Your AI projects are bound to fail - address these common mistakes - EP 018

00:00:01: Edgar: Hello and welcome a new year and new season and a new

00:00:06: episode of the AI Boardroom.

00:00:09: And welcome back.

00:00:10: Thanks for tuning in.

00:00:11: Of course with me, my wonderful co-host, Atlanta.

00:00:16: Svetlana: Hello.

00:00:16: Welcome back.

00:00:17: Uh, happy to be here for season two.

00:00:20: Uh, of our podcast series.

00:00:22: Yeah.

00:00:22: So

00:00:23: Edgar: yeah, we, we tried, um, we tried to improve.

00:00:26: We have some, some new stuff for, for you in the pipeline, um, starting today with

00:00:31: a very special topic that's near and dear to your heart for Anna, as far as I know.

00:00:35: Svetlana: Yes.

00:00:35: Uh, AI risks.

00:00:37: I feel like that's one of the hottest topics right now as more and more

00:00:41: companies are implementing it.

00:00:42: And ai, um, is a bit riskier, um, to deal with, uh, which we'll talk about

00:00:49: seven key sources of risks, um, that any company should really pay attention

00:00:54: to as they're kind of taking on AI projects into their organization.

00:00:58: So.

00:01:00: Yeah, this is, uh, a really, I think, timely topic as, as more

00:01:04: organizations want that positive ROI, but they also wanna mitigate risks.

00:01:09: Edgar: So, um, but it's not, it's not only like talking about the risks, it's,

00:01:12: this episode is about risk assessment.

00:01:14: Right.

00:01:15: Svetlana: We have a guide, which we'll get to.

00:01:17: Exactly.

00:01:18: Well, I think we'll talk about the seven topics and how do you actually, what seven

00:01:23: areas of risk could AI be introducing and.

00:01:26: Those ultimately are sources of, or points of failure in the system.

00:01:30: Right.

00:01:31: So, and we do have a systematic way of kind of laying out the land of

00:01:36: the land of, of those seven risks.

00:01:38: And I think there's flexibility to add additional ones and

00:01:41: really understand which.

00:01:43: Risks do you tackle first, second, last, and then how critical are those to

00:01:48: the success of your AI implementation?

00:01:50: So we do have a systematic approach.

00:01:52: There is a grid, there's a system mm-hmm.

00:01:54: Behind that as well.

00:01:55: But I feel like we need to get into the seven.

00:01:58: Yes, yes, yes.

00:01:58: Areas first.

00:01:59: Edgar: Yeah.

00:01:59: Just, uh, just wanted, uh, everyone to know when you finish this episode,

00:02:03: you should be more aware about.

00:02:05: What to look for in the first place.

00:02:07: Svetlana: Okay, let's get going.

00:02:08: Um, so I think it's important.

00:02:10: So what, why, um, I think to kick things off, why are.

00:02:15: AI risks different than traditional software development.

00:02:19: Um, one of the key areas that I, I feel like it's, um.

00:02:24: It comes to data.

00:02:25: So in traditional software development, you are kind of, um, creating rules

00:02:31: or rule-based logics, and you're kind of validating those algorithms.

00:02:34: You have kind of checks, checks and balances along the way before you

00:02:37: actually deploy things into production.

00:02:39: But with AI, you're dealing with data.

00:02:42: And so not always do you get the clean data.

00:02:45: Um.

00:02:46: How your algorithm algorithms are actually getting the data.

00:02:49: And then what the especially large language models do

00:02:52: with that data is also, um.

00:02:55: You know, could lead to different consequences, right?

00:02:57: So such as hallucinations, maybe retrieving the wrong

00:03:00: information and things like that.

00:03:01: So you can't always control for all kinds of use cases and things that

00:03:06: could, could potentially go wrong.

00:03:07: So, um, AI is just different than the way that you would manage those projects, but

00:03:12: also how you are kind of evaluating risks.

00:03:14: 'cause they are traditionally different from.

00:03:17: Um, traditional software development.

00:03:19: You also, um, kind of, there's a, a source of, I would say, like AI

00:03:24: hype, um, with, with AI projects.

00:03:26: So sometimes people tackle projects or introduce unnecessary risk with

00:03:30: AI into solutions that, uh, should not be solved with artificial

00:03:34: intelligence in the first place.

00:03:36: Um, they're just too simple of tasks.

00:03:38: Um, that could have been solved more simply with traditional programming.

00:03:42: Than ai.

00:03:43: So it's just, you know, misfits sometimes, you know, you're

00:03:48: introducing new risk because of, uh, wrongly applied, you know, uh.

00:03:53: Techniques to the problems, but then also, you know, the types of models

00:03:57: and how they're dealing with data and then how that data is transmitted.

00:04:00: You're also dealing with tools or API calls, you know, dealing

00:04:04: over the public internet.

00:04:05: You know, you're, you're talking about, you know, these large language

00:04:08: foundational models being used as, you know, APIs and things like that.

00:04:12: So.

00:04:13: For larger enterprises who are in especially large, um, highly

00:04:17: regulated spaces, this becomes also, um, a key consideration.

00:04:22: And I, I would say one of the reasons why they take a little

00:04:25: bit longer to, to deploy some of their solutions because they are

00:04:29: checking all of these potential new.

00:04:31: Introductions of risks.

00:04:33: Um, and some of them I feel like I, I have spoken to actually one executive in

00:04:37: the past from a big, you know, fortune 500 company, and he said, you know, it's,

00:04:41: it's not the risks that I'm afraid of.

00:04:43: It's what risks that I don't know.

00:04:45: Mm-hmm.

00:04:46: Uh, that is the biggest fear that I have in my mind.

00:04:50: So I,

00:04:50: Edgar: I I will, I would love to ask a question.

00:04:52: So,

00:04:53: Svetlana: yeah.

00:04:55: Edgar: Before we talk about the specifics, like mm-hmm.

00:04:58: Who, who is this for?

00:04:59: Like who, whom might I be for me to, for this to be relevant?

00:05:06: Svetlana: Yeah, I think that's a great question.

00:05:08: So if you are an executive, um, in a tech company who is undertaking,

00:05:13: um, you know, technical projects with artificial intelligence, so

00:05:17: if you are or you've attempted.

00:05:20: Uh, artificial intelligence before, maybe you've, you've done chat

00:05:23: bots and it's just not sticking, you're not really sure why.

00:05:28: Um, or how, and what points of failures actually like led to your,

00:05:32: um, kind of the following or the stalling of those projects and how

00:05:36: you can proactively mitigate those?

00:05:38: I would say it's for technical.

00:05:39: Founders, their teams or executives that are still exploring, um, kind

00:05:44: of the artificial intelligence space.

00:05:45: And they want to understand proactively what things do they need to look out

00:05:50: for when they take on these AI projects.

00:05:52: And do we have the tools under our belt?

00:05:55: Do we have the skill sets to cover for those risks?

00:05:57: Or is, are these areas that I need help with?

00:06:00: Um, I feel like more and more companies in the technical realm.

00:06:04: You know, even if you're dealing mostly traditional software

00:06:06: development are taking on, you know, um, these AI agents or introducing

00:06:12: AI agents into their workflows.

00:06:13: So this is becoming more and more of a common, I would say, undertaking, but

00:06:19: again, it's like what you don't know, what you don't know type of thing.

00:06:21: Um, and this is where we're hoping to close that gap for you to kind

00:06:25: of have a more structured approach for how you think about risk.

00:06:28: Edgar: Yeah, it's, it's, uh, especially when you think about like actual agent,

00:06:33: uh, applications or genetic applications, they grow in complexity like.

00:06:38: Quite fast.

00:06:39: Um, so, uh, yeah, there's a lot of, um, a lot of stuff to talk about and talk to.

00:06:44: I also have some, uh, uh, had some anecdotes in my mind when

00:06:48: I first read the, the topics.

00:06:50: Uh, so yeah.

00:06:52: Um, I would say let's, uh, let's dive in.

00:06:54: Could you give us an overview, um, what we are looking at?

00:06:58: Svetlana: Yes.

00:06:59: So, um, the framework that we're gonna go over there are seven core.

00:07:04: Areas of risks.

00:07:05: And this comes from, you know, uh, experience, uh, research and, you

00:07:10: know, lots of, kind of big hours being actually in the field of implementing

00:07:14: ai, AI projects that we've kind of consolidated it to seven areas.

00:07:19: Think of it as like a framework of seven different areas.

00:07:21: What could be.

00:07:23: Again, uh, a point a failure into your AI implementation.

00:07:27: So we call it the Clear seven AI Risk Framework.

00:07:32: Um, and it covers, again, seven areas of different kind of parts

00:07:36: of your AI implementation that you need to pay, you know, kind of.

00:07:40: Pay closer attention to, um, and we'll get into it.

00:07:43: I don't wanna kind of, uh, allude or give away too much.

00:07:46: Um, but I think as we kind of walk through it, I think you'll

00:07:49: find it very, very interesting.

00:07:51: And I think hopefully it'll start resonating in, in the idea, or the hope

00:07:55: is, is that, you know, as we kind of go through it, you'll kind of get the ah,

00:07:59: oh, was that what, what was happening?

00:08:01: Um, I've, I've, I've seen that.

00:08:03: So, um.

00:08:05: And we'll have some mitigation strategies.

00:08:07: Yeah.

00:08:07: Let's get, let's get started.

00:08:10: So the first, um, area of risk is, uh, feasibility.

00:08:14: So, and the gist of this question, or the gist of this kind of, uh,

00:08:19: risk, uh, area is can we actually build what we're undertaking?

00:08:24: So is there sufficient data to actually take on this project?

00:08:29: Do we have the infrastructure, do we have the actual, you know, the necessary

00:08:33: compute in our cloud environment to actually support the, these models?

00:08:39: And then, um, are we just kind of.

00:08:42: Ready to, to take on this project.

00:08:44: Do we know the tools in the market to actually orchestrate

00:08:48: the relevant solution for the objective that we're hoping to solve?

00:08:52: Edgar: Yeah.

00:08:53: Level.

00:08:54: This is also, this is also like different depth to it, but at the basis is like.

00:08:59: Data.

00:09:00: I think it's the biggest point of that infrastructure.

00:09:02: Um, tech, uh, and technology of course is one, two, especially technology like

00:09:07: are the models that we have access to are capable to do enough or good enough?

00:09:14: So they don't have to be perfect, but they have to be good enough.

00:09:17: Um, and of course, data is like the main point.

00:09:21: I always give this, um, idea of like, I want to build a marketing tool.

00:09:26: Um, mm-hmm.

00:09:26: And this marketing tool should, uh, like give me at.

00:09:30: Copy.

00:09:31: Yeah.

00:09:31: If I don't have proper ad copy to feed in as an example of a good app copy, it

00:09:37: will be generic stuff that's unusable.

00:09:39: So, um, yeah, do I have the data?

00:09:42: If not, maybe I don't build a marketing tool.

00:09:46: Svetlana: And what I've seen also happen is that, you know, you, um.

00:09:48: Executives might have these moonshot ideas.

00:09:51: So I think part of like also the engineering side is that you, I have

00:09:54: the data, but it's a moonshot idea.

00:09:57: You're looking for something that is like completely autonomous ic, you're

00:10:00: thinking a GI levels type of stuff.

00:10:03: It's just not fricking feasible.

00:10:04: Um, yeah.

00:10:05: Or you will have to, you know, for the type of use case, the return on

00:10:11: investment is just gonna be so minimal that it's like, why would we actually.

00:10:15: Do this, but we'll get into the, um, kind of that, that side of

00:10:20: it, the viability part of it.

00:10:21: But it sometimes it's just like a moonshot idea that I don't think that the

00:10:25: technology of today can actually solve it.

00:10:27: So that's part of the feasibility and can we actually build

00:10:32: what we're aiming to build?

00:10:34: Edgar: Yeah.

00:10:34: Another thing is, um, we also have like trade offs and these trade

00:10:38: offs, like sometimes you have to, um.

00:10:42: Like lose some speed for more precision.

00:10:45: Mm-hmm.

00:10:46: Um, you could also find that, yeah, just from my rough calculation, the

00:10:50: costs are already like, barely feasible.

00:10:55: Um, if, if I, if I go even deeper and, and if you also like just

00:11:00: add some buffer, you might end up in a place where you lose money.

00:11:04: So, um, and that's, it's like, for example, um.

00:11:08: AI coding is like big.

00:11:10: There's a lot of costs.

00:11:11: Like I pay several a hundred bucks a month for the influence for AI

00:11:15: coding, which for me, it's worth it.

00:11:19: Mm-hmm.

00:11:19: But it's just like, there can be substantial costs, even

00:11:22: on a small scale already.

00:11:24: Uh, you have to, you have to consider that.

00:11:26: Svetlana: Yes.

00:11:27: And I would say also, I think, we'll, we'll get to it, but also part of the

00:11:31: feasibility is that, um, especially with reasoning use cases that require

00:11:37: multi-steps to actually, you know, get to the output that you're actually desiring.

00:11:41: Um, you're treating latency.

00:11:44: Of the responses in order to get higher quality of output,

00:11:47: right there is a trade off.

00:11:48: So it's a bit unrealistic.

00:11:50: Again, can we actually fulfill, is it feasible for us to do

00:11:53: sub-second response times?

00:11:54: Sometimes it is not.

00:11:56: Um, just because of the, it's not a dependency on how quickly

00:12:00: your infrastructure moves or how quickly, um, or how skills

00:12:04: skilled your internal team is.

00:12:06: It's just that the inference of these.

00:12:08: Models is slower because in order for for them to optimize for higher accuracy

00:12:13: of output, you are sacrificing latency.

00:12:16: So again, those are the things that you have to look for.

00:12:18: JJ,

00:12:19: Edgar: J, just, just in case you dunno, inference.

00:12:22: Inference is the process of like actually where, um, getting output from the model.

00:12:27: Svetlana: Yes.

00:12:27: And latency is, is how long does it take to actually, um, get the output or

00:12:32: get the response from, from the system.

00:12:34: So yeah.

00:12:35: Okay.

00:12:36: Edgar: Thank you.

00:12:37: That's feasibility.

00:12:37: What, what's, what's the next one on the list?

00:12:40: Svetlana: Yeah.

00:12:40: The next area of risk is desirability risk.

00:12:43: Right.

00:12:44: So will they actually use it if we build it?

00:12:46: So if, um, I. I would say if, if you've built something, you know, it's, it's a

00:12:52: shiny object syndrome and everyone else in the industry has it and you're like,

00:12:57: oh, I think it'll be really cool if we introduced it into our organization.

00:13:01: Um, and the use case you, you know, that you saw was in, in the marketing

00:13:05: agency and you work in the industrial area, and that's not the main.

00:13:10: Business of you.

00:13:12: Yes.

00:13:12: It's very cool.

00:13:13: And it works for them.

00:13:14: It doesn't mean that it actually works for your organization.

00:13:16: So just because it's feasible and then you're able to actually build

00:13:20: it, the, it doesn't mean that it's gonna be actually adopted or, uh,

00:13:24: it's desirable to end your end users.

00:13:26: And so when you're actually prioritizing use cases, it's, um, desirability

00:13:31: needs to be high because, and the reason why it's important is because.

00:13:35: Um, you want more iterations, you want more use of these AI systems for

00:13:40: the, for them to improve over time.

00:13:42: So if you're building something that doesn't actually get used, you're

00:13:47: kind of wasting away, um, kind of the investment that you've put into, bring

00:13:52: that solution in the first place.

00:13:53: So desirability needs to be there in order for, um, for you to align.

00:13:59: To the people who are screaming, we need this tool because you, if you

00:14:02: have hungry consumers, um, and you give them the tool, you're, you have

00:14:07: higher chances they'll actually use it.

00:14:09: Edgar: As a lot consumers also, like if you have users internally, if you wanna

00:14:12: build up an internal tool or something, or automate something, uh, like if the

00:14:16: people don't have trust, which is a big part of acceptance and inability, um,

00:14:22: um, that, that you think about how to build trust, how to build understanding

00:14:25: for this, um, how to also deliver the um, uh, user experience as a whole.

00:14:31: Uh, which is AI is just a small part of it.

00:14:34: Mm-hmm.

00:14:34: If you build an AI solution, like the integration around it is

00:14:37: basically 95% of what you have to do and what you spend your time with.

00:14:42: And then you have this intelligent engine in the middle, like triggering

00:14:47: stuff, but you still have to do all the integrations, which is just like,

00:14:51: I, I had a workshop with a company, they built a complete solution.

00:14:56: Without any wreck or something because they spent most of

00:14:59: their time integrating properly.

00:15:01: And then I ca I came in, helped them with the proper advanced AI techniques,

00:15:06: and then they were able to, uh, quickly integrate them because they

00:15:09: did a lot of work beforehand, so.

00:15:11: Hmm.

00:15:12: Um, so, yeah.

00:15:13: Um, uh, yeah, that's like UX integration is a big part.

00:15:18: Therefore, only invest your time if you really think you

00:15:20: can get a lot of value from it.

00:15:22: Svetlana: Yes.

00:15:23: And I think, um, we'll probably have an episode some sometime in the future for

00:15:27: how do you develop and what are different ways of integrating AI into workflows.

00:15:32: And I, it's a deep passion of mine to talk about the project,

00:15:36: uh, this, this type of topic.

00:15:37: So, uh, would love to kind of dive into that as a, almost

00:15:41: like a standalone episode.

00:15:42: But one of the things that I did wanna share, 'cause I think we'll,

00:15:46: we'll continue down the list, but.

00:15:47: This resource that of the Clear seven framework is gonna

00:15:51: be available to our audience.

00:15:53: I wanna make sure that, you know, for folks who are kind of new to ai, um,

00:15:58: kind of risk assessment, monitoring, just their undertaking this for the first

00:16:02: time, we've created a 22 page or 23 page document, um, that covers this at Link.

00:16:09: So everything that we're covering in today's episode and more, um.

00:16:12: And I just wanna make sure that we disseminate that knowledge.

00:16:16: We've created it for you to accompany the, uh, the episode.

00:16:20: So we'll link it in the description on the YouTube channel.

00:16:23: And how would they find it on, on the, for Pots, people listening.

00:16:27: Uh,

00:16:28: Edgar: you either, you either, um.

00:16:30: Either have to visit spark change, or you have to, um, look in the description.

00:16:35: I, I'll try to put the link into the description.

00:16:37: Otherwise, um, um, I would just, uh, um, say, yeah, check out the YouTube channel.

00:16:43: Uh, we will still, we're still working on the, on the podcast side, but, um,

00:16:47: yeah, if you're listening on Spotify or Apple, um, uh, podcast or something,

00:16:51: we'll figure something out for you.

00:16:52: Uh, I'll put this in the description.

00:16:54: Whatever we figure out.

00:16:56: Svetlana: Yeah.

00:16:57: Maybe if you could message us, we'll, we'll give you, we'll, we'll

00:17:00: drop you a link, uh, for sure.

00:17:02: Of course.

00:17:02: So, um, but yeah, let's, let's keep going.

00:17:05: Maybe, um, yeah, number number three.

00:17:08: Edgar: So we, we had visibility, desirability, and now we are at.

00:17:12: Svetlana: Viability.

00:17:13: So this is, and I think hopefully, uh, folks who are, who've been

00:17:17: in the, the tech industry, this, the first three are not new.

00:17:21: I think we'll get into the unique, unique ones, but it's the DVF framework.

00:17:25: So the third one is, um, viability.

00:17:28: So can we actually afford it?

00:17:30: So part of being able to.

00:17:33: Standup AI solutions is not just the initial takes and a lot of vendors

00:17:38: offer two week or sometimes months of demo or like, you know, kind of

00:17:42: free credits for you to use it, and then they get it into production.

00:17:46: And they realized how expensive it it is to run these solutions, and then they

00:17:50: continued to scale it with more users and those hidden costs, um, basically start

00:17:56: to rub them off the ROI they were hoping.

00:17:59: So, um, part of it is like, again, viability is the cost

00:18:03: effectiveness, but also.

00:18:05: Compliance and regulation, right?

00:18:07: So how do you, is is it even compliant for us to, um, take a use case, uh, with

00:18:13: artificial intelligence one use case?

00:18:15: 'cause I, I do work in, um, you know, work on, on in the highly regulated spaces,

00:18:21: but can you actually give, give advice, uh, and let's say critical settings

00:18:27: and then delegate that task to an LLM, like spec specific to your industry.

00:18:31: So if you think construction, I don't know, you're frontline worker.

00:18:34: Maybe it's military, I don't know.

00:18:36: But there are gonna be certain guidelines and regulations and complaints, uh,

00:18:40: things that you have to hit before you can actually hit production on these systems.

00:18:44: And so sometimes it gets overlooked.

00:18:46: There's lawsuits and all kinds of things, or, you know, there's certain,

00:18:50: uh, guardrails that are not put in place before things get launched

00:18:54: and things like that, that, you know, can we actually sustain this?

00:18:57: Or is this gonna be a very short.

00:18:59: Um, kind of undertaking.

00:19:00: There's also related to guardrails is bias and ethics.

00:19:03: Um, so I think a good use case or maybe example of this was grok.

00:19:08: 'cause they, I think, raced, and correct me if I'm wrong, I know that Edgar,

00:19:12: you're good on staying maybe with the, the latest kind of the timelines.

00:19:16: But I think gr the, you know, by Elon.

00:19:19: Was created within like two or three week timeframes, so they didn't put all

00:19:23: of the appropriate guardrails in place.

00:19:25: And so when they launched it, you know, people were trying to

00:19:28: take advantage of the system.

00:19:29: And of course, like it gave biased outputs and then it was

00:19:33: giving them unethical responses.

00:19:35: And so before.

00:19:37: They actually like checked all of those things.

00:19:39: Of course, they, they didn't take the system down, but yeah.

00:19:43: Um, not everyone can afford to do it.

00:19:45: Edgar: GR is also kind of like grinding on that part that they're kind of open.

00:19:50: Mm. Um, um, yeah.

00:19:52: But um, besides that, yeah, ethics and biases, like the

00:19:56: thing is trained on human data.

00:19:58: Human data is biased.

00:20:00: By nature.

00:20:01: Mm-hmm.

00:20:02: And, um, so you have to, you have to deal with it.

00:20:06: You can deal with it.

00:20:07: So you have to put in guardrails, uh, we talked about that.

00:20:09: You have to test it, uh, like, and you have to, when

00:20:13: you test it, try to break it.

00:20:16: Um, I, we, we, this is about risk assessment today, so we won't go into

00:20:20: all the details, but we will have upcoming episodes talking about that.

00:20:23: Um, but yeah, that's something if you, for example, I had, I had this

00:20:26: in the workshop with a customer.

00:20:28: Um, we had the case of, um, checking cvs and if you evaluate cvs, you are pretty

00:20:38: quickly in the space of like biases.

00:20:40: Mm-hmm.

00:20:41: And, um, and dealing with also hallucinations and

00:20:44: stuff, but especially biases.

00:20:46: Um, uh, like at least in the eu, um, you have stuff like the, uh, EU AI Act.

00:20:53: There are also other AI compliance stuff.

00:20:57: It is basically the, you, you, you are not allowed to, um, accept or decline

00:21:03: someone based on an automated evaluation.

00:21:06: There has to be a human involved.

00:21:07: So that's just the rule.

00:21:10: Um, and therefore, yeah, you have to be really careful.

00:21:13: If you have assistance for evaluation from an automated system like an AI

00:21:18: system, that's fine and all can do this.

00:21:22: And you can also set up your AI to be, to be capable of doing

00:21:25: this, but you have to be careful.

00:21:27: Svetlana: Yeah, because you, someone, some, um, agency might be after you to be,

00:21:32: to require you to take down your system.

00:21:35: There's multiple industry use cases where that has happened, so make sure that,

00:21:40: can we afford it, but is it compliant?

00:21:42: Is it, um, ethical?

00:21:45: Is it following kind of, again, like all of the.

00:21:47: Industry best kinda practices, especially where you are located in the world.

00:21:51: So there are gonna be some variabilities, but nonetheless, it is part of

00:21:55: keeping that solution viable for a long period of time so that you

00:21:59: can get your return on investment.

00:22:02: All right.

00:22:03: Um, moving on to, wait, wait, wait,

00:22:05: Edgar: wait, wait.

00:22:05: One, like, uh, last part is reputation, which also like, it's also vi thing.

00:22:11: Like if the thing you put out there is damaging your reputation, you might

00:22:16: consider, um, to at least assess the risk.

00:22:19: Uh.

00:22:20: I often get this from Google employees when they talk about like,

00:22:24: yeah, opening Eyes is a startup.

00:22:25: They can just put out stuff at the breaks.

00:22:27: No way no one cares.

00:22:28: But if Google Fs up, that's, that's an issue.

00:22:33: Uh, and, and the reputation part is just.

00:22:36: A lot bigger, um, in their case.

00:22:39: So, yeah.

00:22:40: Uh, just keep that in mind, but yeah, we should keep going.

00:22:44: Svetlana: No, that was, that was a, a good, uh, story, so hopefully it, it

00:22:47: helped, um, enhance your understanding and, and what viability is all about.

00:22:53: So next one is, next one is a

00:22:54: Edgar: cool one.

00:22:55: Yeah.

00:22:55: Svetlana: Strategic fit.

00:22:57: Yeah.

00:22:57: So this is where we're kind of getting into kind of the space.

00:22:59: What the heck is strategic fit?

00:23:02: So.

00:23:03: Ai, and I've seen this multiple times.

00:23:06: Times, and it kind of, even though I've, I'm accustomed to it, it still

00:23:09: frustrates and I'm like, why can't, can't folks get this, um, that you

00:23:14: can't implement or you shouldn't implement AI if it's not aligned to one?

00:23:19: I. Of your business needs, your customer needs, or your business objectives, right?

00:23:24: So if it, there's not strategic fit for you to take on this AI project,

00:23:29: it's a shiny object syndrome.

00:23:32: That it's basically, it's, it's shining and then you're trying to

00:23:34: chase the shiny object without kind of having the promise that it's

00:23:39: actually impact your bottom line.

00:23:40: Right?

00:23:41: So you have to make sure that.

00:23:43: Um, you be, you tackle an AI initiative that you can then measure

00:23:49: and you can justify that investment.

00:23:50: How can you justify an investment for a shiny object that has no

00:23:54: promise of bringing or impacting any part of your business?

00:23:57: It's just cool because everyone else is doing it.

00:23:59: Edgar: So, or, and, and, and it also has to fit your vision in some way, like.

00:24:04: What, what is the company going to do and, and how does it fit into

00:24:07: the overall strategy, strategy of our company, of a product?

00:24:11: Um, especially for your product teams out there.

00:24:13: Like you have to be certain that this is kind of like in character.

00:24:17: Like if you are doing like hard number crunching and, and do a, like

00:24:23: finance assistance, an image generator might be not the thing you have

00:24:27: to build, even if it's cool, like.

00:24:30: Svetlana: That's a good, good example I think.

00:24:32: Um, just, yeah, strategic fit where it's aligning, but I think that

00:24:35: there's also, again, frameworks and.

00:24:38: Um, ways that you can identify, well, which use cases should I be tackling?

00:24:43: How can I future proof my organiza my organization?

00:24:46: There's also, again, frameworks that in the future episodes, uh, will

00:24:50: cover for how do you think about short term use cases and how does

00:24:53: that align and impact my bottom line?

00:24:55: And give me the RO I'm looking for the short term.

00:24:58: What about media?

00:24:58: Medium term, and what about these, again, longer term projects, how can I

00:25:03: potentially even reimagine workflows?

00:25:05: How can I introduce new opportunities?

00:25:08: Revenue generating opportunities with ai?

00:25:10: And these tend to be a little bit longer, but you can start now, but

00:25:13: you have to be kind of goal oriented towards what you're building.

00:25:16: As I think Edgar, you said towards the future.

00:25:19: What is that?

00:25:19: Future kind of look like.

00:25:20: And then you wanna make sure that you're kind of making, you're crawling

00:25:24: initially with your initial use cases.

00:25:25: Then you're walking, then you're running with these use cases because

00:25:28: you should be, it should be contributing to your sum goal in, in the timeframe

00:25:33: that you're hoping to, to solve it.

00:25:35: So this, this all about strategic fit.

00:25:37: Edgar: And also don't forget about the competition.

00:25:39: Like how, how does this AI feature that I wanna build and I wanna integrate?

00:25:44: How does this like position me also against my competition?

00:25:49: Um, because yeah, it's just, um, it's the, the core of every strategy decision.

00:25:56: It's kind of the same for AI plus all the other risks we are talking about.

00:26:01: Uh, but, but yeah.

00:26:02: Uh, it, it has to be.

00:26:04: Because the company can profit of it, not because it sounds cool.

00:26:09: Yes.

00:26:10: Like don't do, don't do AI projects just for marketing purposes.

00:26:15: Svetlana: And then I think it'll come bite you in the rear.

00:26:18: I would say when, um, end of the quarter or end of the year kind of, um.

00:26:23: Reconciliations of budgets and how, where are we spending the money on?

00:26:25: Um, because typically projects are assessed, uh, what, uh, value they're

00:26:31: generating back to your organizations.

00:26:32: If, if it's not measured and you're spending money on, on these projects.

00:26:36: And if you can't kind of quantify the value that you are giving

00:26:40: back on these solutions, and if it's not, you know, again, some

00:26:44: of them can be software metrics.

00:26:45: Some of them can be harder, like financial metrics, but there

00:26:47: should be some way contributing to.

00:26:50: What you're doing and how you're do trying to grow strategically as a company, right?

00:26:54: So what does, what isn't measured that doesn't get done type of thing,

00:26:58: um, would be really appropriate here.

00:27:00: Edgar: Yep.

00:27:02: Okay.

00:27:03: Um, next one.

00:27:06: Expertise.

00:27:07: Svetlana: Uh, this one's a good one.

00:27:09: I'm a big.

00:27:10: I stickler on this.

00:27:11: Um, so expertise risk is all about do we even know how to do this, right?

00:27:17: So this is becoming more and more, I would say, in apparent risk with

00:27:22: lots of organization again, who are newer to artificial intelligence.

00:27:25: They're taking it on.

00:27:27: They think, you know, well, there's these established frameworks.

00:27:30: There's, you know, open, um, source frameworks, there's,

00:27:33: you know, established patterns.

00:27:34: Everyone's been doing it.

00:27:35: We're just gonna figure it out.

00:27:37: Um.

00:27:39: Problem is, is like the failure rates of these AI projects.

00:27:42: And I think you've seen this, you've been approached by my, you know, companies,

00:27:45: I've been approached multiple times.

00:27:47: It's like, we can't get these projects beyond pilots.

00:27:50: My first question is, who's working on these projects?

00:27:52: Do they have the, the expertise of actually tackling this or someone at

00:27:56: least on their team to show them how?

00:27:58: Right?

00:27:59: Like you need kind of the guide to show you how to actually,

00:28:03: uh, follow those best practices.

00:28:04: But if you are.

00:28:05: Tackling these projects, these very complex projects, which AI

00:28:08: tends to be for the first time.

00:28:11: I mean, it's probably gonna take a few repetitions and failures for you to learn

00:28:15: and really get good at at the craft.

00:28:17: But if you don't have.

00:28:18: The patience to fail on the first attempt, then I think the next best

00:28:21: thing that you can do is to get a guide to who will show you the expertise.

00:28:25: So the two things that,

00:28:26: Edgar: and also, and also failing might not, not lead to, to, to proper results.

00:28:30: I, I, like, I had sometimes really dig deep into testing and trial and

00:28:35: erroring and, and just knowing some like tricks, tips and tricks basically.

00:28:41: Um, the thing is like a lot of.

00:28:43: Information you find when you, like, do like, um, learning by yourself

00:28:47: and, and just winging it basically.

00:28:50: Mm-hmm.

00:28:50: It's not, it's not that, that easy to get to deep information.

00:28:54: Like there is a lot like you, you can find it.

00:28:57: Eventually you have to dig quite deep and, and quite a lot.

00:29:02: So it might be, might be better to get someone who already did the digging.

00:29:07: I always say I try to save you six to 12 months in pain.

00:29:10: So, uh, and, and I actually mean it like that because I had

00:29:14: that pain and it was painful.

00:29:16: Svetlana: Yeah.

00:29:17: We, we don't want you to go through it as well.

00:29:19: Um, so one of the things that, um.

00:29:23: Kind of that goes into this, uh, category of risk, uh, comes down to two things.

00:29:27: And usually I, again, I ask these, these things, do you have the experience

00:29:32: of actually tackling this a similar problem or working with this technology?

00:29:35: So this is the technical expertise, but the second piece is the domain knowledge.

00:29:39: I. So especially if you're, if you're working in specialized

00:29:42: areas, let's say it's marketing.

00:29:44: If you're working in healthcare, if you're working in finance and you're tackling

00:29:49: this, these use cases where you kind of have to understand how either data

00:29:53: flows through these systems, how does the appropriate responses look like?

00:29:57: To these audiences.

00:29:58: You have to have both.

00:29:59: You have to not just have the technical ex expertise.

00:30:02: Yeah.

00:30:02: 'cause between domains, they don't always translate as well.

00:30:05: So if you really want to have, I would say like, to know the secret sauce

00:30:10: behind addressing this, the expertise risk, and what is my guide typically,

00:30:16: uh, should be consisting of it's technical expertise and a domain.

00:30:20: Domain knowledge in your specific space.

00:30:23: Yeah.

00:30:23: Uh, would be the two key ingredients.

00:30:25: Edgar: I would want to add something that, because I, um, I presented this

00:30:29: to, to my participants in the workshop and they were like, yeah, what,

00:30:32: what, how, how, what does this mean?

00:30:34: Like.

00:30:35: We can't do it.

00:30:37: And that's not what this is about.

00:30:38: All this risk assessment stuff.

00:30:40: Um, and we get to some, some how to on how to assess it in the end.

00:30:45: But, but I just want to add, when we talk about risks, just like we want

00:30:48: to, uh, create awareness and if you, for example, um, find out after risk

00:30:54: assessment, we find there's a high risk that we run into problems with

00:30:58: our technical or domain expertise.

00:31:01: The solution is not like we don't do it.

00:31:03: The solution is like, we have to see how we can mitigate the risk so that we

00:31:07: don't run into it down the down the line.

00:31:10: So it's not about like do or not do, it's about like mm-hmm.

00:31:14: Assessing and then drawing the right conclusions.

00:31:18: Uh, one conclusion can be, we don't do it, don't get me wrong, but it does

00:31:23: not have to be like, it's not 1 0 0, like there is a little in between.

00:31:27: Mm-hmm.

00:31:27: And for example, for the expertise risk, you can either hire new people, you can

00:31:31: hire consultants, you can hire trainers to train your staff, um, or do all of it.

00:31:37: So, so, um, yeah, because, because you, you might also, um, find that you just.

00:31:43: Lack some resources or you, like, you have all the resources, you

00:31:46: also have domain expertise, and you have really good technical guys, but

00:31:50: you just don't have the experience that you get to consult, right?

00:31:53: So that's like, think about all this risk assessment, not about like do or not do.

00:31:57: Think about it, like to assess to, uh, to find out where we have weak, weak points,

00:32:03: which we have to, uh, tackle first.

00:32:07: And then we can start, or, or to, to ensure that we have a

00:32:11: good project in the first place.

00:32:12: Yeah,

00:32:13: Svetlana: and we show you how you could prioritize that risk as well.

00:32:17: So like, where do we think, um, we could, um, it would be the most impactful to

00:32:22: address first, second, third, where should we take a, you know, a bit of

00:32:26: chance, you know, and things like that.

00:32:28: So knowing, I would say is, um, the key, I think, emphasis here.

00:32:33: So I'm glad that you, um, emphasized that.

00:32:36: So.

00:32:38: Um, one thing that, uh, maybe a, a a quick plug here.

00:32:42: 'cause I think we're, um, we have two more risks, uh, to cover.

00:32:45: But one thing that we're going to start introducing 'cause we want to disseminate

00:32:50: our knowledge 'cause we've been learning again in the time that we've been away.

00:32:54: A lot.

00:32:54: Um, and we wanna do more workshops, uh, live interacting with you,

00:32:59: kind of having that face-to-face, um, you know, more regularly.

00:33:03: So we are gonna be hosting our kind of first, uh, uh, workshop next week,

00:33:09: which is the 22nd I believe we decided.

00:33:11: Um.

00:33:12: May 22nd, uh, between 12 Eastern and 1:00 PM Eastern to talk to

00:33:19: you about what it takes to stand up a, a successful AI project.

00:33:24: Um, how do, how do you actually achieve positive ROI for your investments?

00:33:30: How do you maximize the R that, that ROI?

00:33:34: How do you mitigate these risks?

00:33:36: Um.

00:33:36: Kind of strategically.

00:33:37: And then how do you control costs?

00:33:39: And that's another runaway factor that contributes to failure, uh, rates of ai.

00:33:43: So again, we've organized it into six different pillars that you have

00:33:49: to kind of get right in order to.

00:33:51: I would say accelerate the path of your AI adoption.

00:33:55: And so we'd love to share and get your input on that.

00:33:58: So we'll be sharing that link in the chat here as well.

00:34:02: Edgar: Yeah.

00:34:02: In the description down below.

00:34:04: Um, um, yeah, we'll love you to join.

00:34:07: Uh, we'll be most likely on LinkedIn.

00:34:09: So, uh, yeah.

00:34:11: Um, cool.

00:34:13: See you.

00:34:13: See you then.

00:34:15: Okay.

00:34:15: Risk number six.

00:34:17: Svetlana: Awesome.

00:34:18: Um, so this one's an interesting one.

00:34:21: It's, um, we're calling it AI output risk.

00:34:25: And so this kind of goes back to this whole kind of reputation brand and

00:34:30: kind of, uh, the type of, I would say, effect you leave on the market,

00:34:34: especially if this is a tool again, and just if, if this is a tool that is

00:34:39: internally facing where the risks are lower because your internal staff, and

00:34:43: they'll kind of flag these, these, uh.

00:34:47: Maybe biased outputs or whatever's wrong with the system to you.

00:34:50: And that's internally you're able to kind of control the risks.

00:34:52: There's still risks, again, even with internally facing systems.

00:34:55: But if you're doing more of a B two, B2C launch where you're kind

00:34:59: of publicly making it available, and I think a good, um, maybe example

00:35:04: there is, um, you know, Chevy.

00:35:06: Who was, uh, selling, who was enabling AI agent to sell vehicles

00:35:11: basically, uh, off of their website.

00:35:13: And, you know, the software engineer basically took advantage and he basically

00:35:19: tricked the AI to sell, um, you know, a a pickup truck for, for a dollar.

00:35:25: Yeah.

00:35:25: And that ended up all of.

00:35:28: Yeah, and it ended up being, um, all over the news.

00:35:31: People are just like, well, how reckless can Chevy be?

00:35:35: You know?

00:35:35: And, um, you know, launching this, this, uh, kind of chatbot, proper

00:35:40: guard rails and all kinds of things.

00:35:42: Edgar: I think it was a, I think it was a, um, a single seller.

00:35:44: Or like, kind of like, um, car salesmen basically, who did this with his store.

00:35:50: But yeah, uh, you have to be, you have to be careful.

00:35:56: Um, yeah, yeah.

00:35:57: Hallucinations.

00:35:58: Um, uh, been there, uh, everyone who's touched AI knows that, um,

00:36:04: sometimes comes up with stuff it does not have to come up with in the first

00:36:08: place because it does not know it.

00:36:10: Um.

00:36:14: In the end, you can do a lot of techniques.

00:36:16: But you have to test and try and test and try and, and and,

00:36:20: and really evaluate the output.

00:36:22: On, on, yeah.

00:36:24: And, and that, that's why it's also important to, to test as early as

00:36:27: you can with your users, um, to, to actually, um, try to break the system.

00:36:32: Like try to generate the wrong output.

00:36:35: Try to break.

00:36:36: Systems try, like you don't have to go into all the depth of jailbreaking,

00:36:40: like most cases, your B2B, you have kind of a somewhat controlled environment.

00:36:46: Um, if not, you have to be even more careful.

00:36:49: Mm-hmm.

00:36:49: Um, and then you, um, you have to also think about like how to monitor stuff, how

00:36:54: to get the human in the loop mechanisms in, which means that you, um, not rely

00:37:00: a hundred percent on the, a output, but, um, you, you define basically.

00:37:04: Rules or circumstances where you say, okay, now I want the human in the

00:37:08: loop to, um, make an assessment of the situation to, uh, figure out, um,

00:37:16: some, uh, I, I can't get the word.

00:37:19: Um, yeah, some, some, some workflows for, for allowing stuff to happen.

00:37:25: Um, so yeah, that, that's basically where you, where you

00:37:27: have to, um, we have to try to.

00:37:30: Get the system robustness.

00:37:33: Svetlana: Mm-hmm.

00:37:35: Edgar: And for that to happen to, for it to become robust,

00:37:38: you have to try to break it.

00:37:40: Yes.

00:37:41: Yeah.

00:37:41: Svetlana: And I think that there's again, different workflows as you,

00:37:44: as you mentioned, depending on the use cases, it's hard to prescribe and

00:37:46: tell you all of the potential kind of workflows that could work for you.

00:37:50: But there are ways to introduce kind of checkpoints.

00:37:54: Um, through your workflow to be able to, um, prevent some of these

00:37:58: like output risks, uh, for ai.

00:38:00: But it comes down to like, well, why, why does it matter, especially for an

00:38:04: internal facing tool, brand reputation, risks, trust from your employees?

00:38:09: Like there's not basically

00:38:10: Edgar: all of the above.

00:38:12: Everything

00:38:13: Svetlana: that could go wrong, uh, could, could hurt your company, whether it's,

00:38:17: again, an internal tool or external.

00:38:19: So those are the things that you should pay proper attention to.

00:38:23: So, um, should we move on to our last, but not least?

00:38:28: Edgar: Yep.

00:38:29: Svetlana: This is, um, I would say a big one.

00:38:32: And so user adoption risk.

00:38:35: So the first one I think you'll be like, well, how's it

00:38:38: different from desirability?

00:38:39: Desirability is, uh, do they actually want it?

00:38:42: Right?

00:38:42: So have they expressed it or is there a particular need

00:38:44: that I can actually anchor?

00:38:47: My use case too.

00:38:48: I give them a solution and they've then like jumping up and down and

00:38:51: saying like, I want this tool.

00:38:53: If you build it, I'm gonna use it with user adoption.

00:38:56: And this is becoming more and more, uh, of a problem with artificial intelligence

00:39:00: because people fear AI in the first place.

00:39:03: So this is nothing to do with desirability, but this is

00:39:06: being, um, kind of rolling out.

00:39:09: Yeah, but rolling out like these, these tools in the organization and people

00:39:12: are just like, I don't wanna use it.

00:39:14: Like this is gonna replace me.

00:39:15: Oh, you wanna automate that process?

00:39:17: Uh, you're gonna be taking over my job.

00:39:19: So, um, one of the things, again, you could, you could be, uh, working in

00:39:23: the larger organization, you have the strategic fit, you've checked up all

00:39:26: the boxes in the other risks, but you forgot about this change management.

00:39:30: Um.

00:39:31: Aspect of the plan, and you roll this out and you're like, crickets.

00:39:36: Uh, why isn't anyone using it?

00:39:38: And why is the adoption kind of fails?

00:39:41: And a lot of times,

00:39:42: Edgar: let me say, someone who, who's done, uh, 15 years of, um,

00:39:46: roll, rolling out, uh, software automations and companies mm-hmm.

00:39:50: People start to lobby against you.

00:39:55: They start, they start to, to, uh, uh, to, to try to, um,

00:39:59: not, not only not use it, but.

00:40:01: Make others not use it too, so, uh, as well.

00:40:04: So, um, yeah, this is, um, they create scripts.

00:40:08: Yeah, yeah, yeah, yeah.

00:40:09: People can be quite like, um, averse to change in the first place.

00:40:13: Mm-hmm.

00:40:14: And if that change even, uh, like is combined with fear, they get.

00:40:19: Highly creative.

00:40:23: Stop it from happening.

00:40:24: Yes.

00:40:24: Um, yeah.

00:40:25: So, so, uh, and, and the best thing, uh, you can do, and it's, I just mentioned it,

00:40:29: like get your users early in the process.

00:40:33: Uh, for example, if, if, if I have, um, company and I, we need

00:40:37: to find out where to start with ai.

00:40:39: Mm-hmm.

00:40:40: The first thing I ask is like, what's the most annoying thing in your

00:40:43: day you would love to get rid of?

00:40:45: As quickly as possible, and then we can see if we can fix it with ai,

00:40:49: because that's something that people don't wanna do in the first place.

00:40:51: And if you automate them mm-hmm.

00:40:52: That that's a good entry point.

00:40:54: That's also the thing where I always say like, don't start

00:40:56: with like the big solution.

00:40:58: M make like start like, because most of the time what you have in your

00:41:02: head is level three solution, like the high, like agentic, everything

00:41:06: automated, start at level one.

00:41:08: Like just give the people something to work on and to work with.

00:41:12: Um, and, and so they feel, uh, feel for this to be more

00:41:16: of tool than a replacement.

00:41:19: Um, uh, educate people properly.

00:41:22: Um, and, and, and the big one, and I, I can't overstate this.

00:41:26: Trust you have to get the people to a point where they build

00:41:30: enough trust, um, in the system.

00:41:33: And that also is the expectation management part We had in

00:41:37: the, um, in the second risk.

00:41:39: If you manage the expectation, say, Hey, this system won't work all the

00:41:43: time, but it will work enough times so that it really, really will help you.

00:41:48: Um, and, and people know that they're more, more likely to, um, yeah,

00:41:53: live with the errors the system has.

00:41:56: If you say, Hey, this is a perfect system, just use it.

00:41:59: People encounter error first time, second time.

00:42:02: Mm-hmm.

00:42:02: And they start to not trust system anymore.

00:42:05: Um, and even better, you think about the errors beforehand or maybe you know it

00:42:10: from testing and they say, Hey, this error can happen, but then you can do this.

00:42:14: And then it's good.

00:42:15: For example, um, you know that, uh, you have cases where you

00:42:19: just need to retry and the second time it most likely will work.

00:42:23: Um, so, uh, and if you know this beforehand and you give it, like give,

00:42:26: give the people this tool with the, like what's, what to do when it goes

00:42:31: wrong, um, cheat sheet, basically.

00:42:35: Mm-hmm.

00:42:36: Um, then, uh, the adaption will be a lot better.

00:42:40: Svetlana: Yeah.

00:42:41: And I think that, um, one of the other things that I'll kinda mention the other

00:42:44: side of it is like, yeah, starting small.

00:42:47: Getting them a little bit of an appetizer for what AI can really do.

00:42:52: But I would say the other extreme people are fearing their job loss and AI is

00:42:55: out to get them and all of their jobs.

00:42:57: A lot of it is actually ungrounded.

00:42:59: Um, I would say they're fearing a GI and a SI that, you know, the

00:43:04: artificial journal intelligence and things like that that don't even exist.

00:43:08: So sometimes just having that conversation and educating your.

00:43:11: Teams on how AI works, what it's truly capable of, that could

00:43:15: be completely game changing.

00:43:17: It'll open up people's minds.

00:43:18: It's like, oh, um, I shouldn't be worried about a JI or a SI now,

00:43:22: probably not for another 10, 15 years.

00:43:24: You'll be fine for the next couple.

00:43:26: Like, continue using our, you know, go use our tools.

00:43:28: You'll be how be great.

00:43:30: Edgar: Honestly, I would even connect it like this is a tool.

00:43:34: We try to help you, uh, to have less annoying work.

00:43:38: Um, let's, let's do that.

00:43:40: Um, and also give the people, uh, like proper feedback channels so that they

00:43:44: are always like, if something happens, something doesn't work as you expected.

00:43:48: The people have a, a clear feedback channel and, and, and, um.

00:43:52: Best case, even some life support to help you out, uh, to help them out.

00:43:56: Sorry.

00:43:57: Um, so yeah, but yeah.

00:43:59: Yeah.

00:44:00: But I think, again, think, think of it like the big trust bubble.

00:44:02: How do you get trust into the system and, and then the usability will be better.

00:44:07: Svetlana: Yes.

00:44:08: Address the concerns through education, I think would be good.

00:44:12: So we've covered, again, the seven risks.

00:44:14: Um, we talked about the feasibility.

00:44:17: Desirability.

00:44:18: We talked about viability risks.

00:44:20: Uh, we talked about strategic fit expertise, uh, risks, AI output risks.

00:44:26: And the last one we just covered was user adoption.

00:44:29: Risks.

00:44:30: So that basically covers kind of the core seven.

00:44:34: And do you ask like, oh, what about, what about security and compliance

00:44:37: and like all kinds of other things.

00:44:39: Of course, every industry is different and so there's gonna be some, a

00:44:43: lot of overlaps between others.

00:44:44: Some there's gonna be some specific risks to your specific industry.

00:44:48: So, um, the AI risk matrix is quite flexible in accommodating other groups

00:44:53: as well, but, and there's also a lot more

00:44:55: Edgar: details.

00:44:55: Uh, and the PDF, um, which you can download, uh, and there's

00:45:00: also some security parts in there.

00:45:02: So, yeah, we can, we, we cannot, like, dive into stuff

00:45:07: too deeply in the podcasts.

00:45:08: Um, therefore, yeah, read the document, uh, or join us on

00:45:11: the, on the workshop next week.

00:45:13: Svetlana: Yeah.

00:45:14: So, and then, um, if you do get the guide, there is a nice impact grid where you

00:45:20: can actually score all of these risks.

00:45:23: And there's some like nice instructions, like very simple, or try to keep it like

00:45:26: very lightweight on the content so that you can simply just do it and you can

00:45:31: actually visualize it like a heath map.

00:45:33: So.

00:45:34: What things should be are very important than the likelihood of them happening.

00:45:38: So what should we be addressing?

00:45:40: So if it's talent, if it's AI output risk, we just don't know

00:45:43: how to mitigate the output risk.

00:45:44: Maybe that's what we should pay attention to first.

00:45:47: Yeah.

00:45:47: What we're doing is actually desirable.

00:45:49: Probably not a risk to us right now, but I think it's a very helpful matrix to

00:45:54: actually visualize where the risk is.

00:45:56: Um.

00:45:57: For you, and you can again, add others to, to this kind of framework as well, so

00:46:02: you can visualize lots of different risks.

00:46:03: So I think this will be, um, quite helpful for those who are implementing ai.

00:46:09: Edgar: Yeah.

00:46:10: And, and, and like I said, um, in the expertise part,

00:46:12: it's, it's about assessing.

00:46:14: So you, you take all this seven.

00:46:17: Areas, you get your team together and, um, based on whatever it is you want to

00:46:22: do, you um, try to assess all these parts.

00:46:25: And you, you might have parts where you say, okay, we are, we are, we

00:46:28: are totally fine in infrastructure.

00:46:29: So, um, and it's totally feasible.

00:46:31: We can do this.

00:46:33: Um, uh, on the feasibility side, there's a low risk and you

00:46:37: don't have to focus a lot on it.

00:46:38: But you have, for example, um, stuff that, um, regulation for example, where you say,

00:46:44: okay, we have a regulatory risk that, uh, we, we have to account for and we, there's

00:46:50: a high likelihood that we get that this risk is becoming a problem down the line.

00:46:56: Mm-hmm.

00:46:57: Then you just put more focus in, and what you get in the

00:46:59: end is the stuff that you have.

00:47:02: Little to no, um, worries about it disturbing your project.

00:47:08: And you have stuff that's like really high on the list and

00:47:11: that's your high priority stuff.

00:47:12: So you go ahead, you, you evaluate this.

00:47:16: The likelihood and what impact would it have?

00:47:19: You multiply it, put it in the matrix, and then your whole team is

00:47:22: aware of like, hey, this is, these are parts which we either have to,

00:47:27: um, act now and make decisions.

00:47:30: For example, hire consultants, hire new people, whatever.

00:47:34: Um.

00:47:36: And you have also an awareness in your whole team that this stuff

00:47:40: might come up along the way.

00:47:41: So on everything I do, I keep track in my head or like, just like by

00:47:46: looking at the matrix, if one of these things is triggering the work

00:47:51: I do right now, at this moment.

00:47:52: So that's like the, the two ways to look at it.

00:47:56: Um, and yeah, so it's not, it's not a blocking stuff, it's assessing.

00:48:00: And acting accordingly.

00:48:04: Svetlana: Oh yeah.

00:48:05: I think that's, oh, well, nicely summarized.

00:48:07: And again, we were hoping to bring you some more actionable tools and

00:48:10: frameworks like this, um, in the upcoming episode so that you can take

00:48:14: these as frameworks back to your teams.

00:48:17: But if you are ready to potentially in your bin, explore, exploring AI

00:48:22: space, and you're hungry to get.

00:48:24: You know, faster results with artificial intelligence.

00:48:27: Um, we do have an AI ROI accelerator program that we'd

00:48:31: love to talk to you about.

00:48:32: Um, that again kind of focuses on the key areas where people are

00:48:38: hungry to, to really get right.

00:48:41: Um.

00:48:42: With AI projects, it's achieving ROI return on their investment

00:48:47: finally in making sure that they are investing in the right projects while

00:48:51: mitigating these risks and as again, and others that are relevant to your

00:48:54: industry, but also controlling costs.

00:48:57: So how do you manage these projects from taking years to actually a

00:49:02: much more consolidated timelines?

00:49:04: And we'll show you kind of our approach to how we do that.

00:49:07: So if you're ready to.

00:49:08: Talk to us.

00:49:09: Um, we, um, schedule a 15 minute free discovery call with us.

00:49:14: Um.

00:49:15: Myself or Edgar will walk you through the program and see if we're the right fit.

00:49:20: So hope you, uh, hope you give us a call.

00:49:23: Edgar: And if you are in Germany, we can also do this in German, so

00:49:27: it doesn't have to be English, just

00:49:30: Svetlana: so I don't speak English.

00:49:30: So that'll be, that'll be on Edgar.

00:49:32: So I only speak English.

00:49:34: Um.

00:49:34: Not German.

00:49:35: So,

00:49:36: Edgar: uh, you could, you could do, you could do Russian.

00:49:39: Svetlana: Oh, yeah, yeah.

00:49:40: We can, we could also, we have a third language that we could

00:49:42: deliver these workshops in.

00:49:44: Edgar: Um, you, you, 1:00 AM mean No, like, I could listen and understand

00:49:47: half of what you say, but like,

00:49:50: Svetlana: listen to nod.

00:49:51: So Vegar, listen to nods.

00:49:52: Um, duh.

00:49:56: No, that's awesome.

00:49:56: Edgar: Okay.

00:49:57: So yeah.

00:49:57: Um, yeah.

00:49:58: Um, call us up if you, um, if you are more, um.

00:50:01: Want, wanna have a deeper assessment.

00:50:03: Um, and, and yeah, we, we, yeah, we are happy to help.

00:50:08: Um, but besides that, yeah, um, don't let the PDF get this in your teams.

00:50:13: Uh, it will help you, um, to get, get your AI projects, um, yeah.

00:50:18: Better planned and also better executed.

00:50:20: Um, and yeah, that was our episode on risk assessments.

00:50:24: Ana, thank you very much.

00:50:25: Uh, you did most of the work on this.

00:50:26: Um, so, uh.

00:50:28: Uh, yeah, it's your area.

00:50:30: I'm more the technical guy.

00:50:33: Um, well,

00:50:34: Svetlana: I think we, um, definitely lots of examples I think

00:50:36: is, are applicable across both technical and the business sides.

00:50:40: Yeah.

00:50:40: But again, we'll have more technical, we'll try to bring in more of the

00:50:44: strategic views into the planning aspects of your projects so that

00:50:49: we can give you a much more.

00:50:51: Again, comprehensive angle to, to AI implementation.

00:50:54: So

00:50:55: Edgar: this season will definitely, definitely be a mix of, of

00:50:58: different, different things.

00:50:59: Maybe not as many episodes, but uh, we try to, to add a, a lot more value.

00:51:05: Svetlana: Awesome.

00:51:06: Edgar: Okay.

00:51:07: So, yeah.

00:51:08: Thank you very much, Solana.

00:51:09: Thank you very much.

00:51:10: Everyone out there for listening.

00:51:11: Um, like we told you, you find all the links we mentioned down in the description

00:51:15: when you're listening to the podcast.

00:51:16: It's most likely also in the descriptions there.

00:51:19: Um, that being said, um, it was from my side.

00:51:23: Thank you very much.

00:51:24: And um, yeah, Solana, you have the last words.

00:51:27: Svetlana: Yeah, no, I am, uh, I guess, and thank you so much for

00:51:30: your time and looking forward to.

00:51:32: Hearing your feedback and um, on new topics and things like that,

00:51:36: um, that you're interested in.

00:51:38: 'cause we are here to maximize the value for you.

00:51:41: So thank you so much and we'll talk hopefully in in next time.

00:51:46: Edgar: Yeah.

00:51:47: See you next one.

00:51:47: Bye.

00:51:48: Svetlana: Thank you.