Introduction
Welcome back to another episode of TJ’s Technology Tuesday. Yes, automation and AI agents are on everyone’s lips right now – but far from everyone’s hands. That’s why I’d like to share my personal perspective on the topic today. I want to walk you through my own journey with it, because this always comes back to the question I keep asking: What is relevance over gimmick?
I have been guiding companies on productive digitalisation not just since yesterday – I have been involved with the topic privately for 43 years, since I got my first C64, and professionally for 37 years. So:
Why It Does Not Pay to Digitalise Bad Processes
There is a lot of hype out there, a lot of shiny, colourful things. And the most important thing first: the quote from Thorsten Dierks, former CEO of Telefónica, who said: “If you digitalise a crappy process, you end up with a crappy digital process.” And that quote is as true as it has ever been today – because people have always tried to use digitalisation to magically improve bad processes. And now they are trying to use AI, layered on top of bad processes, to simply make bad processes good.
And the bad news is: that does not work. Reality shows that when I digitalise processes with AI, it is a bit like the saying you know from me: A fool with a tool is still a fool – and the fool with Artificial Intelligence makes the disaster faster. Things get worse and worse and worse – not better. There is yet another dimension in the world of AI. If you are in a situation where you simply do not have good data, then you cannot work with AI either, because there is always that motto – already known long before AI – Garbage in, Garbage out, or Nothing in, nothing out. A former colleague of mine, now in a senior role at a large holding company, had Palantir on site recently. They wanted to unlock significant efficiency gains and potential in production using AI. And Palantir’s statement was crystal clear: Feel free to come back in two years once your data is ready. Of course, you do not have to wait two years, and you do not have to be a large corporation to use automation – but let me share from my practical experience how I have worked with automation so far, what I do, and what might also make sense for you.
If This Then That – A Pioneer in Automation
The very first tool I ever used was called – and still is called – If This Then That, I-F-T-T-T, coming more from the consumer space. With it I could already do things like: save screenshots from my iPhone or iPad into a specific album, or say: whenever I post something on Facebook, please create a tweet on Twitter – today on X – accordingly. Or: whenever I post something here, pass it on somewhere else. So this idea of If this, then that. And why do I mention it? Not because it is a service I use today and recommend to you, but because it is a beautiful illustration of a fundamental principle that you find in virtually all of these tools. You have a trigger and you have an action. If this, then that. If you do this, that happens. And you can do that not just with two services, but with a huge variety of them. And 90% of what is being sold as AI agents are not AI agents at all – they are simply process chains: if this happens, then that happens. And that is neither good nor bad.
Why Using AI at Any Cost Is Not a Good Idea
What I keep seeing, though, is the cry: we need AI, AI, AI. In many areas what we actually need are clean processes. And the first step is: do I need this process at all? The second step: if I do need it, look at each individual process step and ask: where might there be steps that are only there for historical reasons? I was with a client on Friday and they said: yes, that’s how we’ve always done it, these processes. Well, we’ve always done it this way – sometimes that can be fine, but it is always worth asking: do I need this process at all? Do I need every single step in it? And many of these steps can potentially be automated. A typical example: when I have a person in the decision chain, I always ask: what are the decision criteria? And if the criteria are very clear and unambiguous – for instance, if an employee has zero remaining holiday, they obviously cannot be granted more – then the process is pretty straightforward. That means I can fully automate a process: Does the employee still have remaining holiday?
Is their cover in place, or are they ill? If that is not the case. And you can still consider whether – after all the pre-checks – a real manager still needs to make a final decision. In the context of AI-aided decision-making, for instance with credit decisions, there must always be a human in the loop. But many preparatory activities can already be handled by automation or AI. That was the first point. And perhaps some of you are thinking: wait, saving things on the iPhone and so on – aren’t there Shortcuts for that? Absolutely – I’ll show them in a moment. Apple Shortcuts used to be a separate application; Apple bought it and it is, in my view, one of the most underestimated features available. Users of iPhone and iPad can use the Shortcuts app. On my iPhone, for example, I have the Action Button – which you can also assign to a Shortcut – set up so that it directly launches the voice mode of Perplexity. So here too you can automate things: when I press a specific key, when something comes in, even directly on the device.
The downside is that I need to have the device in hand to do anything on it, but the upside is that no cloud services are used. That is also a bit of a philosophical difference in the world of AI – especially on-device AI at Apple – while Samsung and Android services tend towards a philosophy of doing it on servers. That is also one of the reasons why Apple is a little behind here, because it is considerably easier to do this server-based than device-based. But it is a question of philosophy, and what matters is always understanding how these systems work. It is always that brain-first approach: what do I need and how? There is no single automation tool or solution that I can point to and say: this is the one for on-device, or in the cloud, or in your own data centre. Instead, there is the right solution for you for different use cases – which you can also combine with each other where appropriate.
Automation Tool: Zapier
One tool I use very intensively and that has been on the market for a long time and is very stable is Zapier – spelled Z-a-p-i-e-r. For me, Zapier is the data hub where I can connect things together.
The idea here, again, is primarily visible in the templates: you can see things like – I want an email notification when a new form is submitted; I want to write new entries into a Google Sheet; I want to push something entered in Google into Slack; I want to send an Outlook email for new form entries. I have deliberately picked these templates as examples. They are nice illustrations for asking: do I actually need a notification every time a form comes in, or does it make more sense to have a Kanban workflow – using a Planner board or a board view list in SharePoint? That often makes more sense. So please do not always orient yourself by these templates. But Zapier’s great strength is that it can connect to an enormous number of systems. When I look at my own Zaps and what I typically do: for example, when something is entered in Outlook, it gets entered in Daylite, my CRM system. When my AI phone inbox receives a call, a call note is automatically created in my CRM system.
That is a typical example: my phone AI has no direct interface to my CRM system, but both can talk to Zapier. That means an important selection criterion for software today should be: does it have a Zapier integration? Then you can say: when a contact signs up for the newsletter, they are automatically added to your CRM system. These kinds of things can be automated beautifully with it. And it again works along the lines of If this, then that. When one thing happens, then the other. Zapier – rock solid. There are simply an enormous number of integrations, and they keep developing the system further. They naturally recognise that they also need to add more AI functionality and database features. So there are also overlaps with further tools, such as Make.
Automation Tool: Make.com (formerly Integromat)
Make used to be called Integromat, is now called Make, and again the idea is similar. In the templates you can see that there are also many two-step templates, but you can see you can also do more steps and very complex, multi-stage things. And here again there is the option – let me bring this in – a typical workflow that you can build here.
You can also get guidance on it, so that you have the possibility to say: I would now like to add AI components into such workflows. And that makes sense especially when you have unstructured information, because these types of flows and automation work well when you have structured information. If this happens, then that. Someone signs up for the newsletter, it goes into the CRM. But if someone calls, we do not want to force the customer to say “press one, two, three or four” – instead I try to extract the relevant information from the conversation. That means I can use AI to identify the right contact person from a phone call, extract structure from the unstructured communication with the customer, and then attach a process to that. So you can combine the two. Make.com, formerly Integromat, browser-based, where you can do all of this – and there are plenty of tutorials available.
Automation Tool: n8n from Berlin
Then there is what is currently being touted as the latest and greatest thing in digital circles: n8n. It is a Berlin-based company, and this Berlin company also builds AI workflows. On their site you can see, on one side, the logic of how it works.
The core element in the centre is an AI Agent, to which I connect an AI model, a memory – a kind of brain – where I say: remember things accordingly, to which I can then connect things like an Active Directory if I want to access employee information, to which I can connect external systems. And then there is a typical example where I say: new employee – when I create a new person here, the system reads from the systems whether this is a manager or not. And depending on whether it is a manager or not, it enters things differently. A classic example of combining these two elements. n8n – a German company – very, very powerful, can also be extended with code. One small downside: it has a relatively steep learning curve before you get your first results. Something that tends to get overlooked is the topic of Power Automate. It does not look as sexy as what the folks at Microsoft do there, but it is a very, very powerful tool – especially for those of you who work with Microsoft 365. Here I can create automated cloud flows where I say: If This Then That – when a specific condition is met, it fires.
An instant cloud flow would be something like a holiday request, where I have a button I press. A scheduled cloud flow is one that runs, for example, every Monday with an invitation to a regular meeting – although here too you have to ask whether every process you can automate is one you should automate. Then interestingly, there is the Desktop Flows area. This is a little confusing because people tend to think it is the accompanying software for running flows on the local machine. No – it is an RPA component, a Robot Process Automation component. That means I can simulate mouse clicks on a PC. Something that many of my Raiffeisen bank clients use. It is a bridging technology where I say: if I have a legacy system, I can combine manual steps in those systems using this. As you can see, there are a great many options here. A great many possibilities, where the answer depends on: what devices are you working with? Do you want to work offline? Do you have software beyond Microsoft 365? Then the most important tip: make sure it has a Zapier integration.
Conclusion
And if you have Microsoft 365, then in that regard you should take a look at Power Automate. And I would even say: look at it before you look at the topic of Copilot Agents – because with Power Automate you can naturally also build AI-based things, and you can combine the two. As always: brain first, then technology. And please do get in touch if you need input or support in this area. I am always happy to accompany you in selecting the right automation solution and of course in implementing it – because in Germany we do not have a knowledge problem when it comes to digitalisation, we have an execution problem, and I am here to help you solve it.
Your Personal IT Coach for executives, Thorsten Jekel.
Key Takeaways
- Bad processes do not automatically become better through digitalisation or AI – whoever digitalises a bad process ends up with a bad digital process.
- Before any automation, the question must be asked: do I need this process at all – and every single step within it?
- 90% of so-called AI agents are not real AI agents but simple process chains based on the principle “If this happens, then that” (If this, then that).
- AI only works well when the underlying data is of good quality – Garbage in, Garbage out.
- IFTTT (If This Then That) popularised the fundamental principle of automation – trigger and action – and this principle is embedded in nearly all modern automation tools.
- Zapier is a stable, widely used automation tool that can connect a huge number of services and acts as a central data hub.
- Make (formerly Integromat) enables more complex, multi-step workflows and the integration of AI components for unstructured information.
- n8n is a powerful, extensible open-source tool from Berlin with a steeper learning curve that enables genuine AI agent workflows.
- Microsoft Power Automate is often underestimated but is a very powerful automation tool – especially for Microsoft 365 users – including RPA (Robot Process Automation) capability.
- The Apple Shortcuts app is one of the most underestimated automation options on iPhone and iPad for device-based automation without cloud services.
Frequently Asked Questions
Why does it not make sense to simply automate or improve bad processes with AI?
Bad processes do not automatically become better through automation or AI. As Thorsten Dürks (former CEO of Telefónica) aptly put it: whoever digitalises a bad process ends up with a bad digital process. Applying AI to a bad process only makes the error faster and at greater scale.
What should you do before automating a process?
Before automating, you should first check whether the process is necessary at all, and then critically examine every individual process step. Many process steps exist only for historical reasons and can be eliminated or simplified before automation is even considered.
What is the difference between real AI agents and simple automations?
The vast majority of solutions marketed as AI agents are in reality simple process chains based on the “if this happens, then that” principle. Real AI agents – such as those built in n8n – connect an AI model with a memory and external systems and can also process unstructured information.
What does the principle “Garbage in, Garbage out” mean in the context of AI?
The principle means that AI systems can only produce good results when the underlying data is of good quality. If the data is poor or incomplete, the AI also delivers poor results – a company like Palantir has confirmed this in practice, asking clients to return in two years once their data is properly prepared.
What is IFTTT and what role did it play in the history of automation?
IFTTT (If This Then That) is one of the earliest automation tools that popularised the fundamental principle of “trigger and action”. Although it is used primarily in the consumer space today, it illustrates the core principle behind virtually all modern automation solutions: when a specific condition is met, an action is executed.
What is Zapier particularly well suited for?
Zapier is particularly well suited as a data hub for connecting a very large number of different software services – even those with no direct interface to each other. A typical example: a phone AI and a CRM system that have no direct connection can communicate via Zapier, so that after a call a conversation note is automatically created in the CRM.
What can Make (formerly Integromat) do better than simpler automation tools?
Make enables more complex, multi-step workflows into which AI components can also be integrated. This is particularly useful when unstructured information – for example from phone calls – first needs to be converted into a structured form by AI before an automated follow-on process can be triggered.
What are the strengths and weaknesses of n8n?
n8n is a very powerful automation tool from Berlin that enables genuine AI agent workflows, can be extended with code, and covers complex integration projects. The biggest downside is the relatively steep learning curve: it takes longer to achieve first usable results compared to simpler tools like Zapier.
Why is Microsoft Power Automate often underestimated?
Microsoft Power Automate is often underestimated because it looks less visually appealing than competing products. However, it is a very powerful tool that – in addition to cloud flows and scheduled automations – also offers an RPA (Robot Process Automation) component that can simulate mouse clicks on a PC, making it ideal for legacy systems without modern interfaces.
When is a human still necessary in an automated process?
A human in the process (“Human in the Loop”) remains necessary when decision criteria cannot be defined clearly and unambiguously, or when legal or regulatory requirements apply. With credit decisions, for instance, many preparatory steps can be automated, but the final decision must still be made by a human.
Tools & Resources Mentioned
- IFTTT (If This Then That) – Pioneer of automation, consumer-oriented, core principle: trigger & action
- Apple Shortcuts – Device-based automation on iPhone and iPad, without cloud services
- Zapier – Stable data hub with a huge number of integrations for business automation
- Make (formerly Integromat) – Browser-based tool for complex, multi-step workflows with AI components
- n8n – Powerful open-source automation tool from Berlin with genuine AI agent support
- Microsoft Power Automate – Underestimated, powerful automation tool for Microsoft 365 users including RPA
- AI overview on digital4productivity.de – Further information on AI topics
