Get Started with AI for Business [August 2021]

Machine generated transcript…

well let us go ahead and get started to begin while we wait for people to kind of yeah get used to the technology and join us i will go ahead and kick things off with a poll so jesse would you care to launch the first poll all right it is going great well um yeah so if you want to go ahead and uh tell us what is your relationship to ai at work currently between these five different options here so it looks like we have a pretty high percentage of considering ai and then a smaller percentage of growing.

ai about three-fourths considering and one-fourths growing wait that doesnt add up what else is is there starting or skeptical yep there’s only four options oh so there’s so there’s nobody on uh skeptical and there’s nobody with fully enabled yet well great great um well that sounds excellent it seems like many people are kind of in the beginning processes of uh.

getting enabled with an ai project which makes sense because that’s what this webinar is going to be about uh so let me go ahead am i back to sharing my screen jesse yep it looks great oh i think i clicked there there we go um all right well lets go ahead and start with some formal introductions then so as you can see our wonderful pictures here i am dr tom wineyandy im a data scientist here at blue granite and i work with our different clients primarily in areas of retail finance and non-profits im certified by microsoft.

as a ai engineer but my background is as an economist my dissertation was on consumer and business decision making within the digital age so im very interested in um yeah how businesses and individuals are applying new technologies uh within their own domain and im joined by jessie uh she is the voice you heard already and will be hearing throughout the webinar shell be answering questions for us so if you have any questions to ask please put that within the question pane and well be sure to address it live um all right well lets go ahead and.

kind of cover what were going to talk about so in this webinar at blue grant we give a lot of webinars that are like technical demos but this is not one of them this is meant to be a high level overview for a non-technical audience to be able to talk about what ai is and kind of the direction its going and how businesses can utilize it.

so were going to be talking more on the ai strategy and enablement side and as i said already questions are highly encouraged so during this event just drop that question in the chat and well be sure to address it but if we run out of time or i dont know the answer off the top of my head well definitely be sure to follow up with an email to get an answer to that so dont be shy these webinars always much more.

entertaining both for the participants and more engaging on my part if i have live questions to respond to so right now were in the middle of our kind of welcoming part of the webinar but were also going to be talking about why ai matters what is ai which ai to use and then how to begin your own ai project at your organization.

and were hoping to wrap up within 50 minutes maybe 55 um but we definitely will leave you with some break before you transition to your next meeting um so now that we already introduced kind of myself and jesse i wanted to introduce our company as well so we are blue granite we are a 20 year old data consulting company and were founded in portage michigan.

and uh were very proud that weve been a partner of the year with microsoft so we work with them as well as our clients our mission is to help our clients unlock the power of their own data and we do so in three specific areas the first of which is with modern bi and reporting so this is with power bi and business intelligence.

just helping organizations understand what their data is saying but often a necessary part of that is looking at the modern data platform of an organization as well which is another core competency where we work with organizations to kind of um modernize their data architecture and how theyre consuming data where theyre storing data and all of the security and privacy associated with that and were also a very cloud-first company in that regard and were big proponents of all the benefits that there are from moving data from kind of.

on-prem to the cloud next there’s machine learning learning and ai which is what im a part of at blue granite and our team we help build different models or use different ai technologies to be able to make predictions provide insights with data or even automate different processes to kind of allow an organization to operate with deeper insights or with a greater productivity um yeah and kind of that wraps up the welcoming um so lets get into why ai is important.

um so any kind of technological progress like ai kind of comes in ebbs and flows and i see artificial intelligence as the next big technological wave its going to sweep across businesses the economy society in a lot of ways it already has and there’s two people that can back me up in this uh the first of which is.

satya nadella the ceo of microsoft who said that ai could be one of the most fundamental pieces of human technology ever created um and that’s its a very bold statement ever created but um yeah to back him up is sundar pichai the ceo of google who said that ai is something more profound than electricity or fire and i really like sundars kind of approach to this because hes comparing ai with electricity which are both general purpose technologies that means that in of themselves there’s nothing really special about what they do or yeah theyre just not very useful.

on their own but its when theyre combined with other technologies that they can really have transformative effects about how we structure our world and ai has a lot of importance for businesses not just in um yeah and how its going to organize businesses but the economic value that it can also have so when one report by mckinsey global institute predicted that by 2023 almost half of companies are going to be using a full range of ai technologies.

across their organization almost half and seventy percent of companies are going to have adopted just any kind of form of ai technology in the first place so although some organizations are going to be kind of much more involved with that the majority of organizations are also going to have already adopted that technology and in total this is expected to increase global economic output by 13 trillion dollars which is a massive amount 13 trillion dollars but just by 2030 all associated to ai.

so really there’s there’s a lot of evidence pointing to the transformative effects that ai is going to have on organizations uh and ill kind of take a pause here jesse have there been any questions yet there havent excuse me im sorry about that there have not you are also okay well uh everyone remember just drop a question in the chat if there’s anything that kind of peaks your interest and um yeah itll give an opportunity for me to dwell on that topic a little more.

now ive been using the term ai a lot already in this webinar so i should probably define it um ai and its this term is commonly used and commonly misused in a lot of different ways so there’s not really consensus about it and part of that is because artificial intelligence is covering this large area kind of of a definition.

so i the way i define ai is any kind of computer program or software that’s imitating in some way human actions or intelligence and and so its this is kind of like a squishy definition um there’s nothing clearly ai or not ai but what its getting at is that its yeah its just saying that we have weve programmed in.

some way a software that is acting as if its a human and at a very high level this could be something is or rather at a basic level this could be something like me writing us writing a computer program that can play poker and i could do that maybe using a series of if then statements if my opponent does something then based on a probability this computer player will respond with a different action.

so that’s like a programmed series of if then statements but its still artificial intelligence however within artificial artificial intelligence there’s machine learning which is a subset of this definition machine learning specifically though its any kind of like predictive model or algorithm ill use those two terms interchangeably but any kind of predictive model that’s trained from existing data to make predictions about.

future data or new data but basically using the past to predict the future and then another kind of layered deeper within machine learning is deep learning and its deep learning is a special kind of machine learning uh that’s using a layered algorithm and its going to form an artificial neural network um so yeah so this is something its uh artificial neural network it gets the name because its like loosely modeled after the human brain but in reality that’s um yeah not completely how the.

brain operates but its a useful analogy to kind of have these like very complicated algorithms that are difficult to interpret that have many many layers inside um yeah and so machine learning is often good for if you have some kind of like tabular data set and you want to make a prediction about it with a dependent variable a deep learning although it can also kind of do that same use case of machine learning its also useful with more unstructured data if you have text.

if you have voice data or video or pictures if you’re trying to predict whether this is a dog or a wolf in a picture a deep learning algorithm is probably whats going to be used to be able to differentiate those two pictures um now when we talk about ai just in this talk i kind of want to like further explain what what specifically ill be discussing um so what im not discussing.

is robotics that’s still ai but just kind of not part of the purview of the uh this hour webinar im also not talking about artificial general intelligence or agi and this is the idea um if youve seen 2000 2001 a space odyssey or the terminator some like super smart computer system that can do everything um but right now we dont have anything yeah anything close to this the computer program that’s able to beat the smartest human in chess is not able.

to just to differentiate between a wolf and a dog or maybe cant even understand some simple text and it might not even be able to do some simple addition so those are all separate algorithms that are solving those not artificial general intelligence humans still are the masters of agi and even a nine-year-old is able to do a wide variety variety of different tasks.

whether or not they’ve seen those tasks before so instead what were going to be talking about in this webinar is that were going to be talking about models or algorithms that are completing just a single task and this is kind of where the state of ai currently is and yeah and i mean we could talk have a whole webinar about the future of ai but this topic is really focused on how.

businesses can leverage this technology so i can kind of illustrate this with an example if you have some kind of home assistant whether that’s an alexa hey siri okay google hopefully i just triggered one of those home assistants right now but if you have one of these they might seem like theyre artificial general intelligence but in reality theyre just kind of a series of different single models or single algorithms stacked on top of each other so for example one of these home assistants the first model it has is.

an algorithm that’s able to convert audio speech into text and that’s the first input of data into this model so itll make that conversion and then from there itll take text data and kind of convert it or be able to try to like obtain the understanding and the meaning behind that text so when you ask a question what what are you really.

asking what do you want to know and by trying to ascertain that understanding then its able to send that as a query and respond with us with some kind of search response to what um it thought your question is to what it thinks is a good answer to that question oops um and then from there its kind of taking that that search result and converting it from text which is kind of.

how its doing the search and taking that text search and kind of converting that into a spoken reply now we dont see any of this as the user we just kind of see that audio goes in audio comes out but in reality its these series of algorithms that are each doing a single task that’s adding up to something that’s more intelligent than what any of those individual steps are yeah so the way different organizations are using ai is very different and i wanted to kind of bring in a parable that’s that’s true this is a.

based on an article that was in the harvard business review that talks about how an organization used artificial intelligence and that organization was the md anderson cancer center and they had what they called a moonshot project so the cancer center they kind of provide services and research around um yeah cancers and how to improve human condition for people affected by that and their goal was to better diagnosed uh diagnosed patients and recommend treatments for those who are affected with cancer and what they ended up doing was.

investing 62 million dollars over a period of four years to try to reach that goal however uh this was paused in 2017 and it was before any of the technology was ever used on patients so no one was able to kind of see the benefit of this very high investment in terms of finances as well as time so that’s kind of a word of caution against maybe not reaching too high in these moonshot.

projects its good that some organizations do it but if you’re kind of looking for quick wins and be able to like provide some kind of roi for artificial intelligence then starting with low-hanging fruit might be better for example the same md anderson cancer center had a series of i t initiatives that used artificial intelligence and with that they built a series of small scale ai models that performed kind of three different things there was.

one model that made hotel and restaurant recommendations to the patients and their families um and then a separate model uh was used to predict which patients would need help with billing so building some kind of like logistic regression model um yeah which are have the highest likelihood of not paying a bill and that way the organization could better like intervene and identify okay what are the.

most at risk patients and how can we provide financial assistance to um yeah to be able to target our limited resources in a very specific way and then they had a third initiative to to build a predictive model that looked at what caused employees to kind of yeah to have high employee turnover and so this was able to look at what are the different maybe areas where theyre not doing well with their employees and to be able to kind of target interventions that could help employees with their jobs where theyre less.

likely to turn over and together all three of these initiatives all three of these initiatives turned out to be very useful and they were successful so the same organization that spent over 60 million dollars in a moonshot project where that failed they were able to succeed with some more smaller scale targeted approach with ai now another factor that depends kind of how successful an organization is and implementing ai comes down to the ai maturity process and this is i break this down into four.

different areas um so these four different areas are going to be or just different stages that an organization goes through as they mature on that kind of ai continuum uh the first of which is an organization that’s just discovering ai and what theyre doing is that theyre um implementing ai maybe theyre trying a proof of concept or two and theyre just starting to learn about what this technology can do from there there’s uh organizations that are building ai this is when they’ve.

tried those proof of concepts and theyre wanting to start to implement them or maybe they have ai working in one department and they’ve just expanded to a second department and theyre maybe still testing things out and getting things ready the third level is scaling ai this is when they have some larger scale ai products that are going into production that they have employees using its across multiple departments maybe the.

entire organization and its starting to kind of um yeah be built into the ah um there yeah just kind of helping helping customers or employees perform specific tasks and assist them in what they do and then the fourth stage is organizational learning with ai and this is a larger scale if they’ve implemented ai throughout the organization if it becomes a key part of the.

decision-making process if a large majority of employees are utilizing ai on a daily basis this is where ai is starting to like transform the organization in a very broad way that the strategy of where the business is going is being kind of assisted and guided by artificial intelligence so here are four different steps that an organization can be at um and i wanted to ask you where is your organization at.

so jesse is going to go ahead and launch the second poll to yeah to see are you at one of these four places or are you have you not even reached discovering are you at none of these so we have the poll going um we have some answers coming in right now give it just a second here now we have about 50 who have voted at this point um it looks like 33 are none of these options are discovering ai and 20 are building.

ai nice so again for a webinar about getting started with ai most of the audience is in somewhere of that stage of either wanting to get started or just getting started or starting to build up a little bit um so you’re regret you’re at the right place everyone okay is my screen back yep you are perfect cool cool um all right so the reason i asked like this just wanted to ask this specific question is because these different stages of the ai maturity process is actually based on some specific research.

of what the outcomes are for an organization so there’s some research this is by the mit sloan review and they said they wanted to look at what is the likelihood of a company achieving significant financial benefits by using artificial intelligence and when they talk about significant financial benefits they define this as between 5 and a hundred million dollars of either increased revenue or decreased cost and that’s such a wide range because it depended on the size of the organization but just to just know that when i talk about significant significant financial benefits its relating to that five to.

100 million dollars of yeah of revenue increased revenue or decreased costs so of those organizations that were just at the discovering ai phase the likelihood they would achieve those benefits was only two percent um and this kind of addresses the idea that many organizations they like they see all the hype about artificial intelligence and they want to get in on the action but they might be disappointed that theyre not immediately kind of earning a lot of money just by starting a proof of concept now two percent are so there still are.

some that can but the overwhelming majority are not now for those who are at the building ai kind of stage in the maturity process theyre they have a tenfold more likely chance of receiving those significant financial benefits by um using artificial intelligence in this stage and it goes up to 21 so again its still a minority um but much more likely than it was just at discovering ai now when you scale up ai it increases.

even more and from there you get a doubling of going from 21 to 39 percent likelihood of significant financial benefits um however again were so were at like scaling your organizations already scaling ai and you only have a 40 chance of receiving significant financial benefits that’s still pretty low and they found that for those organizations that are at organizational learning with ai then it brings the number up to 73 which again is still not a guarantee.

many ai projects do not kind of have like theyre not able to recover their initial investments maybe because theyre costly maybe because theyre poorly implemented or because the organization is not quite ready for them or its just over not a long enough time horizon because maybe its not until the organization is like implementing ai in a large scale you know like high production and its a key part of their organization until its really starts to.

receive and see those returns and i think this is a good comparison with electricity to go back to that analogy because for electricity maybe its not until all employees of a company are using this new technology that it starts to be productive for the company um so yeah so it so it yeah as an organization kind of continues on that maturity process its increasing the likelihood of seeing a benefit from it um and i still dont see any questions that have come up if anyone you know has.

any questions again just drop them drop them in the chat all right uh what i want to talk about next is kind of the different kinds of artificial intelligence so we covered yeah we covered the different reasons why ai is important but lets talk about the more of the what ai or which ai an organization should choose there’s a lot of different kinds of artificial intelligence and machine learning um but i like to kind of think of them along the continuum of rather.

whether its a pre-trained model or a more customized model so those pre-trained models they have different benefits um in the fact that they’ve already been pre-trained like theyre just uh like ready to go right out of the box theyre low code or no code but the downside of these pre-trained models is that theyre highly rigid because theyre only kind of supposed to.

work with this specific use case and you cant have a use case or if you have a use case that’s beyond that its not really useful so you have more limited options there but then more customized machine learning models it allows you to make a much more targeted solution to whatever your problem you’re trying to solve is and it can give you a competitive advantage because if all companies are.

using the same pre-trained model then there’s really no difference into what predictions theyre coming up but with a custom model maybe you’re able to utilize your own employee data or whatever data your company has maybe you’re able to like kind of construct it in a way that gives you that competitive advantage when everyone else is using pre-trained models however this is more expensive there’s higher.

upfront costs required you need to also have the team that is able to work with these more customized models and yeah often there’s a lot of the additional costs come in the labor of whos whos responsible for building and maintaining such customized systems but let me also discuss a few examples along this continuum so i work a lot with microsofts.

cognitive services and these are a set of pre-trained models but they also uh not they dont just have these pre-trained models but they have models that can be that are pre-trained but they can be customized to a specific use case um kind of like yeah kind of shifting it from its initial purpose um but just kind of like nudging it and shifting it only a little bit i have a few examples ill mention uh in an upcoming slide there’s also bot services um there’s.

just been a lot of chat bots that have come up in the past five years and these chat bots they can be yeah theyre like kind of um pre-built in a lot of different ways but maybe its customized to have a specific personality that matches the branding the company wants to have or its a chat bot that’s trained on a specific local language or chatbot that’s able to answer questions specific to that company.

and clients are coming to yeah to have assistance with those specific questions uh but more on the customized side there’s automated machine learning uh which yeah ill ill get into that uh and it allows you it allows a company to train a model using its own data but not really kind of getting to not worrying too much about what model specifically its choosing as long as its getting good results and then finally there’s customized machine learning which is yeah just the most custom approach to building a model that’s specific to your organization.

your problem your data and so on so let me kind of give some examples with these different items with microsoft cognitive services they offer a whole variety of pre-trained models there’s a lot of models around vision to like kind of recognize specific items in pictures or video there’s also speech again using the home assistant example that’s converting speech to text or text to speech or using other kind of yeah trying to tag and identify other things within audio files.

there’s also language a lot of text analysis and associated things with that and then there’s also other decision support yeah used for i guess these things like content moderation and anomaly detection um but uh some of the different use cases ive used cognitive services for uh at blue granite we were part of a hackathon a couple months back where we wanted to um.

yeah we wanted to identify be able to build a model that predicted how many cars were parked on a street and microsoft has a pre-trained model that’s able to identify cars and other like objects associated with travel um however what we did is that we wanted to customize that a little more so we took the pre-trained model and then added pictures that were just right out my window and it was we were able to take those pictures tag them with what the cars looked like when they were parked on my.

specific street and then we could train a model to it was it was even better after we kind of retrained it with some more customized data and then it was able to better identify cars parked on my street than just when we used the model that was right out of the box another example is coming from text analytics so within the healthcare there’s a lot of electronic medical records and um this is like very unstructured data.

because its just text that’s not in any kind of um yeah any kind of like automate or tabular format and because of this uh there is a cognitive service that’s able to identify what are the key words or key phrases within that text and there’s even a pre-trained model specific to the medical field where it can identify different medical terms within that text so what were seeing here is just uh yeah an erm that’s already been tagged for different objects um jesse im getting a notification my audio is degraded is that are you seeing.

that too i am not but i heard you blip for just a second but you’re still sounding really good to me so i wouldnt worry about it quite yet okay great that would be embarrassing if i just went through the next 20 minutes without yeah having any kind of audience wonderful anyway so these are examples of how to use microsoft cognitive services either kind of the pre-trained ones or slightly customized um now lets talk about automated machine learning uh to be able to kind of contrast that with customized machine.

learning so to kind of like present this distinction with an example we can ask the question about how much a specific car is worth so if i have a spreadsheet with data of all these different cars different attributes of those cars and prices maybe i want to use that historical data to train a model to predict future car prices um and we can we all know right now car prices are very high but we want to build a model that can be able to give a.

very specific estimate of what the price is so if i was kind of building an approach using customized machine learning i would have to make all of these decisions for example i would have to choose out of all those features or variables in my data set which of those am i going to kind of include in my model building and once i do that i then have to select the specific algorithm ill use to train the model and there’s a long list of algorithms to choose from but not just that many of those.

algorithms have a specific parameter and parameters are something like a little knob on that algorithm that i can fine tune to change what the result is going to be so i make all these different selections and i train a model and maybe that model is only 30 percent accurate which seems low given the use case but ill yeah but i have that result its evaluated ill store that answer and ill start all the way back again ill go through the same process of choosing features maybe a new algorithm maybe new.

parameters retrain a new model and evaluate that model and maybe the second one is only 50 accurate and its like okay that’s better but i think i can do even better than that and i just can go through this process again and again and again and it can take a week just to be able to identify okay what is the best model.

yeah based on my criteria so this is a long process it can take a lot of work and maybe if i dont even care what the algorithm is i might have wasted a lot of my time in the process but that’s a customized machine learning approach and there might be some benefits to that like maybe you want a more interpretable model and so you want a specific algorithm to use to make these predictions so you can measure something like price elasticities something like that however if you are just interested in.

getting good predictions and that’s it the easy way out and often this is better for organizations is to use automated machine learning in automated ml instead of going through all those other decision making kind of steps what we do is that we just input our data we define a specific goal such as the objective can be minimizing.

accuracy of the model and then we apply specific constraints like saying i dont want to do this kind of feature engineering i want to exclude these models from consideration and i only want this auto ml to train for five hours i can apply those constraints and then automated machine learning will intelligently go through the different combinations of of yeah kind of the different objective function or the objective function ive defined and itll find combinations of features algorithms parameters and identify what is the best model given the constraints i decided and this will can take down what may.

have took taken me a whole week to do with customized machine learning i might be able to have a result within just a couple of hours so that’s automated machine learning um i can get it into another example of how customized machine learning might be beneficial at blue granite at the beginning of the pandemic some co-workers of mine were getting frustrated with their online grocery orders because a lot of things were out of stock.

and then once it was delivered they substituted some really weird items in there so what we decided to do is to we thought we could do a better job so we built a customized machine learning model to kind of make recommendations for which items should be substituted when somethings out of stock uh so for example um in this use case yeah we found that in this use case um if something like peanut butter is out of stock oh im not doing well with these.

there we go uh something like peanut butter is out of stock uh then uh if we identified that that was the justins brand of natural peanut butter um our algorithm was able to predict what are the top five replacement items if justins peanut butter is out of stock so we found that the best particular item is justins brand honey peanut butter and that was the best match and then we see some other examples of peanut butter but then the fifth best.

match for some reason is organic ice cream cones not sure whats going on there but overall we were very impressed with the model and how he was able to make predictions for what was specifically out of stock um yeah and so again this is just another use case of why uh customized machine learning is sometimes better than automated machine learning despite the difference uh jesse do we have any questions we dont have any questions but we had a.

comment come in so im not sure if you would like to read through that uh sure all right uh so yes uh someone is pointing out that automl is best used on balanced data set for classification okay yeah um again talking about the tradeoffs with automl um automl as someone points out is better with balanced data which means there’s not a lot of missing data or um or sorry i mean that’s that’s.

true also um but balanced data is if there’s you’re kind of trying to predict to an outcome that’s more rare like if im trying to say okay whats whats causing someone to have a car accident car accidents are very rare relative to the number of miles we drive so you can kind of run into some problems with.

automl if you’re predicting these rare events and then someone else i see asked about what were the different parameters that we used for this um out of pr or out of stock product recommendation engine um so the different items um lets see the parameters we used uh we were able to like kind of scrape data from some grocery website and uh yeah we looked at what the price of the.

item was what the description of that was uh and the ingredient list and then i think we just stuck with those because it was only a proof of concept we might have gotten into something else like whether or not there was like an allergy or something in there but primarily it was yeah just like the ingredient list product description and the price so maybe yeah the reason organic ice cream cones came up is because this is like a natural peanut butter and it was.

using similar description as what an organic ice cream cone was saying so yep perfect you had one more question in there tom if you want to just quick here we go answer that one okay uh someone asked what happens or so you kind of build a model uh with automl and you find that its highly accurate what would you then do to go and apply.

that model um in day-to-day business operations uh this is yeah this is like a the million dollar question its basically when you’re using like some kind of machine learning algorithm how do you take it from a proof of concept into production and there’s a lot of different ways you can do this it depends uh yeah it depends kind of where you’re training your model um so this is why were like a very cloud first company because we use products like azure machine learning and data bricks um yeah where where you can like.

kind of be able to process large amounts of data train a model very effectively and then be able to deploy it into production with also with also with relative ease so this uh this can be part of your like larger data architecture of where datas coming in um i i cant really uh answer the question with any kind of like specific way because it depends so much on like what the use case is how.

the data is coming in but there yeah there are a lot of different options whether you’re using data that’s on the cloud or some kind of app or even power bi you can apply a machine learning model in a lot of different ways yeah for example this one in this particular example we were able to like kind of just run the analysis ahead of time come up with.

the results and then just use this like selection pane where we went through and all we had to do is click a button and it already kind of pre-populated the the results of there so we didnt have to do any kind of like live yeah live analysis um all right well thank you for the questions um so now were getting into.

the next stage of okay how do you actually like start with your machine learning or artificial intelligence project um and so this is if if id had to pick one this is the most important slide of the webinar and i call this the ai solution roadmap of kind of how you get from not starting to starting your initiative and this is set up as a seven different questions because as you go through this process and as you’re like planning out.

what you’re going to do you should ask these seven questions or even if youve already started a project its worth asking them about this pre-existing project to make sure you’re not forgetting something very important and the first question is looking at what is the business problem you want to have solved and you should start with the project in mind because if you’re just saying you.

want to do ai for ais sake then you’re missing yeah you’re missing something big there should be greater purpose behind that and then once youve identified that then you can look at what is the actual business value of what you’re trying to predict um because again you dont dont just want to like say you’re using ai you want to be able to have that.

significant financial benefit from your ai project and then from there you can start to identify what data you need to answer that question some companies theyll also start their ai project just based on what data is available but maybe your biggest business problem maybe you dont even have the data yet to solve that and by asking these questions then you can identify what data do you need to start collecting in order to be able to answer that question.

and provide value for your company and then from there you can also i also need to identify the key result that’s being optimized in statistical terms this is the dependent variable so what is the one thing you’re trying to maximize or minimize or just something you’re trying to improve what is the kpi yeah what what results do you want to have and if there’s not a very clear answer to that if you cant say i want this one.

number improved that might be harder to do i worked with a client who um was a operated a call center in latin america and their their objective was to kind of um yeah was to ensure that their like employees at the call center were doing a good job or following a script and that was a very hard project initially because it took a lot of.

questioning of the of the company to be like okay how are we measuring that something like script compliance how are we making sure people are yeah kind of doing their job so the more specific question the more measurable result you can get the better and after that you can start to identify what is the best machine learning approach for that specific use case and that’s when you’re choosing.

do you want to do a pre-trained model a more customized model or if you are doing a customized model which specific one is best for that problem next you also want to see what is going to happen to your model after its been trained so how is it going to be used in production where is it going to live a lot of proof of concepts they kind of something is trained on an individual.

laptop but then it doesnt leave that laptop so you should be able to or start thinking about before you even begin the project is what the algorithm is going to do after its been trained um which was yeah a question that was asked already and then the last step is kind of bringing it all the way back around um is this model going to improve the.

end user experience that end user may be the customer and maybe an employee whos able to like better automate their job based on some kind of bot service but who is going to benefit from this and will it actually benefit what they do and if you dont have an answer to that question then you should not be starting an ai project in the first place or you should kind of go back from the beginning and kind of.

reassess what you’re doing so this is something we highly recommend and um yeah to be able to kind of guide you through that initial process uh there are kind of a few other things that are important when you’re starting your ai project uh and this is also kind of identifying who are the team members who are going to build this and move it to production.

and for the sake of time im going to go through this one a little fast but everyone knows a data scientist someone like me often whos going to build and deploy a specific model however there’s other people involved too there might be a data analyst that’s kind of visualizing data or explaining what the data is that’s being used there’s the data architect whos like designing where the data is being stored how its being structured some of the bigger back-end questions like that also a data engineer that’s kind of processing the data from where its.

being stored and kind of serving it up to the data analyst or data scientist in a useful way im not a data engineer and i rely on them so so much so if there’s any data engineers on the call thank you you’re the best and then there also is a machine learning engineer someone whos responsible for like kind of scaling the model up in production or monitoring the.

model after its already been deployed and depending on the size of your organization it might just be one person that’s doing all of these different tasks or might be a team of people that have different specialties within their team but it might involve more than just a single data scientist to implement your project and this is something that i appreciate working at blue granite is because i.

have colleagues that are going to serve all of these different roles and often will be put together on a specific project or i can at least confer with them if i run into a roadblock that’s out of my specific domain and finally its um its also important to talk about the ethical the ethics behind artificial intelligence so we follow the six different ethical principles of.

responsible ai uh that microsoft kind of uses and there’s a lot of different ideas here i only want to kind of touch on a few uh first of which is the accountability and is saying that any ai system that your organization produces that ultimately you’re responsible for it uh we cant just blame the algorithm because what its doing but before we can like yeah be accountable for what the algorithm is doing we also have to know like okay why is it making that.

prediction um why might my algorithm um why might the facial recognition not be the same for different ethnicities and we need to be able to understand or interpret our models accordingly that also goes into fairness so i did a webinar um a year ago about uh yeah using data bricks to predict which customers are going to default on a loan and we used actually like anonymized real world data for that but we could also be able to audit that model to look at.

are people of different genders being treated differently by the algorithm we didnt train the algorithm on gender as a prediction but we want to make sure that a specific group isnt being unfairly focused anyway so yeah i uh its important to consider those ethical principles as well but we are just about time so lets go ahead and formally wrap up um what i recommend.

is you’re going to get a follow-up email from this event with a link to our website uh this is specifically bluegranite com machine learning and uh what i recommend you to do is that we have a white paper on building an ai strategy for business so if you liked what we talked about today if you want to kind of consider this even more you can go ahead and download this free white paper.

we also have additional resources we have a blog that we maintain and there’s a lot of posts specific to ai and machine learning you can go ahead and check those out we also show solution briefs which are more of like industry use cases of projects we actually have worked on although anonymized about how we use ai for different businesses and its in a variety of different uh kind of industries so you can check that out as well.

and then lastly the best compliment you can give to this webinar is to recommend uh to your colleagues and friends to also attend this webinar we offer a series of free events on our events page and yeah if you like this today then definitely pass the link along to someone else as well well be giving the same one at a similar time next month so thatll be updated shortly however i can also say yep that when i started this webinar i was.

able to get permission from my higher ups to be able to have some kind of a no commitment ai consultation check so this is a one-hour meeting that you and i can have as well as with an industry specialist and this is a free like kind of one-hour meeting to just talk about ai like where your organizations going where you’re at um just to be able to like kind of start a conversation and again this is completely free its no commitment i wont be offended if you dont take us.

up on it but this is something that we are able to offer you uh even if you just like like this conversation and want to like chit chat about ai um you’re totally welcome to and you’re welcome to sign up for this and the link that jesse is going to send out in the follow-up email even if you’re attending or just watching a recorded version of this you can definitely email us here and well be able to get you that link but.

with that we wanted to thank you for attending todays webinar jesse and i really appreciate being able to kind of talk to other people and get excited about artificial intelligence you can email either of us were also on linkedin um im also going to stick around to see if there’s any like concluding questions so definitely type any questions and ill be around for a little bit uh jesse is there anything i am missing no i think you covered everything that was fantastic.

oh shucks thanks we will uh also in that follow-up email uh its going to include a video recording of todays webinar um in case you yeah arent fast enough to write down our emails now um so you can also check that out but yeah lets definitely uh stay in touch and i hope to see you at some other blue granite event.