How to identify and approach repetitive tasks with AI
To view this video, please accept marketing cookies.
Posted: Tue 17th Nov 2026
Most small businesses know AI matters, but it is hard to know where to start and how to approach it in an effective way that will drive genuine results for your work.
Business owners estimate they spend an average of 11 hours a week on administrative or finance-related tasks. (American Express SME Business Barometer, 2026)
In partnership with Enterprise Nation, Kiran Jayaram from Maikai delivers another AI session to help demystify AI, understand where to start in your organisation and illustrate how to automate common processes.
Topics covered in this session
How to avoid the noise around AI. Get going without feeling overwhelmed.
Understand where to implement AI within your day-to-day. Find the jobs that eat your time and tackle them in simple stages.
A live automation. A demonstration on how to turn a spreadsheet into a financial report, with the checks and the human decisions shown.
Maikai has funding for small businesses to support with AI upskilling. You will find out how you can access this.
About the speaker
Kiran is Maikai's education lead, helping drive impact in small businesses and schools by upskilling staff and ensuring training opportunities deliver maximum value.
He loves sharing the transformations AI can bring to individual roles through small, practical tweaks.
Watch more expert webinars
Access a growing collection of expert-led webinars covering marketing, sales, finance, growth and more – ready whenever you are.
Transcript
Lightly edited for clarity.
Beth: Hello, everyone, and welcome to today's Lunch and Learn. My name is Beth, and I'll be your host today. For those of you attending a Lunch and Learn for the first time, Enterprise Nation is a vibrant community platform for startups and small businesses.
I'm very pleased to introduce Kiran Jayaram, who is MyKai's education lead. In the session, Kiran will demystify AI, discuss where to start in your organisation and illustrate how to automate common processes.
As always, if you have any questions as we go along, please do pop them in the chat or in the Q&A, and Kiran will do his best to answer them at the end of the session. Today's webinar is being recorded, and we're going to share the recording along with a few further resources later in the email. Over to you, Kiran.
Kiran Jayaram: Thank you very much, Beth. It's lovely to have you all on, and thank you very much for taking the time. This session is to go over where to get started and how to see through the noise when it comes to AI, looking at it through the lens of your business and the work that you do.
At the end, we'll go through a live demonstration of automating a spreadsheet into a financial report – I know a lot of you have asked how you can do that. You'll also be able to take away the complete folder showing exactly how I run this process.
So you'll all leave with a template on how to take a finance export and turn it into whatever report you're looking for. It should be something you can go and play with and work out how to apply to your own business.
So, how to approach AI and repetitive tasks. The first piece is how and where to start. We'll go really quickly through how to think about AI and how to make sure you're not getting distracted by, quite frankly, all of the crazy media stuff that's going on at the moment.
We'll then talk about two lenses to look at every piece of work with AI through, and where to start with your business. It's really important initially to ask where you can begin to cut a little bit of manual workload to free up some of your time. A lot of us are running around manically with so many things to do.
If we can carve away at some of that workload, it then allows us to move on to the second layer: how do we make things bigger, better and faster, and really add leverage to your business by making your service much, much better? Then, for the third piece, as I said, we'll go through that working example, and you'll be able to leave with a template on how to do it.
As a starting point, here's a lovely word dump. I actually asked Claude to go through all of the crazy buzzwords that have come through articles, particularly across UK agencies, over literally the last three weeks. This is just to say that there's so much noise going on around AI.
There are a lot of reasons for this, but fundamentally, there's a way of thinking we want to reverse. It goes something like this: there's all this stuff going on with AI in the world, people are telling me I should be using it, I'm feeling like I'm falling behind – what tools should I be using?
All of those questions are not relevant to businesses like us and like yourselves in the first instance. The most important thing we want to avoid is starting with all of this noise and asking which AI you should be using and what you should use it for.
The most important place to start is: what are the priorities for your business? What is the service you provide? You want to address it by asking where you can apply AI in your day-to-day in ways that are actually going to drive value for you.
Just as an example of some of the noise, this is just the last three weeks. OpenAI has cut its funding round target by US$10 billion, which is no small feat. Anthropic is talking about what would, quite frankly, be the largest IPO there's ever been, by a long way.
We've also had some issues with the supply chain into AI companies with NVIDIA. This is just the last three weeks, and it's not going to slow down – all of the noise in the media is going to carry on.
But the most important thing is that, quite frankly, none of this really matters to you, not in the short term. And it's actually unlikely to in the medium term, as in the next couple of years. It's not our race to run.
All these tech companies are driving and investing towards a generative model that can do all your tasks. That's brilliant if we can get there, but it's a marketplace race. They know that if they get their tools into as many businesses and organisations as possible, then as we become a little more dependent on AI moving forward, theirs will be the ones we use.
We saw the same with Microsoft across the public sector and schools – they own that marketplace in many ways. That is the race, and it's not our race to run.
If AI development stopped right now, we would have everything we need to completely rethink the way we run our business and the services we can provide. We could also completely reimagine the amount of work that can be done by one person in your team. So it's about starting with the business first, and starting with the work that you do first.
It's really important to understand where your priorities are, but also what's limiting you. So we'll go on to the horizons – the two layers you really need to focus on. There are some key questions here that are really important to ask.
The first one should always be: where does my time go unnecessarily? What am I spending my time doing that isn't actually driving business outcomes? There's an initial question to ask there about what you should be doing anyway.
But if these are important but not urgent tasks that are incredibly manual, which goes to the second question, they're brilliant opportunities. You can start to really reduce the amount of time you're spending on those activities, jump yourself up a level to be more strategic and put better processes in place.
Then layer two, as we said, is to focus on the lens of reimagining how your organisation runs. It all has to be anchored in the service you're running and how you can really improve that for the customers you're serving. It's not about adding AI for AI's sake – it's about making the delivery better.
As long as you come at it from that lens, you'll really be able to start breaking down where AI fits and where it doesn't. A really important note is on ethics, which is a super important piece. There's some brilliant work out there on the ethics of AI and computing.
This is something I genuinely believe in, and it's probably the best framework out there for AI and ethics. There are articles you can find from Luciano Floridi, who sets out five principles of AI ethics. Very simply, this is something you can build into your processes.
You can take the articles Floridi has written and apply them to the context of your processes, and the AI will operate within those. As a really quick summary, we've got beneficence: it has to do good. Is this a good thing for people, is it genuinely helping, and is it improving your work and the service you're providing?
It's also got to do no harm. Lots of these are common sense, and you'll be able to feel it when you're looking at your processes, but non-maleficence is all about doing no harm.
Autonomy is something that's really important: you need to understand where you fit in with this and how you retain your agency within the process. Where are you making the key decisions, and where is AI offering its input?
Justice, a lot of this is around accessibility. Who retains the cost? If you're producing products that are AI-heavy or you're incorporating AI, is it fair for everyone, and are you considering that when you incorporate it?
The same goes internally with your teams. Then there's explicability, which is the key difference. The four previous principles actually came from bioethics research, which is really interesting.
Explicability really simply means: can you explain exactly what happens at every part of the process where AI is used? And can you explain why you got to that particular result? That's the most important one, and it's the key difference from ethics in the past.
So how do we get into the practical layer? You might be thinking: I feel like I'm starting to understand ethics, I know my business priorities and, to some extent, where I think AI should be applied. You'll end up with a list of workflows after asking some of those previous questions.
Stage one is to get a list of all of the manual workflows across your organisation – things you do every day, monthly processes, annual processes – and write them out. Some examples are here: chasing invoices, templates you apply to enquiries, or how you run your bookings and your stock. These are all manual tasks that can really be automated with AI, and this is where the real operational gain sits.
Today, we're going to go through turning a financial spreadsheet into a report. It's really important that the framework I'll show you today can be applied to any workflow, and you'll be able to see that once you've opened the template.
You've now got a list of workflows, and you're trying to work out how AI works within them. Using our example today, we've got a bank export that we're looking to turn into a financial report you can share across your team. We have that workflow defined.
So the first thing is to write out your workflow and describe what it is. We'll then break the workflow into stages, and break those into the tasks that sit within each stage. You'll then start to see which tasks are human tasks and which are AI tasks.
This slide illustrates how I've broken up the financial reporting workflow into the key stages and the tasks required. What's really important about this is that it allows you to define the process. When we go back to explicability from our ethics perspective, you can tell whoever you like exactly what is happening at each moment of the process.
This is what informs your prompts, and it's what informs the way you store this process moving forward. Really importantly, as you build these processes out across your business and new staff come in, you don't need to spend lots of time training them. It's all defined right here, and they can go through it as they need, improve it and take it away.
It allows you to really break up each element. If you need to make edits, you're not relying on one massive prompt that you're then trying to edit, wondering where to make changes to get the better output you're looking for. You've already got it broken up naturally into stages, so you just tweak as you go.
This is super important, and the most important thing to understand from today is that this is your stage one: break it apart into the stages and the key tasks. At the bottom, it then gives you the check you want done after each stage.
This then manifests itself in the next practical layer, which is, quite frankly, files and folders. I'll show you this in a minute, and I'll also share this folder with you so you can try it yourself. It gives you an example of how we break these stages and tasks into a folder that holds all of this context.
What do I mean by context? Firstly, it just means asking really high-level questions: what is this process for, who are you and who is your business, and what are your ethics and your company policy? We'll have all of that within a file.
In this case, the file has got ".md" at the end of it, but don't worry about that – it's really simple. It's called a markdown file, and it's much easier for AI to read than, for example, a Word document.
I'll show you what one of those looks like in a second, but it's just text, almost like a text file. You'll have a context file that describes what the process is for, who you are and what your company's ethics are. Then you'll have a process file, which is almost like your instructions: it describes what's going to happen in this workflow, the steps and where the human checks are.
What's really important about these two files is that they act as a map through the entire process, describing to your AI what it needs to do and where it needs to go. Imagine having a library full of books, with all your workflows in it. The context and process files act as a map, and your AI is like a librarian.
You say, I would like this particular process, which sits in these folders, to be delivered, and your AI will be able to go and find it by following these processes. Really simply, all I've done is break up our three stages – prepare, compare and report – into their own folders within the slightly larger folder for the workflow. It just allows me to break up the instructions.
I'm going to show you what that looks like in a second. I'll be doing this in Claude, but you'll be able to see the files and folders. For the workflow we're going through today, I've made up a company, Bramble and Co Bakery – I think it's a really thriving business.
For Bramble and Co, we've got a bank export of the financials for the month. We're going to go through how I've broken those stages out to get me a financial report, which is going to look like this. I've just skipped a few slides, but don't worry about those – that's going to be our output at the end.
What I want you to be able to do is look at the template I'm giving you and apply it to your finances and your process. If we skip back a couple of slides, there are a few different ways you can automate this process.
The first thing I'm going to illustrate today is how you would run this with your files and folders through a basic chat. We recommend that everyone has some form of paid AI licence, because it's much more secure from a data perspective. Using the chat option does mean you can do this in a free tool, but obviously be very careful about putting financial data in there – it's not something you really want to be giving to an open-source AI model.
The second layer: in Claude, we have projects, and in Copilot, you have notebooks. You've got Gems within Google, and I can't remember what OpenAI's one is called. But everyone has a version of what is, in Claude's case, a project, where the same repeatable context stays within one particular chat.
Whenever you input something into that chat, it reads all of the context each time and then executes what you need it to. I'll show you what that looks like in Claude as well.
The third layer is the wider automation. I'm not going to go through that today because it takes a bit more time to set up, but I'm very happy to answer questions on it. In terms of having your AI act as an agent moving forward, this folder gives you the exact framework to automate the process even further.
What's the difference? You might want your financial report delivered automatically on a particular day every month, and that's the level of automation you'd get into with layer three. So I'm going to jump out – this is the output we're going for, and I'm going to show you the setup.
Bear with me a second. The first thing I'm going to show everyone is this finance report folder. As you can see, the whole process I've described sits within it.
If we open it up, we can see our prepare, compare and report folders. Within this finance folder is the entire workflow: our three stages, with the tasks that sit within each stage and how they're delivered. All of those tasks sit within an instructions file, which you can see here.
This is what I described as a markdown file, so what does that look like? Don't worry about this application – it's not going to be relevant to you, but it allows me to illustrate what the instructions look like. These are actually the instructions for the writing part of the report in stage three.
You can see how I've set this out. It's really simple: it goes through what these instructions cover, the purpose, which files should be read during this process and, ultimately, the steps that are going to be done. You've then got the ethics and the human checks in here, so all of that will be accounted for.
What's really cool is that you can ask your AI to help you build this out. I didn't write all of this out myself letter by letter – I built it with AI. Within each of my folders, I've got an instructions file describing what we're doing in that particular stage.
We've also got the ledger, which is our messy spreadsheet, so to speak – the bank output we're going to be analysing today. You'll notice a couple of things around the categories and corrections here, and you'll ask, Kiran, what is that, and why is it in there?
When I was actually doing this for our finances, I found you might get a report through that requires corrections, or data that needs to be cleaned into some format. So I have a corrections bank. When there's a duplicate, for example, it goes into the corrections file and gets stored there.
I'm going to pause for a second, but what I'm going to do now is pull up Claude, and we'll go through what this looks like. It's very simple. First, I'm going to open a new chat here.
Like all of the different interfaces, you'll have a chat bar here. You might ask: what's the simplest way I can use AI to help me with this process? You could go into your first folder, take your ledger, the instructions and the categories – remember, the ledger is just the spreadsheet of data I'm analysing – and copy that whole group of files straight into your chat.
Because of the instructions I've put in there, it will go ahead and run it, which you'll be able to see in a second. All that's happening here is that Claude has seen there's an instructions file, read it and found every step it needs to run. That then goes into a few different outputs, which it writes into the categories and corrections files.
Why am I doing this instead of writing the prompt directly into the chat? It goes back to the explicability part of the ethics. It's also much more efficient and allows you to start building layers of automation.
It's really important that you can break this process up, and it's going to save you so much time. You've got that one prompt stored in an instructions file, clearly set out in a way your AI reads it, so you save yourself having to constantly write those prompts. It also means you can share the entire folder with your team and say: I've worked out how to automate this process – here's the answer.
You give it to everyone, and not everyone has to relearn it. You can see here that this is stage one, our prepare stage. It's gone through the finances and talked about the differences in there.
It's told me there's a decision that needs to be made, and that there were some duplicates in there, which it's removed. Why has it removed them? Because in my instructions, I've told it to remove duplicates.
It's also flagged possible duplicates – things it wasn't sure about – and asked me to confirm, so it's come back to me with some questions. What you'd be able to do is make sure that after each stage, you've got your human-in-the-loop piece.
But this isn't the most efficient way to run this process. You'd have to answer these questions, get an output, copy and paste all those outputs back into your folders, navigate back over to your stage two folder on the left and put that into the chat as well. That takes a bit of time.
It's really great while you're building the process out, testing and trialling. But in the long run, that's not quite what you want – you want this to be automated and effective. So what I've done, as you'll be able to see here, is to build a project.
As I said, in Copilot it's going to be notebooks, and in Google you've got Gems, and there are workbooks you can run in there as well. In Claude, you've got projects. In this project, I've copied the entire folder, so all of these elements have gone down here on the right in the context part of the project.
That means whenever you put a prompt into the chat within this project, it's going to read all of that first. I'm also able to add some instructions. They don't need to be long – I've just said always read the context and process files first, and then execute the steps once the ledger, or my data sheet, comes in.
You can imagine logging in every month, giving it the new data sheet, and it running your whole process for you, because I've copied all the files over. And remember, in those instructions I've said always read the context and process files first.
That's because that's stage one. It's like going up to the librarian: Claude goes up to the desk in the library and says, I need to know the context – what's in the library – and I need to know the processes, so how do I go and find the stuff in the library? Once it reads that, it knows everything that's going on in all of these folders.
I'm going to pull up Claude full screen now so you can see, but I'm not going to run it live because it takes a little bit of time, and the buffering on my side isn't brilliant. You've got a bank of the entire folder, and I then added the ledger – the new dataset – which you can see in the top right here. It started taking me through the steps.
I didn't have to ask it to, or tell it how I wanted the process done, because I've got everything in those files and folders. The most important thing about having the files and folders, from a longer-term view, is that we've not spoken once about whether this has to be Gemini or Claude or whatever new model might come out in the future. It doesn't matter.
Let the tech companies run their race. We'll get our processes sorted, and whichever model comes out on top, or whichever one you prefer, we can then use. You might have three models that you want to use for different processes within your team.
It gives you the agency to write down your automations in a way that lets you apply any model, no matter what. If you imagine having this process mapped across your entire organisation, you'll be in a place where you dictate how AI is going to affect your business – not Google, not OpenAI. It all gives you the agency back.
So it started with step one. You can see it's gone through preparation, told me it's done and come out with some questions – it said it needs my decision on three things. So I answered those questions.
It was telling me there's no category for this particular purchase, and asking me to explain it or say where it fits. It said these lines have different references, but it's the same debit that's been charged. It's asking me questions about what I want, and this is really important.
The human in the loop is built in, and your ethics are built in. I've given it my answers, and it's asked me a couple more questions because it wasn't quite sure. It asked how I'd like it to handle adding the corrections for this duplicate into the file, and I've said keep both of them – don't worry about it.
That's what I want. Again, I'm making the decisions; the AI is not doing it for me. Then it's gone on to step two – it said, Tell me when you're ready, let's go on to step two – and it's started flagging different things again, which are all in the instructions.
When I send the folder afterwards, you'll be able to read through the instructions. Step two is then complete, and step three is complete as well. I've ended up in a place where it gives me a headline and summary so I can double-check everything's good.
It's also told me what we've been through in the previous two steps, just to make sure everything is right. And then this is a really important line, guys: it says nothing should be shared until someone has signed this off.
That's your human in the loop – you're built into this process, and your expertise is built in. I've said it's signed off, because it's not a real report or a real business, so that doesn't matter. And it's given us a monthly report.
Everything that's in this chat, guys, you should be able to go and do, so go and try it. I think you'll be able to see as you go through it how this template maps onto your work. That's the demo, so I'm going to pause here.
Anyone who wants further help with this, do let me know – we're very happy to give time. As a final message, I'll just pull this back up. Go and find somewhere to start small – the simplest process is the best place to start.
Pick something that's going to save you a bit of time, and just have a go. You'll start to see how you can really build your business off the back of this.
Beth: Brilliant. Thanks so much, Kiran – that was great. Lots of people are saying excellent presentation, love this, great demo.
Just a reminder that we're going to send the ZIP file later today, so you will have that. I know we're out of time, but if we could maybe do one or two questions, we've had a few in the Q&A. Just to clarify, we've had a question from the audience: is the corrections file essentially a log of errors or changes made to AI outputs, which it can then refer back to for future responses?
Kiran: Exactly. Let's say I run stage one and something goes wrong – there are duplicates I didn't know about before. That's how your AI learns.
It's going to ask, is this a correction you'd like me to store for the future? The action might be: whenever there's a duplicate that looks like this, flag it to me and then remove it, or don't do anything until I've had a look at it. That information will then get stored in your Excel sheet, and it will read and use it every time moving forward.
Beth: Perfect, that's great. The second question is from the same person: when you've set up a project like this, what happens when you upload a new version of the ledger? Does AI automatically know to follow the same process, or do you need to use a new prompt each time to get it started?
Kiran: If you go into that project, you've put all of your context there – the entire folder. You'll have written in your context what the folder is for, and instead of just this ledger, you'll have, say, the November ledger. When you upload that, if you need to, you can say, " We've got a new piece of data – run the monthly report.
It's very intuitive, and it will do all of that for you. Do try it, but in theory you can just upload your new monthly data and it will go and execute. That's basically how it works.
Beth: Perfect, that's great – thank you. Unfortunately, that's all we have time for today, but please do reach out to Kiran, who I'm sure will be very happy to answer a few more questions after this call.
Thank you so much for joining us, and we'll send you the follow-up email later today. Thanks, Kiran, and see you in the next one – bye now.
Explore our webinar library
Unlock more on-demand sessions designed to help you sharpen your skills, grow your business and stay one step ahead. Find more Lunch and Learn webinars