Hey, everybody. How's it going? Happy to see so many faces here tonight. I appreciate you taking the time to be here.

So I need the remote. That's it. And let's start with a very quick show and so that we get moving a little bit.

So how many of you are actually who's here tonight night that used AI this week? OK. And who here used AI this week for work? Lovely. And who here runs the business? Thank you. OK.

Why AI Initiatives Often Fail to Deliver

Let's start with something a bit spicy. According to an MIT Nanda preliminary report, 95 % of the organizations they studied were getting no measurable financial return from their generative AI initiatives.

So a question that probably went off in your head should be how come that with the tools this powerful the The return on the business is so disappointing.

And you know what's another question that should be going off in your mind? Who the hell is this guy that keeps suggesting new questions? So, to answer that one, I'm Fabio Rosato. I'm a software engineer and AI consultant.

I have had the pleasure to work over the years with many interesting companies, Reply, Bending Spoons, and San Rafael Hospital, just to drop a few names. And now I help established businesses put AI to work.

So, you see, no matter the starting point you have tonight, I want you to show you three things mainly. First, how to choose a problem where AI actually can help. Secondly, how you can begin with a prototype, experimenting maybe even a simple conversation we're gonna do that and third how building and experimenting on what works could actually lead to something a lot more interesting for a business so back to that 95 % of

businesses getting abysmal results why is that the case well there's no single solution, but a good starting point to try to think about this is, what are businesses for?

Oh come on, another question? Yeah, I hear you, but hear me out for a second.

Start With the Business Bottleneck

Businesses solve problems. You know, a customer just wants a good service, plain as that. Adding AI on top of that only helps if it makes something that matters work better.

And there's actually a framework behind this intuition, it's the Theory of Constraints from Goldratt. Now the gist of it is actually pretty intuitive.

You have to, in order to improve a system or a business, you have to understand what is holding it back. back, because in that case you have to focus on that instead of trying to squeeze as much performance as possible from all the other components.

The Bakery Bottleneck Example

Let me give you a very quick textbook example to make this concrete. You have a bakery. You prepare the dough, you bake it, and then you pack it. At the moment, in this particular example, the baking process is the bottleneck.

You have 60 loaves an hour. That's your entire system output. You have 60 loaves. You can be ready to sell in an hour.

Let's say that you are an AI practitioner and you apply some fancy AI magic to the packing process. And voila, you double the throughput and you get 240 loaves an hour. Magic. Well, the thing is, the baking is still the bottleneck.

So no matter how fancy your packing technique is, at the end of the day, you still only output 60 loaves an hour. You still only have 60 loaves an hour ready to sell. So I think you get the gist of this idea.

And that's why, with the companies I work with, I always start with an audit, because the whole goal, and that's like 90 % of the work, is understanding where things get stuck. Once you know that, then we can work together and improve it. But until you know that, you are probably flying blind.

A High-Value Opportunity: Customer Inquiries

Let's dive a bit deeper. 50 % of established businesses, in my experience, the first point where I go look for improvements is handling customer inquiries. Why is that?

If you think about it, these are people that are contacting you, contacting the business, because they want a service, they are asking you how much it costs and they are also asking you what's the next step. This is like prime time for a business. It's what you live for.

Unfortunately, it's annoyingly frequent to mishandle these customer inquiries or maybe having a process that takes too much time to get back to them. And let's say how many, like who here, ever had to send a second email to a business in order to have a quote or ask for something? Yeah, the business might simply be busy.

The thing is that in that time the customer has probably contacted somebody else and even when your response finally arrives that initial delay has already shaped the first interaction that you had with the customer so also the first impression let's make it a bit more

The Catering Business Scenario

concrete imagine you have a catering business and a new inquiry just arrived Yay! Alex here wants foods and drinks for his event of 50 people. His budget is 2 ,000 euros and he's still a little bit on the fence about the date and the venue.

Okay, you are the owner, so you receive this inquiry. Maybe you are busy right now because you are handling another event, so you put it off for tonight. But tonight you're tired and so tonight becomes tomorrow. tomorrow. And Alex is still waiting for your response.

Tomorrow a rushed reply could very easily miss some information and you know how the story goes.

Three Friction Points to Solve

So when we think about this, if we think about this, we can identify three smaller problems or issues with the the matter. We can decompose it in basically three main annoyances.

First, the reply, the actual creating the quote takes time. It's boring, it's necessary work, but takes time.

Secondly, following up is easy to forget. Third, new inquiries are very easy to miss in your inbox. Okay.

So this is the scenario. Let's work on it together.

Prototype the Solution With AI

The incredible thing about these days is that you can have meaningful progress on this matter just by talking to Judge EBT or Claude.

You don't need no fancy setup, you don't need no magic prompt. It's bad to say but that's mostly a marketing gimmick, having the right prompt to get the results.

Let's start very simple and easy, high high -level, treat it like a colleague, and say what you need.

Drafting Faster Quotes and Replies

For example, here, help me reply to the scattering inquiry. We paste the request. Is that enough? No. Why?

Because we need to provide the AI. It's like a new hire. You need to provide it with additional context, the business information you would need. So you actually give it the information it needs.

In this case, we paste in the menu and prices fairly easy stuff obviously we could have also attached the menu file whatever then you say a little bit more concretely what you want back give me a price breakdown and a short friendly reply i can review use only our prices is an actual guardrail we'll see it in a minute but the important bit suggestion

suggestion I have for you and a piece of advice is that always, maybe not, but often, giving it an example of something you like works a lot better than providing it with long but somewhat vague information. So in this particular case, we just say, for the tone, use this example that I like. That works perfectly and goes far beyond this very simple example.

Then, we tell it to use our prices and ask for missing details. These are guardrails. Basic guardrails so that it asks us for information or flags missing information instead of making it up. That's the problem with AI hallucinating and all that jazz.

All right. We submit the ask and lo and behold, in less than a minute, we got a nice breakdown and and a nice reply, 1 ,720 euros, entirely within Alex's 2 ,000 euros budget, and we also have the draft ready, detailing everything, and you just have to sign your name at the bottom. Obviously, you, as the owner, do your due diligence and check the numbers and the promises, but then off it goes, you're ready.

So in case the AI returns you something that you do not quite like, simply ask, work with it, talk to it, make it shorter in this case. Just a silly example, what I want to convey you is the idea that you have to drop the mindset of one -shotting the perfect answer for AI. It's a lot better to work with it exactly as you you would with a human, and then get to something you like.

Obviously, you do not have to stop there. You can ask it to turn this into a PDF code. There's no reason to stop with chat messages. I know a lot of people that use a chat GPT or cloud, and then just keep it there. No reason to do that. Again, you review it, then you send it.

Provide Context, Guardrails, and Feedback

Finally, this is another very important habit to get into. to, say you are not sure what to tell the AI, what information to provide in order to actually have it do its job. What can you do? Ask it. What information do you need from me to do this well?

Again, this may seem like basic stuff or simple stuff. But again, from my work, it's annoyingly frequent to completely ignore this kind of basic stuff and go for fancy solutions that do not always work as well as you think.

So one job, checked. Replies and quotes take a long time. Not anymore. A minute and we are done.

Use AI to Improve Follow-Up

Let's tackle the second one. So you send the request. Three days pass and Alex has not replied to you. Bummer. So what do we do?

You are probably the owner and you are probably very busy so you carry on with everything else and you forget about it. Good? Hell no. You have to follow up.

Because otherwise, you know, there are statistics that 50 % or more of requests are completely ignored. You have to follow up. But following up is easy to forget. So what can we do?

Let's go back to our conversation with ChatGPT and ask for a gentle follow -up in case you need assistance to create a follow -up. If the conversation goes too long and gets too messy, obviously the advice is to create a new one. In that case, you have to provide the context again.

We can actually do even better. We can ask it to remind us. Now, granted, there are a variety of ways to set a reminder, but we're already in ChatGPT your clothes, so might as well. In three days at 9 in the morning, remind me to check whether Alex has replied.

And you get a reminder in the same conversation that contains all the context you have and you need for handling this. Again, it might seem like simple stuff, but compare that to, I don't know, setting a reminder on Apple reminders or whatever other platform you have.

Make Sure New Inquiries Are Not Missed

Two jobs made easier, now let's tackle the final one. one, new inquiries are easy to miss.

So, you are the owner, you are preparing for an event or maybe you are networking, who knows, and while you are running an event, you cannot like keep watching and scrolling the inbox all day, right? Yeah, you have stuff to do, plus you get a lot of invoices and newsletters and spam and just like that, new inquiries inquiries are very easy to miss in the inbox, in the deluge of emails you get. If only there was a way to get notified automatically when a new inquiry comes in.

Well, maybe AI can help you with that too, because you can simply connect a Gmail account to chat GPT. You connect the account, then again, the good habit, try one request manually. Tell it to fetch some e -mails, maybe e -mails that are interesting to you, for example, find new requests for gathering a quote. Tell it what to ignore, so ignore newsletters and sales e -mails. Then you can also specify if you need precise information back.

Why am I telling you this? Because this is the habit. You try one manual request, you check what it finds, then you refine the request, and And when you do all that, again, we're in the same pattern. A clear job, give it enough context, a result to check, and iterate.

After you do that and you find that it fetches the correct emails, you schedule it later. So at the end, you can say, run this inquiry check every weekday at 9 AM and 4 PM.

Three problems tackled. Again, I'm not doing any kind of magic. It's not wizardry. It's super basic stuff, but I hope I gave you the kind of habits and mindset that's a lot better to use in these cases.

Also notice that to do all this work, we used basic church GPT, and we made some amazing progress on this particular annoyance. ones.

From Chat-Based Experiments to Structured Systems

Experimenting with it gives you also a clearer idea on where it's important enough or worth it to take things a bit further.

For a business, that next step of taking things a bit further usually comes when there are too many conversations happening at once. So maybe you have a customer that wants to change the number of guests, or maybe another customer doesn't know if you replied with a quote or not, maybe didn't receive it, or

maybe, again, you have a teammate who does not know if somebody replied to a particular customer. In that case, having a chat interface, so doing it through chat GPT, becomes very impractical and start actually getting in the way.

So in those cases, if you experimented with it and you found value in this, it's easily worth it to invest a bit more and get a structured solution.

Build a Workflow Around What Needs Attention

Imagine one way, one screen, where there's basically everything you need. And for a business owner, that's what's happening happening and what needs your attention.

Obviously, to do that, you need a bit of software engineering and proper investment. But then you could integrate the manual work that you did, you experimented with on ChartGPT, and get a solution that's built entirely around your bottleneck.

In this particular case, for catering, we have one place to see what needs you, what needs your attention. So emails come in. They are automatically processed by the system. We have status of each inquiry and next action. So you as the owner can come in, open this page, and immediately see what's going on.

We have Alex with his proposal ready and we just need to review. Marta with a follow -up, review the follow -up, and finally Elena whose inquiry needs a bit more details. Probably she misrepresented something or we just have to provide some details.

You see, as the owner these kind of things is probably invaluable because you experimented with it and you found value in the first iteration. Here you have the request, the price breakdown, the draft reply and date and venue are clearly marked as still needing an answer.

The owner makes the call, we have the human in the loop. That's optional but very recommended usually. You as the owner check, approve and the system automatically manages it and keeps track of what's next.

Apply the Same Approach to Any Business

That was for catering but I hope I convinced you that this approach can be used with a variety of businesses. And though the details may vary, the underlying idea stays the same. Find what's holding things back, improve it, and when you find that you would benefit from a more structured solution, then you invest in it and you get a solution built around your bottleneck so that customers get a proper service and your team has more room to serve your customers and do the work that matters.

Key Takeaways and Next Steps

So, what I want you, and what I hope you take home from this tonight, three things. First, applying AI is useful if we apply it to a problem that matters. And that's why we start with an audit and apply it to a bottleneck.

Secondly, begin with the simplest iteration, a prototype, experiment with it, even a conversation, as I showed you, can realistically solve 90 % of the annoyances. Then, when you find that that helped and you need a more structured approach, in that case, build on it with a proper investment and an ad hoc solution.

One job, think about it, think about one job that's taking longer than it should or maybe that's giving you more pain that you are willing to tolerate. And try, even tonight, as you go home, try to think about the actual annoyances with that problem and try to have one honest conversation with AI.

It may sound simple, but trust me, people don't do this. Try it tonight, give it context. Again, that's the mindset. Check the result and iterate on it. Keep talking to the AI.

You can scan the QR code to get these slides so that you can try some of these tonight at your home, at your pace.

If you are a business owner, I would love to see you at my event for Rome Future Week tomorrow night. It's going to be in Italian, and it's going to dive a bit deeper on how AI can actually help you grow your business.

The details and the booking link are all on that page, along with how you can work with me. so scan the QR code all the goodies there I look forward to it thank you so much