Building AI For Business

https://www.youtube.com/watch?v=dAyR4cgBDTw

During an interview last year, my interviewer asked me a question; he said, "How can I reduce the risk of my AI projects failing?" After stumbling over some educated guess (I really was not prepared for this question) I admitted that I was unsure how to answer but that I would research what other experts in the fields were doing. This video is just my response back if I could go back and answer the question again.

Well, after dozens of hours of research, I saw the same problem repeated over and over again. I also found three core principles that organizations were implementing Ai into their organizations successfully.

Organization That where successful at implementing AI:

- Focused on Business Objectives 📈

- They Built Strong Teams, then Focused on Ai Models. 👩🏿‍🏫🤹‍♀️👨🏻‍🏫

- Lastly, They all focused on having clean, well-labeled, and relevant data 📡

Please don't take my word for it, be sure to check out the reference document in the comment below. It has all of the top resources I used to make this video!

Additional Resources

Check out this one page document if you a curious to learn more about the challenges that building Ai for Business faces. Resource

Transcript.

hello my fellow tech enthusiasts 2021 brought an insane amount of growth to the ai sector and with it we saw some fantastical stories of people's successes and what they can do with these ai models and i bet with 2022 coming around you are looking to start some of those projects for yourself and in your own organization ai initiatives that are sure to drive up revenue drive down costs and unlike the nearly 9 out of 10 ai projects that reportedly yield no to minimal results yours are sure to drive impact into your organization for years to come just in case you need a little bit more assurance stick around as we cover the top three reasons why ai projects fail

but first a little story for the research for this video i scoured the internet for every article every research report and every white paper i could get my hands on and over and over again i saw the same story being told the story goes like this there's a company called moogle and moogle is a non-descript company who has a name that's unfortunately sounds like that of an alien cow and moogle is run by a spiffy man named norman and one day norman reads this research report that says 45 of executives perceive some form of risk from a.i and norman thinks to himself i don't want moogle to get left behind i better start focusing more on ai so mughal calls in his techies or norman calls in his techies and says i would like us to become more ai focused and when the techies finally ask what would you like us to build norman proudly declares that he wants an ai that focuses on forecasting sales given these vague instructions the team of techies set out to build an ai project that can predict sales a small budget is approved in a team of traditional data scientists and software engineers with a little experience if air put on the project because moogle was caught in the last big wave of big data they have loads of customer data but it was never really utilized and so it has been left unlabeled unorganized and overall a complete mess but the team's budget wasn't approved enough to completely refit the debt organization's data pipeline and so they're going to do the best they can months pass and that mughal's ai team has built an ai model that can predict sales but to so to move to norman's surprise no one wants to use it see the problem is that when each department looks at it they all wonder to themselves was this designed for warehouse managers was this designed for store managers maybe it was designed for the marketing department the product in the ai model was built with no consumers in mind and so it fits no one's need well enough for them to be bothered to use it and so this ai model that was that took months to develop sits on a shelf with no one ever using it

so when you read over the story of mughal one of the things that should stand out is the complete lack of business objectives that norman had norman simply wanted to project a project sales with ai and well this plan might sound simple and appealing it lacks some of the directional help that your teams need to be successful and so let's start by identifying a business issue after going back to norman for clarification we found out that the reason he wanted to forecast sales was that overstock costs have been rising 50 year over year and he wants to better predict sales so that he can better predict demand and merchandise on hand and so this can easily be turned into a business objective i.e reduced the expenditure of storing goods in a warehouse by better projecting demand and having less merchandise on hand now notice something interesting about the business objective it doesn't actually mention anything about ai models and that's because norman doesn't need to be building ai models he needs to be solving business objectives and ai is a tool that can be used to accomplish that goal but you shouldn't just be building models because that's what all the cool kids are doing now after explaining the importance of focusing on business objectives to norman norman and moogle decide to go back and do some research on their warehouse storage problem they researched traditional and a high methods for solving this problem they come back and they determine that yes this is a solvable business problem and that ai can be used to help better predict sales and demand and reduce the amount they spend on warehouse costs from overstocking now now norman needs to ask himself a very important question and that is does norman actually have the ability and expertise and talent required to build these ai models because if not i mean no one's gonna they don't build themselves and that brings us to our next point building teams that can accomplish ai objectives and this is one of the biggest problems that norman and moogle missed is that they completely underestimated the importance of having multi-disciplinary teams who could handle all of the unique tasks that ai projects bring to the table and the skill gap and need for talent is so important that a 2019 gartner survey of executives listed it as the number one uh one number one obstacle that was preventing organizations from adopting ai more broadly and the talent challenge that moogle faces is real like you need cross-functional individuals who can work between product finance business of strategy data science software engineer that is a tall order to ask for most people you need business experts who can understand how ai can fit into the overall business strategy you need product managers to handle talking to consumers so that the product ends up being useful you also need them to build up kpis so you know how it's performing you need data scientists to manage the vast sums of data that you've collected to organize it to label it and that's utterly but you can't skip that part because if you do the model doesn't function and then you need software engineers to actually build it and if you're missing any of these pieces you're not going to get the results from the ai model that you want and it's it's not even just the team right you need leadership involved in every step of the way process and one of the more interesting things i found through this research was that oftentimes that the best position for the ai projects to be under is under the ceo and this isn't because your cios can't handle this type of job but because when it's under the ceo or another department it becomes more of a strategy more of a whole business a movement than it does become these one-off side projects in the back

after highlighting the importance of talents to the success of norman's ai projects norman went out and he hired a ai expert to handle and lead his ai divisions he also transferred over individuals who were willing to campaign for for ai inside of the organization and move them over to reskill and upskill them up and lastly norman brought on a few consultants to help ease that pathway and close that knowledge gap in the short term and if you'd like to understand why norman took this route then i suggest checking the resource document that i'm going to add to all these longer form videos where they they talk more about the challenges of talent and how different organizations have come them and norman really here is taking a a kind of a multi-faceted approach to upskilling but wait wait i think there's one thing that's kind of glaring and we haven't talked about it much yet is that you you do have the data for this right because you're going to need a lot a lot of well-labeled organized relevant data for these models to function otherwise they're not going to give you the predictions that you want and so that brings us to our last pillar the secret sauce is in the data one of the reasons that norman's ai model was so unused at the end of the day is because its predictions were so unreliable and i think data is something that we tend not to want to talk about a lot but data is the probably the most important piece right like if you get everything else right but you leave out data you can't even lock your way into good predictions and like that that's what your model uses you need to feed it as much good well-labeled sorted and relevant data as you can to get the best predictions i i think that there's a great analogy here between ai models and olympic olympic athletes olympic athletes go under this incredible nutritional dream they eat only the best ingredients they eat just as much as they need to everything is monitored and sorted so that they are eating only what they need and only as much as they need and it's very well managed and in fact they usually pay people to manage this

so you need to do the same thing with your ai there's a great uh saying i love here and i don't remember who said it but it was trash in trash out and you eat trash food and you end up eating what you're being lethargic you're slow you're not on the top of your game and ai is the same way you feed it trash data you're going to get bad predictions out of it so for norman that means going forward he needs to make sure he's investing in clear pipelines for data normally needs to start implementing data governance processes and structures to clean up label and sort through all of their data it will help immensely as they go forward go in the time being though um the team in the research department that has been pushing this new project forward has determined that in fact they can use some apis to to support and supplement their own data in the time being it's not a great long-term solution but it will help their predictions in the short term as they're building up these data structures okay so quick recap of things to keep in mind as you're planning your next ai projects

don't make the same mistake as norman use ai tools to accomplish business objectives not just because that's what all the cool kids are doing second take inventory of your talent and make sure you have the multi-disciplinary teams that are able to overcome the unique challenges that ai faces and additionally make sure your leadership is involved every step of the way through model development so that ai can be part of your business organization strategy and lastly and probably most importantly make sure you're developing robust data government systems and have the requisite data quality quantity and relevance needed to build your model because if you don't all this is kind of just a mute point and that is it guys as a far final note even though this video was kind of long i only got to cover a fraction of the things that i wanted to cover which is why i'm going to start including a research document that's full of all the links and resources that i thought were crucial to this video so if you'd like to find out more about that be sure to check it out in the footnotes and that is a wrap i hope you enjoyed this video and that it was worth the watch and be sure to leave a comment telling me about which element you think is the most important when you're running your ai projects in 2022