Collaborate on Professional Invoice for Product Management with Ease Using airSlate SignNow

Professional invoice for Product Management

hi everyone and welcome to this uh exciting webinar that we have for you today uh thank you for tuning in it's really a pleasure to host this series of webinar usually every two to three weeks we invite a fantastic guest um to talk about a data topic or something in general related to the field of tech today we have a great guest for you but before we go and just to wait a little couple seconds before everyone has the time to to tune in um maybe very briefly can go into what and who is data science test in case you're not familiar with our work we are a Next Generation training institution uh specializing in all things Tech but especially data and AI so if you're ever curious about our courses to become a data analyst a machine learning engineer an analytics engineer a data engineer so on and so forth please feel free to check out our website site um where you'll always be able to book an appointment with a counselor to ask any and all questions regarding our remote but very engaging Tech courses in any case uh today we have a great guest um Yan nitka who is a managing partner at 11 owls um a consulting firm in data and analytics uh without any further Ado I'd love to introduce yen to you all and um leave the floor over to him hello thank you very much I'm still missing my slides there we go cool well thank you very much um for that brief introduction um so my name is Yan and together with my friend Lucille I founded 11 hours some three years ago and before I jump into the topic I want to give you a brief introduction into uh what we actually do um we are a strategic uh consultancy and we also do technical implementation in two uh different fields the first one is and you'll probably be very familiar with that one is the field of data analytics uh so we help companies derive and realize data strategies we help them conceptualize and Implement data warehouse Solutions uh and we build data models uh on top of that and then enable companies to build reporting cap capabilities and then we help internalize what we've built by um helping them build data teams or building them the organizational structures that they need to uh work with what we have built and then the second pillar of our work takes a more holistic approach at um at this whole problem uh and we are looking more at a data architecture level so we understand how a business process is reflected in the individual systems and tools that you have in your organization and we help you streamline these data flows or integrate new tools into your existing data flows and the talk I want to give today mainly explains why we think that these two go very well together and why we think that you all as as data analysts have a very high stake in what we call Enterprise data architecture so let me um Jump Right In um with something that probably most of you are very familiar with so we as data professionals we often might feel like firefighters right um we often are confronted with situations where we need to act quickly a dashboard shows a wrong number something in your Source system is not right um there's a whole bunch of fires that we constantly have to put out and over the years we have equipped ourselves with a whole range of tools to protect us from this virus I like to think of these uh Tools in categories we have check on in the first category so that's all the data warehouses that we have the uh data pipelining tools like airflow for example or the data modeling tools that we have like DBT to define a data model um so all the tools that power the data platforms that we build second we have a range of like governance tools that allow us to very strictly control how um we publish the data to test our data so we have things like great expect ations which help us to Define very concrete tests on the data or we have something and that has been talked a lot about throughout this year the metric layer so things like like Cube DBT has its own metric layer as well um tools that allow us to very specifically control how we publish information and then lastly we have communication tools so essentially the the wide range of dashboard tools that we have to communicate the insights that that we generate so when we think back of of firefighters um what they do in in bush fires or in forest fires is they build trenches throughout the bushlands here in this case right so you have the fire on the left hand side of the Bush lands and the Fire won't be able to jump over this fire trench uh so at least the right side of the Bush lands um remains intact right so even though we have a big fire we can at least save some uh of the Bush LS and I want to argue today that in in data we're doing something very similar right all these tools that we have given ourselves they formed this fire trench um so that we can protect ourselves from these fires that are burning throughout organ organization yet still deliver the reports and the insights that are expected from us but that is a very frustrating situation sometimes right because we we really only protecting ourselves from the fire but we're not doing anything about the fire itself right so we at 11 hours we put a lot of thought about how can we um be more proactive about these fires coming from the data analyst perspective um that we are in and to do so I want to introduce you to a model that has helped us a lot thinking U to think about this problem and um we want to understand how the process of making an analysis works so let's picture a pro or an analysis where we want to understand how our marketing team performs so in most cases it's starts off with some business process that you want to analyze right so in our case we have a marketing team that runs a campaign on a number of AD platforms and the goal of these campaigns is to generate new purchases in um our shop so you implement that process in a set of operational tools that will um create some data um for us right so we have the ad platforms uh and we might have website or a shop that generates tracking information we then pull that information out of these operational tools into a single repository a data warehouse a data Lake um or maybe directly to um AI tool and we then build a data model on top of this right so we as analysts we think about the question well now that I have these two different data sources the ad platform and the shop how can I bring these two together to answer the question well which uh purchases do do attribute to which marketing campaign right essentially the business logic of matching shops and campaigns happens in the data model step and then last but not least we'll create a report and a kpi uh that will uh provide the insights that we want to provide so we'll say well this campaign can be attributed to say 10 purchases that we see in our in our shop cool so this is the the model that I want to use to um to think about this question a more now in more concrete terms what does it look from a tooling perspective so we start off with the business process we have our marketing manager who uh who runs the campaign who creates the campaign um so that's where where things start we then have maybe a Shopify based shop uh and we might run the ads on Tik Tok and and Instagram maybe we then pull the data into our own infrastructure or data infrastructure so that might be a snowflake based data warehouse and then we have DBT on top to to build a data model and then finally we'll build a report in the dashboard tool that our company uses now what I think is is important here to notice that we as data analysts we need to understand all four of these layers in order to be able to perform our analysis we have to have a good line of communication to the to the marketing manager to understand what was the intention of your campaign we need to understand how Shopify works and how Instagram data and Tik Tok data looks like right so that we can then bring them together um so we need to walk U work across all four of these layers but what's funny if we think about which of these layers we actually have an influence on this becomes or we have drawn a very distinct line between step two and step three right we assume that all these tools that that we pull data from given they're not operated by us right Shopify and the Instagram and the Tik Tok um we don't have any influence on them so we just pull their data we just take it as it is and then we continue moving forward in our systems right once the data is in our system there we have full control and we can manipulate and fix the data as we want right but we don't touch the tools that the data is coming from we just we're mere consumers of that information now the problem that we see is if we look at um the projects that that we do very often the fire is burning in the operational system and the source data right so the problem that we have to deal with originates from a systems that we have very little control over and that can be um very frustrating so let's think back of our example that we looked at right so in our case we would have a marketing manager who still builds a campaign and that campaign is then reflected in Shopify Instagram and Tik Tok now the issue here is or the start of our problem is the campaigns are not properly passed to our Shopify shop system so whenever somebody clicks a campaign in or the ad on Instagram that information is not properly passed to Shopify so in my Shopify system I might get a track event that is not associated to the ad platform that is coming from so how does that progress throughout this model um we'll build a data model but we cannot connect this uh the track event to the actual ad um and eventually we will publish a report that will under represent um or under attribute the relevance of that specific ad right so what do we need to do to to solve this more sustainably we need to start thinking about step two right we need to start thinking about what can we do to support people uh to build better operational systems so that we can fix the fire where it actually started how do we do that now I want to argue that you as data analysts your task should not be to fix the issue right you're not the owner of the tool so it shouldn't be your job to fix the issue but what you can do and maybe what you need to do is you need to create awareness of this issue and I want to present you with three arguments why I think that we as data analysts have a very unique perspective on this issue why we are the ones that should create awareness or while we in a position um of of creating awareness for this issue so um what can we do the first one is you can create urgency right you can create urgency for the issue by saying no um so try not to solve every data issue that you have in your tool stack right as long as nobody sees the fire in your organization nobody will understand uh why you are calling the fire brigade so to say right so it's important to make the fire visible as long as you'll cover up the issues that are coming from the source systems why should anybody worry about it you're solving their issue um in in your systems the second thing that you can do is you can create transparency for the issues right I'm sure many of you will know situations or will have stories to tell about Source data being messy um about inconsistencies that you've uncovered and I think it's very a good first step or it's it's a very good first step to just share that knowledge with the people that actually owns own own these systems right if we think back of the case of of the Shopify and the ads go back to the person who actually implemented the ads and let them know hey I think there's an issue with passing the ID to to Shopify and then last but not least offer your help when there is change in the the uh in the source systems right so whenever maybe a new tool gets introduced into the tool landscape of your organization or um when a tool under goes a major refactoring urge that person who is responsible of that project to include your perspective uh into the process of um of implementing that new tool right don't wait for the tool to be implemented to be then faced with the shortcomings of the tool something you can also do in a more proactive way would be to uh inspect the data um early in the process so do request access to the data that these systems will produce and then compare that to the other information that you might have from other source systems so these are three things that you can do to raise awareness for this issue and to make that issue more visible in your organization in an hope to get those that actually own the tools to fix these top topics now why do I think that you as an analyst should do that why why should that be your issue why shouldn't that be someone else's problem now we argue that we as data analysts we have a very unique perspective on this issue that in many cases we see is not presence anywhere else in an organization and we think so because of three uh reasons that I will go through in detail um throughout my next slides the first one is we understand the oper the difference between operational requirements to data and analytical requirements of data the second reason is we understand how data flows across different systems and then the third reason is we have a single data repository that allows us to combine information into quickly analyze information from different Source systems so let's jump into to each of these three uh perspectives in in more detail let's start with the operational versus analytical data requirement now what do I mean by that exactly I think to data there's always two types of requirements right there is an operational requirement that this data needs to fulfill so in most cases the data needs to provide some kind of information about how something is right I want to know um what permissions for example user has in my application that's the operational requirement that this data set that this tool holds need to fulfill from an analytical point of view that can be very different right I want to I might be interested in how a certain property has evolved over time right yet the same data needs to be able to answer both of these questions now I brought a little case study or use case to explain that um in in more detail let's picture a customer relationship management system like Salesforce right so you're you have an organization that has some salese uh and these sales people operate in Salos that's their tool that they do their job with and now these sales people they do a couple of sales activities they write emails they do phone calls they might meet with people and in Salesforce they have like a table or a view that allows them to log that these activities right and they log these activities as free text right they just type well today I had a call um with Marie or I had a meeting with Peter right I just write that down now you have two requirements to this data the first one is you have an operational requirement and that might be you want to provide the sales agent with an overview over the sales activity that that person has done with a particular potential client right so as a sales agent I want to go into my client I want to be able to scroll through um the activities that I've performed with this client now the free text uh solution might work just fine here right because it's probably going to be four five six touch points I had with this client can just read through them um operational requirement uh is is met now from an analytical point of view I might have a very different requirement to that data I might want to be interested in understanding well how many calls did a particular SES agent do last week or how many emails did a particular Department do in the last week so for that requirement having the data as free text might be more problematic right because it's probably harder to um aggregate all the events that are calls and distinguish them from those that are meetings um so we here have an architectural gap between the operational requirement and the analytical requirement and we argue that we as data analysts our unique perspective on this architectural problem is that we can translate the analytical requirement so in our case being able to analyze a certain type of activity across time or across an a range of sales agents and we can uh translate these requirements into concrete data requirements we can go to the tool owner and tell the tool owner hey your data needs to be able to answer this question for me so this is the first perspective that we think is is unique to data analysts what is the second perspective cross- system data flows what do we mean by that we data analyst when we perform an analysis we need to combine information from different systems right so we have a very good understanding how a particular piece of information is represented in different systems again let's look at a concrete use case um picture uh a task of creating a user Journey so you want to be able to understand how a user traverses through individual steps uh in in your business right let's keep this simple maybe you have a marketing team that sends out emails to potential clients you have a sales team that will pick up the phone and call these uh call these people and then you have your product or your your um your product which the user would then end up buying and the user Journey you want to you want to be able to analyze understand well out of those who bought the product who received the email right or how long ago that that person received my initial email so the requirement that we have here to the data is that we must be able to track that individual person throughout all of these three systems right that person Remains the Same um but maybe when we send the marketing email uh to begin with that user doesn't have like a unique identifier in our system right so we need to make sure that we can track that user throughout all three of these systems so the architectural Gap here that we are solving is we can understand how does each of these systems represent a user and how can we make sure that we are able to identify the same user across all of these three systems right so yeah to summarize we understand how these uh data flows work between systems and then lastly we have a single data repository and that is super powerful because probably nowhere else in your organization do you have the technical capabilities to combine data from different sources with such ease as you might have in your single right repository now that might be a data warehouse or a data l or it depends on um the the technical platform that you're that you're running so what would be a concrete use case for this now picture a situation where you have a subscription management system so A system that is intended to uh to manage the subscriptions that user have in your product um and you have an Erp systems that is used by accounting to to do accounting stuff now both of these systems will have to deal with invoices right maybe the subscription system creates the invoice and sends it to the client but the invoice also needs to end up in your Erp system because accounting will need the information from the invoice um to do their bookkeep bookkeeping now what happens if once the invoice lands with or sent to accounting they realize oh we need to change something uh about the invoice maybe the address goes wrong so they go ahead and change that in their system right but that change is never synced back to the subscription management system so what is the situation you end up end up with is you have two systems that have the same uh should contain the same information yet they diverge um from one another right and these inconsistencies can happen very quickly if you haven't properly synced up your data flows between uh between your systems now the architectural Gap that we um uh that we can overcome is with this single repository that we have we can uncover these inconsistencies very quickly right because uh it will boil down to running if you have a DAT Warehouse to running a SQL query right and that way you can uncover these inconsistencies otherwise you would probably have to do a manual export of your Erp system a manual export of your subscription management system and then you would have to bring these two together to uncover these inconsistencies so we as analysts we have tools and Technologies to combine these information from different sources and thereby we can make the fire visible very very easily so that's the last reason why we think that we as data analysts we have a unique perspective on this problem of understanding what is wrong with how data flows between our operation systems now I want to conclude with uh why this is important like why is this something that you should care about as data analis and there are mainly three reasons the first one is probably pretty obvious right you want to make your life easier it can be very frustrating to have to fix data issues all the time um because you don't have time to focus on your actual work right you your work becomes very reactive and and not proactive so the less fire there is the less you have to worry about putting them out all the time time right it will also reduce your the the the number of tools that you have to deal with right you won't have to put up such a high barrier to protect yourself from fire the second one is you can extend your impact in your organization bad data architecture uh has a very can be very harmful for your entire organization not just from a from a data perspective right if the data flows in your organization are not synced up properly um then that can lead to very bad operational um impacts as well and then last but not least use that unique perspective that you have uh in your organization make uh leverage that uh position that you have because we very much think that you have a perspective on this whole issue that probably not so many in your organizations have so that is why it is so important for you as an analysts to contribute that perspective uh to the resolution of that problem this concludes uh my short presentation thank you very much for uh listening uh feel free to scan uh this QR code to download the slides if you want or uh to reach out um there's also my contact information on on screen here um so I'm happy to take questions now if I understood correctly but of course if you have any questions um maybe later uh feel free to to reach out on on LinkedIn or or via email thank you so much and for very insightful talk uh yes please any anyone and everyone feel free to ask your question in the chat whether you're tuning in from LinkedIn or from YouTube um we can always see these questions um so if you have any comment question ideas whatever feel free to put them uh in the chat and we'll be sure to try at least to answer them um all right this is kind of the typical question y while we wait for any questions that we may have uh that we like to answer that we like to ask most of our guests right uh working in the field of analytics what is and I think you're seeing me coming from a mile away what is your take on geni and how that might help um or not help your profession Maybe defuse some of the knowledge maybe also get people interested in these topics the non-technical folk um it's just one of the typical questions we usually like to ask around it's a lot of changes these past few months um what's your take on those topics in just a couple words maybe yeah yeah very interesting question I'll have to um to be careful not to spend the remaining half an hour talking about this um but so what we have seen in in our space is people got excited um about gen because they thought oh great now I can start chatting with my data I no longer have to write SQL queries um but essentially we saw um a host of tools popping up that would promise US um that we can interact with our data warehouse in a more chat type fashion right I could ask like okay what if we stick with that ad problem uh how many how many purchases did that ad produce essentially right and then the LM would kind of understand that question write the SQL query for us and return us the result now the problem with this is this doesn't work H why does this not work because we have so much um context information and we have so much uh details that we need to to to take care in our analysis that um it becomes very hard if not impossible to provide the llm with that kind of information um so that it can answer the question right so for very simple questions your LM will probably do fine but as soon as you get more complex and as soon as your business Logics become complex then that becomes an issue so I think there's two two solutions to that the first one is we need to understand what information do we need to start collecting um so that in in the future because I think this is the future though I mean it will be able to talk to our dat at some point but today we need to understand um what is the information that we need to gather now so that at some point I will be able to throw that at an llm so I can actually contextualize my business better right so that's question one and then question two is how do I need to simplify and and that brings me back to the to the topic of my talk how do I need to simplify the architectural landscape that I have um to make the life for an llm easier right so I think that's the two ways that you can approach this this issue yeah definitely and I don't think we're at the point yet where non-technical folk can speak directly with these gen just because if you're not capable of understanding the SQL query that came out of it you're going to have a hard time making sure that the data that came out is the the one that you were actually looking yeah exactly so I think we are still not at a place where um an llm will produce an a result that is so reliable and so good that we can um share that without a person that actually understands the business context and is able to translate that into SQL kind of checks and validates that query right and as as we haven't really solved that problem then I don't see any benefits from from running an LM maybe last last um comment on on that what we do see and things that actually do work is these systems work very well in very contained environments so think for example about e-commerce shops right especially in early stages they all look very much the same from a data perspective right they have some uh Performance Marketing they have a shop system and then they might have some uh uh like an Erp system or something but they're all like highly standardized and in these scenarios um an llm can be very powerful because essentially you have a a a standard data model that you can apply to the data of that shop and in these cases um uh applying or putting like this chat interface on top of your data Works rather well but for the majority of cases and especially if you're getting to larger organization uh this will fail very quickly definitely defitely and we have a very interesting question from Richard marara um Yan how do you approach getting by in for good data architecture I found it to be a hard pitch since the people feel since the feel people get is that doing things properly means slowing down and that the benefits are hypothetical this is coming from someone who tried to sell a data modeling product that gave you great architecture yeah that's a that's a good question and obviously that's super important especially if you want to get Buy in from uh from management to to these questions so I think the first thing that you need to understand is um bad data architecture does not only lead to you no longer being capable of um uh of steering your organization right right because the more complex kind of your upper funnel gets the the more the more problems you'll have to actually distill insights that are anywhere near meaningful uh to steer a business process but what is more important to make or to understand is that if you have beta bad data architecture Um this can have very negative operational impact as well and this is where people tend to start listening in right so um two examples the first one is um let's assume you have a company that has a whole range of Legacy systems um that restricts them a lot in how they can develop their own business right so if you have a situation where you've you you are faced with a data architecture that is so complex that no one really in your organization understands how that works it becomes very difficult to build the product that you want to build right or to extend the product that you have so that's something that people tend to to listen to um and uh the second thing is there's a there's huge inefficiencies in all of this right because if we think back of the invoice example that I that I brought to you earlier right where you might have two systems that both contain information about invoices um yet they uh they have different versions of that invoice at some point you will have an operational problem where this clashes right think of a situation where if you have Subscription Service you want to issue a new invoice right so now who do you issue this new invoice to the corrected address in the Erp system or the old but wrong invoice uh address in your subscription management system um so the the solution is well some somebody will have to go in and manually fix that issue right hugely inefficient um so these two verticals I'd say are what I think we must understand and we must put the point point towards thank you any other questions uh from the folks tuning tuning in we still have a few minutes so please feel free to ask any and all questions all right and maybe one last question for me uh what is a tool you discovered in the past I'll say I'll limited to three or four months that you think is worth giving a shout for it's always interesting to have a a lookout for some of the new technical elements that might be popping up it can be a tool or a feature of an existing tool that you think is worth giving a shout because it's a uh it's a nice feature that was designed by people that actually had data in mind or business data in mind yeah yeah um so I think I first of all we are very uh uh proud and happy to not be bound to specific tools right that's something that we are at 11 hours hold very high because it allows us to a great deal of flexibility in how we Implement our project so I think in terms of data tooling I wouldn't want to um promote yeah promote or like say this is going to be a tool that definitely solves the question um because it's it's very much dependent on um on the use case that you have now Richard the the guy who who posed the question I think has an interesting idea that he is currently working on and that might be something that's definitely worth U worth sharing because um what what he does he essentially he builds a notebook for data analysts right so Richard and his friend they are currently building reconfigured and um they are building a very low entry barrier for analysts to take notes about what they're doing now why do I think this is important because what we've looked at today uh when we think about data architectures these are highly complex systems right the problem why we need to worry about them in the first place is because they have become so complex that nobody really understands uh how they work uh or or even how they interact with one another right we might have individual tool owners that might understand how individual tool works but as soon as we connect these tools into a larger Network as we do in data analytics then we we very quickly lose information and I think a first step in solving that can be let's try to aggregate as much information about these tools as we can right the nice thing about llms is they're very good at you throwing a whole bunch of information at them and then they distill um the the relevant features for you so if we can set a very low barrier for people that are interacting with these tools anyway without being analysts and have them input as much information about these tools as we can I think that is a highly um uh uh or a very valuable Mission um to to also tackle this issue of of data um of data architecture and well thanks for giving a shout to Richard's work and I think Richard gave it back to you by asking another question a followup question if bad if bad data architecture is already in place how have you managed to start or Greenlight a re architecture project yeah good question in most cases um we don't we don't do full re architecture because as I mean as as we've discussed before like fully rear architecturing a an organization is a very costly and B is a very very large um project to begin with right so what we see very often is um when you introduce a new tool so let's say I know you need to change your Erp system because maybe you've been using vision for example and know you're ucing sap or something or you are you are introducing a new tool for your Sal team for example these are points in time where it makes perfect sense to just include the analytical perspective as well into this process of introducing the tool because I mean you're the decision to introduce the tool has been taken anyway right so um finding uh these spots um or these point in times uh we feel is very crucial or can be very um um or that can be a very good point in time to start thinking about okay what do we need to change to our data infrastructure um to to make our lives easier all right well then but some of most people have already said thank you so much again and I can only join them in thanking you for taking some time out of your day to help us and give us some great insights here today uh I don't think I see any more questions so thank you so much Yan thanks for accepting our invite good luck with 11 owls and maybe talk to you very soon thank you very much for having me and uh I hope everybody has a good rest of their day cheers everyone goodbye ciao