not to use them, but they kept reappearing, and eventually I went down a rabbit hole to find out why.
What I discovered was rather ironic. Some of the documents I’d created to govern the AI had themselves been written with AI and were full of the very patterns I wanted to eliminate. My first paper, The Feed Loop, examined how those documents could keep reinforcing the behaviour they were supposed to prevent.[1] I was adding more instructions when part of the problem was the material the system was reading in the first place.
The second paper, Gates and Prohibitions, went further, examining why telling a system what not to do was less reliable than establishing checks its work had to pass.[2]
So AI wasn’t being moody after all. It was responding to patterns I’d unwittingly been feeding it. The answer wasn’t another instruction; it was fixing the underlying problem and establishing controls that actually worked. All that because I didn’t want a pesky em dash!
We’ve been sharing data with technology for years; what should people and businesses ask about where their information goes and how it’s used?
We’ve had smartphones, apps and voice assistants in our homes for ages. AI didn’t invent the appetite for collecting data, but connecting a powerful AI tool to your inbox, calendar or business documents without understanding the terms is a different level of trust. I worry that adoption is getting ahead of people’s understanding of what they’ve agreed to.
I’d want to know where that information goes, who can access it, how long it’s kept and whether it’s shared or used to train AI models. Can I delete it or take it elsewhere? Those questions are better asked before connecting anything, not after something goes wrong.
A business may also be uploading information about customers, staff or clients, not merely its own material. That raises questions about confidentiality, accountability and UK data-protection obligations. Different providers have different terms, so blanket assumptions aren’t much use.
I’ve responded to government consultations on information and technology policy because I’d rather participate than simply complain about the outcomes. My concern isn’t that people use technology; it’s whether they understand what they are handing over.
As technology becomes capable of making recommendations and taking actions on our behalf, how should we decide what to delegate and what responsibility to keep?
I start with what happens if it gets something wrong. If AI suggests something I can check and reverse, I'm happy to let it do the donkey work. If it can act on something affecting a person's livelihood, money, health or rights, I want to know what information it's using, who authorised it and how anyone can intervene.
That includes the assumptions it builds about us. I discovered that Claude had labelled me INFJ despite my never taking the test or asking it to profile me. It had inferred this from our conversations and stored it as part of its picture of me. I spotted it, but couldn't change what it had recorded. Whether the label happened to be accurate is beside the point. An inference had been recorded as though it were an established fact.
Now imagine those assumptions influencing a recommendation or action with real consequences. Before delegating, we need to understand what information the system relies on, whether we can challenge it and how to stop or reverse a decision. Otherwise, we risk discovering afterwards that it acted on assumptions we never agreed to.
In plain English, what problem does contAIn™ solve, who is it for, and why isn’t buying another AI tool necessarily the answer?
contAIn™ gives individuals and organisations a straightforward way to work out what they need AI to do, identify the problems getting in the way and establish the foundations for more consistent results. It’s a repeatable, question-led methodology, not software. You can work through it yourself, using the letters in CONTAIN to remember and apply the principles, from one task to processes across an entire organisation.
A lot of it is the unsexy groundwork: deciding which information is authoritative, where to find it, how to name and organise it, who keeps it current and how to carry that context between tools. Without those foundations, people can end up repeating instructions, correcting the same mistakes and buying subscriptions before understanding what needs fixing.
I’m sceptical of the fashionable ‘second brain’ label. If you don't know which information to trust, giving it a clever name doesn't make it any more dependable.
For businesses, CRASH is our diagnostic-led approach to identifying duplicated effort, inconsistent results and unmanaged risk. Think of it as a CRASH course in understanding the problem before paying for another solution. Sometimes the answer is better use of what you already have.
If a business owner rang you tomorrow and said, “Our tech’s a mess and everyone says we need AI”, what would you ask before recommending anything?
My first question would be, ‘Who told you that you needed AI?’ Not to be awkward, although I do enjoy asking awkward questions, but because ‘everyone says so’ isn't much of a business case.
I'd want to understand what's actually going wrong. Where are people wasting time, what's being duplicated, which processes aren't working and what are they trying to achieve? Then I'd talk to the people doing the work, not just those making the purchasing decisions. They're often the ones who know where the real problems are, including why Sharon from HR has a spreadsheet that appears to be holding the entire business together.
I'd also want to know what technology they're already paying for, how it's being used and whether anyone understands where the information lives. Sometimes AI might help, but the answer could be a clearer process, better organisation or simply making proper use of what they've already got.
After thirty years working with people and technology, I've learnt that understanding the problem is worth far more than chasing the latest sparkly thing masquerading as a solution. My advice is always to look at where you are and what you have before buying anything new.
If that sounds like the kind of conversation your business needs, email me at sam@contain.digital.
Read more from Samantha Maeer
References:










