Why Most AI Training Fails at Work | Elise Masurel de Laval

AI training usually fails for a simple reason: it is too generic.

In this episode of Business AI Explained, I sit down with Elise Masurel de Laval, co-founder of Catalyst.ai Academy, to break down what makes AI training actually work inside companies. Elise and her team focus on practical AI training for non-technical roles, with a strong emphasis on role-based use cases and learning by doing rather than abstract demos. 

They discuss why AI adoption needs to be tied to real business workflows, why peer-led learning often beats formal training, and how companies should think about shadow AI, governance, and long-term capability building. They also explore why AI agents create excitement early on but often become difficult to maintain unless internal teams can truly own them.

They cover:

•  Why most AI training programs fail 
•  The case for role-specific AI enablement 
•  Why peer practitioners are often the best teachers 
•  How to build relevance through real company use cases 
•  Why AI communities can outperform static courses 
•  How to think about shadow AI without overreacting 
•  Balancing experimentation with governance 
•  The hidden maintenance cost of AI agents 
•  Why practical adoption matters more than tool knowledge 
•  How non-technical teams can build AI confidence quickly 

This episode is for founders, operators, enablement leaders, consultants, and anyone trying to move from AI awareness to real adoption across teams.

About the guest
Elise Masurel de Laval is co-founder of Catalyst.ai Academy. Catalyst describes its approach as practical AI training for non-tech roles, and the company’s about page highlights Elise’s background in marketing, digital, sales, innovation, and executive leadership in education. 

Where to find Elise:
→ LinkedIn: https://www.linkedin.com/in/elise-masurel-de-laval-2579b999/?skipRedirect=true
→ Company: https://catalystacademy.ai/

Work with Vlad:
If you’re implementing AI in your operations and want hands-on help building real workflows:
→ https://www.elementsagents.com/

Subscribe / follow Vlad:
→ LinkedIn: https://www.linkedin.com/in/vladeziegler/
→ AI with Vlad: https://www.youtube.com/@aiwithvlad

Transcript

SPEAKER_00 0:00

Hi everyone, today I'm joined by Elise LaSurel. Elise LaSurhel is the former head of Culinary School from this very famous chef called Anna Dicas and a former executive at ClubMed. She is now running an academy called catalyst.ai, which is essentially a coaching program for to upskill people on how to use AI tools in their companies. I think this podcast is a gem because she's gonna teach you exactly what makes companies to like adopt AI tools today. How do you strike a balance between encouraging people to take ownership and be creative at their job while also ensuring that there's the right governance and support from the execs? And finally, because she's been growing this company like crazy for the past year, she's gonna tell us more about the different workflows that she has implemented to drive more sales, but also automate a bunch of her back office processes, which used to take up so much time around onboarding new clients. So very excited about that one. As you know, this is Business AI Explain, where we bridge the gap between an AI demo to implementing AI in production in your operations. What I've experienced with Elements Agents where we build AI applications is that companies get very excited about AI, but when it comes to actually implementing it and trusting it, this is a completely different game. And so the point of this podcast is to bring people like Elise and other AI experts to tell us exactly the different learnings that they've had in implementing the solutions, their preferred tools, their ideal use cases, and the kind of impact that it had on their businesses. So, with no further ado, let's get going with the podcast. Perfect. So, like I mentioned, um you have an amazing background, very diverse, and there are a lot of things that I want to unpack in this episode. Um, but first of all, if we just cover uh Catalyst AI Academy today, can you just brief you know give us a brief overview about your about the company and you know what's your sp your special source and angle in how you deliver uh AI training to companies?

SPEAKER_01 1:54

Yes, uh so um when we come to Catalyst.ai Academy, the specificity and the strong convictions uh we have is that uh first of all, all AI trainings we do will be very practical because it's key when uh you start to uh understand how AI works, it's really important to um uh do and to train yourself on uh reality, on the real tools, on the real use cases, and on real business examples, uh, because it's the best way to understand how you are going to use AI on your daily life. And if you don't practice, you are not able to activate uh the day after the training all the things you have learned. So that's the first key conviction, and I think uh it's a real um USP, I would say, uh, of our trainings. The second thing is that uh AI is not used the same way from one function to another. Uh in marketing, in finance, in HR, you will not have the same use cases, you will not have potentially the same tools, and you will not have the same processes. And uh, as all of these things are different, you need to be trained per function. That's the second uh key conviction. And the third one is that uh you need to be trained by peer, and that's why all our trainings are done by people who have a very, very strong experience in the department, in the field, in the job, my meaning uh financial director, marketing director, HR director, um uh lawyers, etc. etc. Because they will be the best one to train their peers because uh they know from I think uh three years now how they should implement AI in their daily life. Because they do that for themselves, they use AI on a daily basis for their own uh department, for their own field, and that's why they are the best one, the best one to train um the people who do the same job than them. So that are the three main convictions. I think the the main uh key elements of our uh secret sauce, as you were uh as you are mentioning it. And um people are very impacted by our trainings because they say to themselves, this is exactly my day-to-day job, so I can actually work with AI when I see these people doing that in her or her job on a daily life. It's the best proof, I would say, to showcase how AI can be efficient on a daily basis for the same function.

SPEAKER_00 4:50

Now, I okay, so if we if I try to summarize, uh not summarize, but like the three core uh elements you said basically you need to build the muscle. So you need to be given the tools to play around and be forced in a room to try out these tools. The second one is you need to have something that is relevant, so a demo for like a marketing department to show us basically what you can do in marketing. So uh whereas like you would probably have like different demos and different examples if you are in the design team or like ops and sales. And then the last one is you want to be like those demos need to be delivered by people you can trust and you can relate to. And so they are generally delivered by experts. Like you said, you have this network of experts who have been doing this for three years. Uh so you would say like this is uh basically like the core components on how you deliver the program.

SPEAKER_01 5:36

Yes, exactly. And even more than demos, it's real business cases, I would say.

SPEAKER_00 5:44

Interesting. Can you uh tell me more about this real business case? How does that because this is like the main problem is like you can get people excited with the demo, but yeah, it's very hard to tie it back to like all the subtleties and the edge cases that happen in a real business environment.

SPEAKER_01 6:01

Exactly. And that's why we always have a time to prepare and personalize the trainings we do. Uh we personalize uh in line with uh, of course, the field, the function, the job, but also to the sector, for example, or the key priorities of the company. And then we will define very concrete use cases that will be relevant uh for the people we are going to train. To give a concrete example, uh, if we do, for example, um training in hospitality, uh, we are going to work on how uh the people can uh optimize their sourcing, their suppliers' sourcing, how they can optimize the way um they are uh uh interacting with their clients, how they can enhance personalization, reactivity also, because uh thanks to AI you can be much more reactive with your clients. Um to be also uh much more accurate on all the financial and PL part and how to uh accelerate uh the way you produce and you analyze dashboardings, etc. etc. So we are going to work really close to the needs of the people uh we are training and based on very, very concrete business use cases adapted to the reality of the people we are training.

SPEAKER_00 7:23

Okay. And um so basically you have a lot of moving parts. On the one hand, you have the the business you're gonna prepare the training for, who will like give you data and like specific problems that they're trying to solve. So you're tying this back to the strategy, the business strategy. And then you also need to find in your network of experts the best people who can actually give good advice for that specific niche or business uh problem. Is that yes? Is that yeah?

SPEAKER_01 7:50

Yeah, yes, it's uh really um we we are building a real community to be to be clear. Uh so um on one side uh we are working on uh all the relevant uh use cases that we can develop with AI uh on the different fields and on different sectors. So now we have a strong and big glossary of all those uh very specific use cases per function, per sector, etc. Uh, and through all the trainings we already did for sure, and all the transformation we can see. And on the other side, um, it's really how uh we build this network, uh, as you were mentioning it, uh, with uh strong experts, uh, which are at the same time experts of AI and experts of their field. Uh and um and we are very lucky, to be honest, uh, to have this uh big uh expert community uh of experts because um uh we uh are able to share among all of us all the news because it's uh doing it's going very, very fast, as you know. So it's great to see all those experts on so many functions that are going to share all their best practices, all their best use cases, uh, and uh sharing on everything they see in many companies. So it's a fantastic knowledge that we can share all together, and we multiply, I would say, uh, the expertise thanks to this uh very strong community.

SPEAKER_00 9:31

I I think uh when we talk about AI and how all the costs to development are going to zero and people are trying to think about how to make a business competitive in the long term, it's often about proprietary data or proprietary network. So basically, uh your company, you're building this exclusive network which really makes you stand out. I end up like just advertising your your I think your your your program. But I I'm actually just learning as we go, also, and I find it like very fast, very interesting, and a lot of nice moving parts that you're connecting together to make it valuable um for companies. So that's great. Uh maybe just a follow-up question on these profiles. You said that they've been working and playing around with AI for the past three years, uh, and probably you know it's gonna evolve and more and more people will actually pick up these tools. What's the typical profile of these people? Are there like independent consultants? Are they still in the main functions? Um, you know, why are they interested in like, you know, I'm sure there's like some financial incentive, but what really is the most uh exciting part for these people and what's the kind of profile that they have?

SPEAKER_01 10:33

Uh you know, no, it's uh it's uh we are very lucky also because we have a different type of profile, and it's also uh something very rich uh to have this different profile. We have uh, for example, um people who are independent. They were in previous life uh um head of uh marketing, uh head of finance, uh uh head of legal, uh, etc. etc. Um, and some of them decided to become independent uh because they are passionated about AI and they really want to focus on it and develop their own activity on AI, mixing usually and often uh consulting and training. Um, we have also people who are uh, for example, uh head of uh innovation or um sometimes head of AI also in companies, which is also something very interesting because it gives us also uh uh a vision of the way they transform their company, etc. And very often they come from other uh previous functions. Uh so they could be uh head of marketing, they could be head of IT, head of data, and they became um AI transformation officers. So it's great to have these kind of people with us, also. And uh the third profile we see and we we have in our community is uh consultants, uh notably in AI, yeah. Uh because uh they are really happy to share and transmit what they see also. And I think uh to be a trainer uh is a good competence to get when you are a consultant, so they are happy also to enrich their competences by being also uh a trainer. Uh so we we we are also working um sometimes with consultants, uh uh, notably also for very niche topics, uh, because they can be very uh interesting and they see many things too. So that's why it's also uh quite interesting that to include their view uh in this community.

SPEAKER_00 12:45

Yeah. Yeah, I think it's uh it's very interesting because I would say maybe a year ago it was it's still a bit intimidating to people to pick up tools like plot code and and and you know uh cursor just because there is code and you open a terminal, but very quickly within two weeks, people will get familiar and you know get very good at the profession at these tools. Um is this also what you see? Like what's the kind of stack or you know, what are the typical tools that all these experts that you work with uh generally use?

SPEAKER_01 13:16

Alright, uh first of all, to answer your question, we have one big uh conviction, again, another one, uh, which is that uh we really want to remain agnostic, meaning we don't want to push one for these or these tools, we really adapt ourselves to uh the best tools for the function. As we are uh doing uh a work really function, it's key for us uh to ensure that we are going to train people on the best tools today for their function because we know that everything evolves very, very, very quick. So we want uh to remain um uh a step ahead, so meaning that's why we are not we don't want to stick to one or another tool. Really, we want to be open and to be sure that we are always uh working with the best tool for the function. And um, the second uh part uh of the answer is that uh when um we uh work with companies, they usually are already working with a stack of tools. And the idea uh is not to change everything, you know, because uh there is a past, uh, there are many things working well, many processes linked also to those tools. So the idea is to see how we can infuse AI within the existing processes, within the existing uh stack of tools, and be able to adapt also ourselves uh to uh what exists in company. Sometimes the companies want to choose new tools, so in that way we can recommend a few uh solutions uh because they are going to be really relevant in the context, in the sector, in the job. Um but we don't have one stack we recommend for everyone, never. That's really not the approach. The approach is to be very personalized and uh choosing the best tool for the function.

SPEAKER_00 15:21

Yeah. Um, which probably means that every time you create those workshops, it's a lot of work because not only you need to identify the business cases, talk with the leadership, lead leadership, sorry, and um and then the second thing is you need to pick up new tools, and then you have to be able to teach teams on how to use these tools. So I guess like you've become really good at abstracting away all the intricacies and you know the uniqueness of each individual tool. So if you are to pick up a new tool tomorrow, what are the top rules that you try to follow? Is there like a systematic approach that you use to pick up a new tool? What you know, what are the key ways to pick up a new AI tool essentially today?

SPEAKER_01 16:03

The the the good news I would say is that um when you are able to work with uh LLM, for example, it's always the same approach. And that's why in the training we do, in any case, we are going to work in order to make people autonomous on any tool because it will evolve day after day, week after week. So our mission is that we are going to learn you how to plant, how to create assistants, how to create agents, and it's quite always quite the same rules, you know. Of course, there are some uh specific functionalities that you can find from one tool to another, but globally the methodology, the approach is the same, and we are going to teach on that, not on the tool. Of course, we are going to practice on a tool, but first we uh train on the methodology, the way to work with AI, uh, the good way to prompt, the good way to create an assistant, the good way to create an agent, and after that, you are autonomous to do it on many different types of tools. And we teach also agility because that will be the new story. It's not a new story, but uh it will accelerate even more, I would say, with AI. And so we we work also uh to uh to uh we train also people to be prepared, to uh um uh uh try new tools, to be always curious, and uh to um adapt themselves uh when uh new functionalities will arrive.

SPEAKER_00 17:47

Okay. So how do you how do you teach that? Uh you know, if we take a concrete example, how do you teach someone to become agile? Like do you force them, do you have their bosses to like force them to pick up a new tool that they have to present to their team every week? You know, how do you build this muscle? Because I agree with you, but it's hard to teach people to enforce them.

SPEAKER_01 18:05

Yeah, yes, and it's it's not only in one day to be honest on that, but first of all, we give them the panorama, we we we give them the panorama of the different AI available because people uh sometimes don't know that there are so many types of AI. So, really to understand how AI works and the different types of AI, the different types of tools also, because I think it's really important to know the panorama, at least to understand what is behind all of this and uh the type of tool for which type of use case so that I at least they can see what is possible. So, really to to open their minds um and to be aware of what is going on today everywhere in the world, yeah, and then to uh really um test a few tools so that they can see uh some differences but also some similarities in the way you work with AI. And you see when you uh work on the parameters of your tool, when you uh work on the prompt, when you create your system, there are similarities on LLM, etc. So it's really to make them agile on the key fundamentals of working with AI, testing different tools, and after that, we take always time uh at the end to um make them be aware that uh we we have trained on topics on day one, but AI is a journey, and they have to be prepared and remain curious about um the the tools they are going to work with. Um and I think uh when we say that it uh uh it's always very interesting to see how people say, Okay, so how can I build my um analysis of the new functionalities, the new tools, and being uh always aware. So we help them to build this kind of assistant, for example, so that every week they are aware about uh the new tools, the new functionalities, etc. So it's a starting point to be agile and a concrete example on how they can remain uh completely aware on a weekly basis, I would say, on what is going on. Um of course uh uh keeping uh in mind that uh uh they have to progress on one tool and not change every day, but be aware and change when it would be the good moment. And if the companies allow it also, it's easier the the way it needs to be uh uh of course subtile and adapt yourselves to the architecture of uh companies.

SPEAKER_00 21:02

Yeah, yeah, it's uh I was talking to uh uh a bank in Switzerland and they were using like an instant, like a I think they were using a ChatGPT version that was hosted in Switzerland, but the model they had access to was GPT 4.0. And I was like, it's crazy that you know, because it's secure, it's hosted. It's you know, they're basically falling behind by you know, I don't know how many months. GPT 4.0 feels like uh, you know, a relic from uh ancient uh Egypt. Uh but um so it's it's kind of crazy how the legacy and the governance and you know the licenses that your companies will give you will will set you for success, or possibly I mean, I don't want to say failure, but you can't you're very much uh you don't have so much control over that basically as an individual contributor and employee in a company.

SPEAKER_01 21:49

No, but uh many companies are moving on on that topic, and um of course it takes more time for uh a sector which is a with a strong um uh

SPEAKER_00 22:00

uh legacy uh it's uh quite difficult also for uh uh it's more difficult sometimes for big companies than smaller ones because they have a lot of uh constraints in terms of data of confidentiality etc but really it's moving on and um all companies today are working on the best stack for uh to be integrated in the um the the official and the historical i would say uh stack yeah uh and they do test uh they really are in test and learn approach uh to test new tools um so really i i see a strong movement since i would say six months on on that topic for sure yeah um if we tie this back maybe if we switch uh um gears and tie this back to the fact that you're doing consulting today i feel that today consulting is a little bit hard when you want to advise these companies because on the one hand you need to teach them about governance and about these tools and finding use cases and so on but at the same time you're telling people when you train them like okay you need to master these tools come up with their your own use cases you know we're saying that it's super easy to build um you know so there's like those conflicting forces between top down where you need to basically have the leadership to decide what to build and then you're also teaching people to basically fly on their own pick up new tools learn on their own build their own thing so how do you actually strike a balance between defining a strategy at the company level and having people to also build their own into internal productivity tools?

SPEAKER_01 23:34

It's true that it's a strong paradox you have to work with um and I think it was perhaps a bit the case already for digital but it's really um it has been completely enhanced with AI revolution because uh what is uh specific with AI is that the option the adoption was really easy and started with B2C. When uh ChatGPT was launched uh everyone wanted to test it to try it etc so the the the usage came from B2C and not from B2B and that's uh what is really changing the approach because um many people took chat GPT for their personal life and they they tried it for their business life even before their boss even before the XCOM uh even before uh the board etc etc so um it really um came from um the ground I would say from the function and not from uh the the the the XCOM or uh the CEO or the C level but that's really the change because and it's what we we all call uh shadow AI uh because uh all the people tried and they found it very efficient and uh so uh without any rule they tested and um so I think and I and for my side I think it's a fantastic opportunity to be honest it has to be uh of course um yeah of course companies need to have a vision uh they need to have a clear governance but they don't uh they they should not kill local initiatives because I think it's a fantastic opportunity that so many people uh have been uh empowered on so many uh in in so in such a short time on a new technology and have been enthusiastic because often you'll have some resistance you know on digital or on tech on tech sometimes you have resistance here you have a strong enthusiasm not from everyone but I would say uh from uh a good majority so it's a fantastic opportunity so for me what is important is that people are playing with AI now so you you cannot do anything uh with that but you can have a vision for your company uh so you have to think about uh where you want to put AI in your company the idea is uh perhaps not to put AI everywhere because you have a DNA you have a history you have a relationship with your clients so where you want to put AI where it's relevant it can be for productivity it can be for personalization uh it can be for um uh stronger uh reliability on figures etc etc so it's really where it will be relevant for companies to put AI so it's the first question you need to ask as a CEO and as a XCOM or as a board um and then you have to uh play with the governance uh which will um of course give some rules and I think that's really important because there are questions about data confidentiality and GDPR also which is a topic that sometimes we forget but uh it's completely linked with AI so it's also very important.

SPEAKER_00 27:08

So you have to give rules to be in line with your historical strategy and the way you manage data in your company but uh and you have to also uh organize I would say the transformation so taking one people who can be the AI transformation official so who will have the leadership on AI and then um you have to uh get some uh AI champions per function because thanks to thanks those AI champions you will be able to capitalize on local initiatives and that's why in your governance you need to have a clear vision clear rules but giving the opportunity from all functions to give their output and to bring some uh new ideas new opportunities new business case on AI yeah uh do you so if we get into specific do you have specific examples of how a company you know set this up in terms of governance did they just give licenses of you know cloud co-work to everyone and they said play around let's talk again in one week about the business cases that came up make sure you don't throw like emails of you know our clients in the in a cloud or you know how does that look like in practice?

SPEAKER_01 28:29

The first thing is how you you can um find your AI champion I would say um so um the the trainings are good moments to identify uh of course uh AI champion but you can ask also to people who would like uh to be part uh of uh who would like to to be uh an AI champion and I think it's always good also to ask people if there are some volunteers uh because uh people are even more um enthusiastic and uh able to give time when they are volunteers I think so that's the first point the second point is um that uh you um need to define their role precisely which can be different from one company to another but from what we see um often they have the role to uh first um um uh ask people uh how they use AI bring uh all the new uh initiatives or the new um use cases which have been built by their peers in the same team um and then they can have the role to build agents for the team because I think not everyone can build agents today at least perhaps tomorrow but today uh it's still um uh a competence that is not uh um uh embraced by everyone and it it's a more technical competence even if you don't need to be a tech uh for doing agents you need to be trained on how to do a good agents how to maintain it and how it works with all the other tools so it's a real competence so I think it's important to train your AI champion on how to make agents so that they will be the reference in the team to build new agents. People will come to them saying okay I would like to automatize this process how can I do it and the champion will be the one who will first work with the team to see which step of the process they want to automatize and then they they will build the agent for one person or for all the team so that everyone can benefit from this agent within the team.

SPEAKER_00 30:59

Yeah um you touched on something I think that is also super uh relevant um so building agents today is maybe not something that anyone can that everyone can do but maybe tomorrow so what's the one you know if we focus on specific workflows or use cases you know with the evolution now of like the LLMs in the past six months what's like one thing what's one use case that you have observed in the last six months that has become possible that wasn't that wasn't possible maybe uh you know one year ago um I would say uh to be um uh but first I see a lot of things around AI and creation for example so one year ago you were not perhaps two years ago should I say you were what able to build uh an uh image or a video with AI.

SPEAKER_01 31:54

So that's something uh really uh new for marketing team content content factories uh agencies etc so um the the the here you you can really uh build uh agents that will help you to create those images etc but um even more uh which will help you to uh adapt this image or this video to any can any um channel uh you are going to target meaning uh it can be for uh social media but for very uh it can be uh for meta or uh for uh LinkedIn uh or for TikTok so the the the formats the content the wording is not the same but today you can organize agent to really build a content planning uh that will go very quickly with a strong quality and uh very uh well adapted to each canal in a very short timing.

SPEAKER_00 33:02

Crazy yeah I think there's this uh post that uh came out recently about how cloud you know um or the growth team at Antropic uh basically I think they have one person in the growth team running and and running a bunch of ads uh with a bunch of uh cloud skills and workflows that they have implemented uh with integrations with Figma and so on is this uh is this what you see also in companies people connecting different tools together to come up with creatives but that's the the magi I would say of agents is that you are able to connect uh different tools um by prompting and that's really the the the the change of the game uh because uh you don't need to code anymore you can do it very easily and it goes very quickly by building these connections so um it's uh it's quite crazy and it's true that cloud cloud the the the uh all the new functionalities uh launched by Cloud since uh three weeks now uh notably on connecting uh you can do um much more things and build agents uh not very sophisticated but at least uh daily agents I would say yeah uh very much more easily and that's where it's really uh a game change uh for people who are non-tech uh uh or who don't have any uh technical historical expertise. Yeah I think it's um so I was playing around yesterday with uh Claude Cowork uh to try to edit my podcast automatically so it would go and I said okay shorten it by you know 20 minutes uh and so on and what I've realized is it's super powerful because it can connect to this platform called Riverside but then what it does is it takes screenshots of the text and it scroll through it's it's gonna scroll through everything and take screenshots, screenshots, screenshots and sometimes you always think that the systems are very sophisticated. But what happens under the hood is actually just like you said like a bunch of prompts taking screenshots and then it's feeding this to the LLM so there's this magical element but uh it's actually like yeah agents are actually rudimentary. The the real power is the LLM but it's just connecting all these integrations together that make them very powerful. Yes and it's um it's a question uh we all have I think is that um now that the LLM uh are working on orchestrating everything and also uh agents uh and integrating uh MCP uh etc and a connection um the question is uh how will llm work with um um make uh and it etc uh will those tools still be relevant yeah uh when llm will orchestrate everything yeah and that's the way they are working a lot on that part uh but today it's not uh uh it's still complementary for sure but it's a question for tomorrow yeah it's true uh I mean there is a whole thing around uh infrastructure to cut to tie this back to what you said earlier like basically when you build an agent you need to be able to maintain it um evaluate the results and there's also this part about uh you know infrastructure and A10 you can deploy it to the cloud whereas if it's cloud code it's you know locally so there are maybe like still a market for both today at least uh what do you think makes it hard to maintain an um you know an agent today uh and to be able to evaluate the output how does that work in in practice you know do you have a specific example on you know if we tie this back to the creative workflow that you mentioned um but I think you you need to be well equipped uh in um uh I think the the it's a question always of uh in um taking people internally uh doing agents or uh taking external expertise and I think that's a key question uh because if you train uh your own teams on agents meaning uh in functions in the the tech teams in the data teams I think um uh you can have uh uh a better maintenance on the long term because everyone uh works inside the company um and uh uh all the function teams the tech teams and the data teams are at the same level of competencies so they can really work together on the best integration in the current stack of tools and there will be much uh they will be really able to maintain it because they have built all the agents together and they can also uh write all the processes and document everything so that uh it's um something that will be maintained on the long term so I think that will be easier but it's not possible for all companies because you don't have always the money to uh train all your uh champion uh and all your tech and teams etc so sometimes it's it goes it goes quicker and faster uh to uh bring external uh competences on that notably for smaller company um so I think what is really important is to ask to uh the the people who are doing the agent for you the external people who are doing the agent for you to ensure they will be able to maintain yeah or at least at end potentially train internal people so that they can be able to maintain the agent uh on the long term yeah so well interesting yeah it's important to keep that in mind because uh there is always a question of maintenance yeah so it's true it's true um and it it's it's a bit tricky uh like you like you said uh depending on the skills and the department not everyone can can do it and people don't necessarily want to do it and have the time and so on. So if we think about you know what should be built internally versus uh have like consultants to implement there's this thing around budget and having the resources to basically be able to invest in teach in training people are there like specific use cases or specific projects that you think um could still benefit from having external parties or how do you think about these topics between um you know building internally versus having some external consultancies but I think it's uh again a tie a question a bit of uh company size because when you are a big company you are able you have people to betrayed first of all people can have a bit more time not always but well and um you you ensure and you have uh normally a more sophisticated stack of tools when you are a bigger company so uh I think uh I would really uh recommend that um on the long term they really train people inside to be able to have to to do agents and to maintain them they can start with uh external expertise but then they have to make it internal so well for the big group I think it's easier in that way for the smaller companies um sometimes they will uh be uh they will need to get quick external uh expertise and I think it's a good thing because they can really work from people who are doing that for many companies who have the vision of all the new functionalities the good tool and have more agility while um uh bringing new tools in pardon in bringing new tools uh within the company so um I think uh it will be uh easier for them to start with external approach for sure uh but then uh what is uh interesting is that they they choose few very few people in the company uh it could be in the tech team if they have one because it's not always the case um which uh will also get these competencies so uh that again when they stop one day uh the contract with the external people they are people internally able to maintain the agents but I think for a smaller company it will be easier to start externally yeah yeah for usually I have uh personally I have this rule of thumb where if they don't have basically you hire someone internally to do this full time only if you have like three projects in parallel because it's easy to build but you rely on people to give you feedback to improve the product and usually giving feedback having people to play around with the agent to evaluate the output and give feedback takes much longer it takes around a week. So basically you like the developer is going to build for one day wait one week for the results and so basically he's working on it at like 20% capacity. So unless you have this roadmap of you know a bunch of agents you want to implement it's not really profitable uh to to have you know someone full time to do that.

SPEAKER_01 42:44

Yes I completely agree.

SPEAKER_00 42:46

Yeah. If we switch the if we connect this to your company today, um Catalyst AI, I know that we've talked about this before the the the the call you have a lot of exciting workflows. Should we maybe go over uh one of them and uh and you can walk us through how you came up with the idea and why you actually built it you know because we're teaching companies how to do this but it's interesting to also hear your understanding on you know why did you decide to build this over something else the challenges along the way and how it's helping you today uh yes so I um we are um we have started this company one year ago now so we are a young company I would say um it was uh important for us to um uh really um be able to automate automatize many things because we are not so numerous you know inside the the company uh and we thought um where and and the the way we thought and we chose the processes we wanted to optimize was where um it would be a game changer for um really the the acceleration and the growth

SPEAKER_01 44:00

I would say one side. And on the other side, it was how we can really uh increase for decrease the the tasks which have no added value at all. And that we did as the co-founders, you know, and as many entrepreneurs. And how we can uh change um those tasks uh how we can uh gain time on those tasks uh for bringing more added value on uh the way we want to uh develop our companies, etc. etc. So for us, it was not a question of productivity as we are, it's kind of, but we don't have people, you know. We were a small company with few people, so it was not a question of uh how we could uh reduce potentially the productivity, but it was really how it will help the growth on one side and um give us uh more time on a uh task with added value. So that's why we choose on one hand how we can really accelerate and optimize our um B2B prospects uh lead generation. So here uh we work on a workflow to uh scrap uh data from uh databases uh which are very qualified, for example, uh LinkedIn says Navigator. Uh how we use those data um to uh bring uh very targeted prospects in line with what we do, uh very qualified also with very quite precise information to um uh uh uh reinforce uh the quality of our uh SRA. So we have uh we are working with UpSpot so to really enrich uh the database in uh UbSpot, and uh we are able to uh increase the number of contacts but also the quality of contacts uh every week because it's automized now and it looks for uh it's crap in the initial database, external database, to increase to um uh reinforce our uh CRM and to be able then to target uh all those new prospects, much more qualified, uh in uh through different channels. It can be uh it can be uh emailing, newsletter, but also um calls, etc. etc. So uh that's the first uh automated process uh which works quite well and help really the the the targeting and the uh the the way uh we work on prospection and it accelerates uh things. Yeah, uh it's well again it accelerates, but it also increases the accuracy and uh the the targeting is much better. Uh on the other side, uh we work to automatize the um administrative process because in France uh the the the training pro the procedure uh around trainings are very very heavy because uh it's linked to um stay uh it's linked to certification given by the state, so it's very heavy. You have a lot of uh uh steps uh to follow, a lot of documents to send, a lot of questionnaire to send, etc. etc. So um we worked here to uh really reduce the time we spent on administrative tasks to reallocate this time on uh uh much more uh added value uh tasks. And uh here we uh really automatize uh all after the client has said okay, go we do the training, then we automatize all the rest of the process, meaning uh sending the invitation, uh sending uh the common the contract, uh, sending uh the questionnaire, uh sending um all the elements they need to know before the trainings, etc. etc.

SPEAKER_00 48:15

Amazing.

SPEAKER_01 48:16

So and it's really a game change.

SPEAKER_00 48:18

Yeah. Yeah, I I think France is one of the countries that will benefit the most from uh Asians to free ourselves of uh of uh paperwork. Um what's what are the tools, if I can ask? Uh you know, what are you know how does it work under the hood? Maybe if we get into like a couple of specifics, you know, on how did you connect to send out emails, you know, at what point does it know? How did you think about uh setting this up without sending the wrong thing to the wrong client?

SPEAKER_01 48:45

Uh not sure to understand the question. Sorry, about the tools we choose, the stack you choose.

SPEAKER_00 48:50

So so for this specific example of the questionnaire, basically like you have the client agreeing by email, and then you're like, okay, you know, let's you know, how does it look? Like you're gonna press submit and it's basically like a full workflow that is gonna run. You know, is this like on cloud uh co-work? How does it look like?

SPEAKER_01 49:06

Wait, kind of. Um, so we we have um uh a formular that uh we are filling in with the help of the clients, taking oh it's a Google form to be a bit simple, uh in which we uh bring all the data we need to know. So the name of the clients, uh the the the address of invoicing, uh the number of participants, the name of the participants, um the the the type of trainings they choose, the date, uh etc etc etc. So we fill in this Google form. And after that, we have uh created connectors um to uh ensure the sending of emails at the good timing with the good document. But to do that, and I think it's really important because that's the rule for any company, you have to have very clear process, so you have really to um uh detail uh uh micro task by micro task your process and then you have to uh have to to uh to be if you don't have them, you have to build them, but sometimes you already have them, you need to get very um standardized uh templates because if you want uh to uh ensure that the documents uh will be well filled in, it has to be perfectly uh it has to be very uh clear in terms of templates because you will send uh the contract is always the same, the questionnaire is always the same, the um you you need to have very uh uh standardized templates in order to be able to automize. Because uh, if you don't have that, you you you cannot ensure that people will receive the good documents, so you need to be really clear, and you have to to ensure that you are perfectly in line with your contract before uh launching. Uh it's a good occasion to review, you know, yeah your documents, your templates, your contracts, because when you push on the button, it goes to uh many many people who are going to sign the document. So it's a good opportunity to review your process to simplify sometimes also, and to um make very clear templates uh uh in with which uh you are perfectly uh at ease because you will not have the occasion to review it before sending. So I think that's something very important, and then we have tested uh, of course, on a few uh customers uh before launching that to everything.

SPEAKER_00 51:46

Like please don't sign, this is just a test.

SPEAKER_01 51:48

And there are always bugs, for example, on the questionnaire, we we didn't get uh we didn't know where the results were, so sometimes we have some surprises, you know. Uh but finally we found them. But voil. Uh it's also very important to uh structure well where all your documents will be located when they come back from the clients, and um very uh important uh to uh ensure that uh it uh all your emails have been also reviewed, etc. Because sometimes you you see all emails and you forget one that you don't reread before sending, and you see, oh my god, it was not the good.

SPEAKER_00 52:30

We committed to a five-year agreement.

SPEAKER_01 52:32

Uh exactly. So it's a question of uh it takes time, honestly, but when it works, it's fantastic.

SPEAKER_02 52:40

Yeah.

SPEAKER_01 52:40

Uh yeah, it's people think uh it doesn't it doesn't take time, it takes time because you have to review the processes, you have to have the good templates, you have to test to resolve bugs, and then you can deploy. It goes much quicker than any uh digital project uh before, but it's still time uh because it forces you to rethink the way you work. Yeah, the process, the templates, and it's a fantastic opportunity, yeah, honestly. But it takes their time.

SPEAKER_00 53:10

Yeah. Yeah, I mean I think it's um I read somewhere. So Harvey, I don't know if you've heard of Harvey, but it's uh a legal AI tool.

SPEAKER_01 53:18

Uh I heard about it.

SPEAKER_00 53:20

I think there's a say yeah, I think there are a couple of companies like uh doing really well in that in that space because it's a lot of documentation as a lawyer and so on. And I think they're the the whole bet investors are investing in this company because they believe that um it's not only the tool, it's basically at some point you're gonna become a lawyer who's like very good at mastering AI. And I think coming back to your example of the contract, I think you need to be a good lawyer or be good with contracts to know what you can remove, to like simplify, you know, what should be kept, what should you watch out for, and then understand how the LLM works so that you can actually populate it with the right field. So there are a lot of moving parts, and it sounds easy because technically you could just put a template in ChatGPT and be like update it, and it's gonna feel like it worked. But when it comes to actually having someone to sign like an agreement that you're gonna commit to for like several years, then you have to really go through all that like tedious exercise. And this is where I think the functional expertise of being a legal, paralegal, or you know, jurist or you know, lawyer or whatever, uh and being able to master like AI tools is really gonna bring the best of both.

SPEAKER_01 54:24

Exactly. And uh, if you are not mastering your function, you will never be good in uh integrating AI in your function. And I think uh that's where it's important to keep this function expertise. And uh even for young people, uh they need to continue to learn the functional expertise, the soft skills. Even if of course the ways of working will change, but they still need to bring uh they they will need to be able to check. So if you don't have the expertise, you are not able to check anything.

SPEAKER_00 55:04

Yeah, yeah, I I I think it's uh you probably saw that in France. There's this teacher who gave an exam, like at uh at one uh at home exercise, and everyone got like amazing grades, and then uh the teacher realized that it was impossible, they had like an 18 out of 20 average, and then he asked them to take the exam again in class, and the average was like five. And obviously, people had used LLMs, you know, uh like ChatGPT at home. So I think like education uh as a whole, we need to rethink, but yeah, I I think it's very hard to tell where this is going, but it's it has to be yeah, reimagined, that's for sure.

SPEAKER_01 55:36

I for sure. And it's a big topic that people uh and states need really to take as a as a as a strategic uh issue to tackle. Uh because um of course people need to learn differently, but they still need to develop their cognitive uh competences. Uh and it's something that you learn uh at school, at junior. So um it's how we can uh bring AI in education uh while keeping uh working on uh many uh cognitive uh ways of thinking and cognitive competences. And um you think well you need to rethink everything from uh the way to teach, um the way you evaluate, uh the way uh you um say, okay, on this topic no AI. On this topic, yes, but not on this one, because I want you uh uh to be able to develop uh something on your own. Uh so uh develop your curiosity because I think AI is a fantastic way to uh develop your curiosity if you use it well, okay. Uh if you use it, uh it can be even addictive for many people uh because they're so curious that they never stop, you know, uh trying new things with AI. But if you are not curious, I think it's a drama.

SPEAKER_03 57:07

Interesting, yes.

SPEAKER_01 57:08

Yeah, it's uh new, it's I think to to really uh work um on uh it's really to change the way we learn um and again finding where we want to keep AI because it will it will be the new way of uh working and and in people and and young people need also to learn how to work with AI, but where we think uh we should not uh uh use AI because uh it's something it's topics that people need to think on on their own.

SPEAKER_00 57:43

Yeah.

SPEAKER_01 57:44

And to develop their own thoughts, and I think that's something uh we need to really work on.

SPEAKER_00 57:51

I think I mean we both grew up probably uh doing presentations where we use Wikipedia, and already back then, you know, we're like, okay, we need to question the source and you know do more research. So it's gonna be this times uh 10,000. Question everything, uh do the research, and uh, but yeah, I think those the critical skills, whenever you do research, uh will need to stay.

SPEAKER_01 58:12

Of course.

SPEAKER_00 58:13

Um maybe just one last question because I know we uh were running over time. Um how did you learn all these tools? Because you were you know a marketeer, uh, you know, you were uh running uh operations and departments in in much bigger organizations. So how did you pick up these tools? And you know, do you have any other tips and recommendations to give to people who were in your position uh you know two years ago, let's say?

SPEAKER_01 58:35

Yes, no, um I started to work on AI tools uh when uh OpenAI uh arrived, and um and I started to to and I wanted to understand uh the way it would change, so I I tested first with OpenAI, as many uh people uh started uh with. Um and I was still the managing director of uh Alain Ducasse school, and there uh I had the intuition that um there could be uh interesting use cases for the chefs we were at training, uh pastry chef, culinary chef, and um uh I wanted to test it to understand which kind of use case could be relevant and if it would be uh useful to train them also on AI. So that was really the starting point of my AI journey testing if the tool could help uh the chef on uh building a PL, on uh how uh they could um calculate their carbon footprint footprint on how uh they could be better in uh marketing and content, on how they could uh source with um more information available. So and when I tested it, I saw the huge number of use cases for a function which is very far away from technology, yeah, and I thought to myself uh it's a revolution. So, first we are going to teach uh chef on AI because they will gain time on tasks on which uh they don't take so much pleasure. Where they take pleasure is on how they can create fantastic menus, how they can um uh have an excellent relationship with their clients, how they can manage better. So that's where they bring added value, and that's what they love to do. And all the other things they do at night, you know, because they work uh day and night, uh the test, yeah. Uh perhaps they could gain time and um and uh take more time for the other field they really love. So that's where that's and that's why I really started first to test myself to be sure that uh it will be relevant for them. And then uh after that, I develop a passion for AI. So uh I learned, I look, uh, train myself. Um uh I because as I wanted to launch this project, I had to be also very aware of what we could do, what are the concrete use cases, uh, because I'm always obsessed by the results, you know. Uh we are training the impact impactful, so I really wanted to understand how AI could be uh really helpful for the different functions, and I wanted to test that myself before uh launching the company. Yeah, uh, and when I saw all the use cases, I was completely convinced, and that's why I launched Catalyst. And um, to be honest, it's a question of curiosity, I think, of uh trying and uh uh accepting to uh not have the good answer uh in the first uh uh step, but accepting to iterate, and I think it's a way of work, uh and uh last but not least, uh we have the chance to have a uh very large community, as we were saying before, and they nourish us a lot also on the way they see the changes, the new functionalities. So, on top of all uh the um the news we receive are that uh of course uh we also have the chance to work with people who are uh uh working with the eye on a daily basis for their function. So they are the best to give us also uh many insights um and to give us the envy to test uh new functionalities.

SPEAKER_00 1:02:44

Yeah, interesting. Yes, that's I think this is why I'm doing this podcast also, is to get ideas from others because I'm running out of inspiration and uh I love uh hearing from uh from others. Um thank you so much, Elise. Uh, this was amazing. Um uh I think we covered a lot of use cases and uh and good insights into how to drive initiatives, uh, finding champions, driving from the top to give licenses and uh keeping people excited to try out and stay curious. Is there anywhere people can like follow you and like uh learn more about you uh besides LinkedIn? I will be sharing your LinkedIn profile and the name of your company. Is there anywhere else you share some thoughts or any specific things?

SPEAKER_01 1:03:23

I think the Glim Gene remains the best channel, and of course our website, uh which is um catalyst.ai Academy, also.

SPEAKER_00 1:03:32

Yeah. Amazing. I will share it here. This is the website, so please give everyone a follow to Catalyst Academy and to Elise. Thank you so much, Elise, for your time, and uh we'll be in touch.

SPEAKER_01 1:03:45

Yes, thank you, Vlad.

SPEAKER_00 1:03:48

Thank you so much for listening to this podcast. This is Business AI Explained, and I am Vlad. I am the founder of Elements Agents, and we build custom AI applications. If you do need support to implement custom AI solutions, you can book a session with me below. Or otherwise, if you want to hear from more AI experts and execs implementing AI in their companies, come back next Tuesday for a new episode. Ciao.