Dale McGeorge, Founder

Friday July 26, 2019 2pm - 3pm MST

Watch on YouTube:YouTube

Dale McGeorge, Founder Jepto

Dale McGeorge completed a Bachelor of Computer Science in 2006 and decided to Join the military instead of going into the Software industry. Spent ten years in the Australian Special Forces and left the Army in 2017 to work in the Development (Software and Websites) and Marketing Agency I started with my brother. I ran the Marketing department for 18 months and then decided I would go back to my Computer Science roots and develop a SaaS product. Having been in the Marketing industry for a short while I saw an explosion of Apps, but not many that helped with the problems we were having scaling the Agency. Which led to the creation of Jepto, a Marketing intelligence and automation App.

Show Resources

  • Jepto
  • Exploratory - A UI for R - a popular data science library.
  • awql.me - A killer free tool to run queries on your adwords data to join with other marketing data for deep insights into your marketing.
  • Hangout Deck

Show Transcript

Noah Learner 0:01 Hello, hello it's Friday 2pm Mountain Standard Time. That means that you get to chill with Jordan Choo. Hey Jordan.

Jordan Choo 0:09 Hey everyone,

Noah Learner 0:11 and myself Noah Learner. And today we're exceptionally excited to welcome Dale McGeorge, the founder of Jepto. When we started our automation Hangout, and I guess it was late February, George and I talked almost deeper. Dale and I talked super early. And we've been planning this world domination since then. So it's been a long time coming. And our schedules finally sank. And I'm really, really stoked for this one.

Noah Learner 0:41 Dale, I don't know that there's going to be tons of people in our space that are super familiar with you or know a ton about Jepto, I think a lot of people in the SEO land especially people who do local SEO, have heard of your Google My Business connector for Data Studio. Can you tell us a little bit about you and tell us a little bit about your company.

Dale McGeorge 1:02 Sure. Um, so I'm fairly new to the marketing industry. So my background, I did a bachelor computer science straight out of high school, and then do whatever on that does computer science degree does and join the military. And did that or 10 years, spent 10 years in Special Forces, and then decided that was enough for me and I would get out and get back to my roots in it. So

Dale McGeorge 1:30 I started an agency with my twin brother. We did marketing and software and web design. And then we did that for a few years. And I ran the marketing team there. And then my love of software sort of kicked in and I said, Yeah, let's get out and leave the agency side and start a software company focusing on the problems that we discovered during that agency time. So, chapter was a company sort of formed me last year.

Dale McGeorge 1:59 We basically took the the idea of automation and marketing intelligence and thought, you know, there's probably a gap here for some of the stuff, there's a million different dashboarding tools out there, but nothing that really sort of solved the problems that we were looking at. So we, in our research, we built a Data Studio connector for Google My Business, which seems to be fairly popular, so totally free, we decided that, you know, we were doing things a little differently in that regard.

Dale McGeorge 2:34 And had some great feedback on that. And then, obviously gave us an opportunity to start to talk to agencies about their problems and just continue to develop Jetta, which is is really aimed at marketing agencies to help them scale across the portfolio and make their life a little bit easier.

Noah Learner 2:54 And where are you now my understanding is that you're still in beta mode. Right?

Dale McGeorge 2:59 Right. So we're coming. Yeah, we've already just really soft released, we are still building out the first few features that were released to so far. So KPI management, which is our machine learning, predictive forecasting and budget tracking, which is a, we think just a bit more modern way of tracking advertising spends than a Google Sheet. And we're currently in the process of building out anomaly detection feature. And we should have all features sort of baseline out sort of by September, but I'm really keen to hear what people think about what we've built so far, and the direction that we go.

Noah Learner 3:45 I found it to be super cool. I think most most people who have watched this podcast know that we feature really cool software products. And just in case anyone has concerns about how ethical we are in terms of how we behave. Anything that we feature, we're not getting paid or free anything from any of these software tools, we subscribe to them, either with free trials or we also paid subscription so that, you know, we're just totally ethical about it. And we only feature stuff that we think is super cool, that's going to help agencies make more money.

Noah Learner 4:24 This is really neat. I think the automation stuff that you have coming down the pipe is really cool. I love your your slack integration now from an alerting perspective. That to me is something that I'm using every day watching how one of the KPIs that I'm tracking for one of my clients is rolling. I think it's really neat.

Noah Learner 4:44 Jordan, you're using it too, right?

Jordan Choo 4:46 Yes, I'm using it right now to make sure we you know, we keep on track with the KPIs that we set month over month and it's been great so far. I think I've been loving it.

Noah Learner 4:57 What the burn down in terms of ad spend is incredibly useful as well, I think that's a that's a key feature. And I know I'm making you uncomfortable because we didn't want to talk about your tool too much.

Noah Learner 5:09 But, but, um, can you let us know how you got started with automation? and talk a little bit about about that?

Dale McGeorge 5:17 Yeah. So I guess my sort of background in process in from the military side, and then really trying to dig down to what would make ourselves more efficient.

Dale McGeorge 5:31 I had a lot of challenges within the agency as we were starting to grow around a growing client base and then a lack of consolidation as to how things were being done, and just wasting time on administration and not giving value back to the customer. At the end of the day, you know, we were services, company time and so selling our time meant that every minute that we were spending on that client if it wasn't billable was was kind of wasted.

Dale McGeorge 6:04 So this particular slide sort of goes over the the way that we set up the company in the different kilos in which people were responsible. I was personally responsible for the collection and objective pillars. And so my main role there was looking at from a daily operational perspective. What is the team aware of the other people are working on for their same client. So we had channel specialists, like most agencies to some folks on SEO sounds like stone sem or social, but there's a lack of cohesion across the board. So when we started using teamwork, calm as a project management tool, we sort of saw the capability of the API and their, their web hooks that they provided. And so we had a few web developers internal to the company and we said, Look this is probably a good opportunity to automate some of the manual tasks. And I really got excited by what we could achieve by applying a bit of Process automation and what that meant for the team on their day to day basis. So our first sort of foray into that daily automation was to automate the task closing and filing process within teamwork. So we had a basically the goal of having every client having a separate project within teamwork.

Dale McGeorge 7:39 And then we used a time tracking system that everyone would log their time. But in the way that teamwork works is there's a board view which you're looking at now. All of the time logs every client was then put into the one project so you couldn't pull that out for analysis. So what we wanted to achieve was single view for everyone for work that's currently in progress. But then once work had been finished, we wanted to push that task over to the individual client project.

Dale McGeorge 8:11 So then we could then analyze all that data within the client project. So slightly off topic, but we built a Data Studio connector for teamwork as well, where we pulled out the data from from teamwork to then display internally, as well as to the client they can see all of that data about what's going on what's in progress without us having to actually do anything else. And that's kind of where I fell in love with our studio and what its capability was in the direction that Google was taking it. And then you know, on the automation side, and honestly, it was it was it was fun. It was cool to play with it. Do you have an effect obviously but I enjoyed it Monday. So if I use this one is a great example to to walk through.

Dale McGeorge 8:58 I guess the the main objective here is that closed column is.

Dale McGeorge 9:07 And we this is we had a naming convention across the board. And we basically just had a script that ran that said, look for this clear light, everyone loves acronyms. So we made our own three letter acronym. And so that that code of the client then matched another to a within the project and then it found the right project. And then we even had the opportunity to look at the tags and tags made the task move to a specific task list.

Dale McGeorge 9:43 And then we we built a slack notification somebody was dropped in the closed colon and didn't, didn't have the appropriate mechanisms to be able to fall away. We were kind of name and shame the individual to say, hey, you need to fix up that that task before it can be filed away.

Noah Learner 10:00 So you did you make your own slack?

Dale McGeorge 10:03 But yeah, that was just a web book. Yeah.

Noah Learner 10:08 Okay. Oh, ok. So I wish we could see this live becauseit blew you showed us this in March. Right. You showed us the life?

Dale McGeorge 10:22 Yes. I'm no longer involved.

Dale McGeorge 10:25 Yeah, it's like if they could see it work live it's kind of mind blowing. So yeah, instant, which is cool. You were talking about the challenges associated with tracking time. Can you give us a visual for that with this board view? I mean, is it just like they were clicking the time icon and then open it when that pops up? Like I don't do a ton of board view. I'm almost always in the task list view myself. Yeah. So for time tracking, is obviously various ways you can input time within a project management tool like teamwork, so you've got a slack, you've got the task, you've got the extra apps, they've got a like a desktop app for Mac and Windows. So all that time just goes MBC, like, quite grabbed and put into the task. But when, when we're looking at what our customers paying for on a monthly basis on their retainer, then we need to allocate a certain amount of hours of time that we're going to be spending. So what we do is, what we do then is we analyze the time that people were spending. But we found that trying to pull that out on a single project within teamwork was difficult because it's every single client that is currently on active tasks. So the most people I think, would probably use an individual project for the customer, but we want them to have a real consolidated view across the board across every client. What is kind of In Progress and not have to switch around.

Noah Learner 12:03 Okay, that's what not Bridget that's what I didn't get from this view. Okay. So to put it a different way, we're looking at a consolidated agency wide board view versus a single client. Okay, so that's why this is so baller. Sorry. That's so cool. Okay. Totally neat. So why did you go with a lambda script?

Dale McGeorge 12:29 So we had a few guys that had some experience with Python. And the whole service architecture really made sense to me it, you know, the opportunity to pay for it when your code runs, and the visibility you get around logging and stuff made a lot of sense for me as a business owner at the time and then trying to understand the cost implications of automation. So what Automation itself is that there is a cost, obviously involved with the setup, the maintenance, and then any compute or infrastructure that you've got underneath it and the cost and that even the free tier within AWS is quite generous.

Noah Learner 13:18 Did you so what's funny is that as I was building out my script that if we have time I'll share at the end, I looked at lambda, and then I quickly was like, a, like, it just, I was like, did you did you find that you kept in the free tier? Or did you bump up over? Yeah, so this, I'm going to free to you for that one. But it really does depend on how you set your your account up.

Noah Learner 13:44 And how often that particular lambda is running. Okay, got it. You're building 100 millisecond increments. So it's um it's pretty good. Okay, sweet. So Do we need to refer to this slide before we go on to the tools? You want to go?

Dale McGeorge 14:09 The year that we we tackled the different phases.

Noah Learner 14:13 Okay, you were breaking up a little bit. So can I move forward?

Dale McGeorge 14:21 Okay, yes.

Noah Learner 14:23 Do you want to tell us? Do you want to blow our minds away with this?

Dale McGeorge 14:28 Um, yeah. So I guess what I started to get into once we started to talk about automation and build that into the agency was what other ways could we use data for our clients and how could we separate ourselves within the industry and being something a little bit different? it the agency world, where we were, was fairly commoditized, and we found that a lot of people were trying to play compete on price only. And we wanted to try and get out of the norm, and focus less on the execution side, and then talk to clients more about the strategy side and how we could provide value there.

Dale McGeorge 15:17 So that got me sort of searching around for different tools. My programming background was over a decade ago, so I couldn't write code and I still can't write code. So I started to look around for tools that would help me advance in using data without having to write code. So two tools that I found and got a lot of use out is exploratory IO, which is a tool basically a GUI for the programming language art, which allows you to apply a lot of mathematical models to your data and then keep a complete history of any of your data wrangling, which is like a data transformation process that you do.

Dale McGeorge 16:09 And then the AdWords, query language.me, which is a visual API Query Builder, that you can set up schedules for, which I found to be interesting. So I didn't have to set up a server or do any kind of complicated code, you could just as long as you can sort of read, which fields in which particular queries to make, you can get data directly back from the AdWords API.

Dale McGeorge 16:42 Can you share an application using these two tools? I think we talked a little bit about a car dealership example that you guys drilled a ton of data into, is that worth sharing

Dale McGeorge 16:56 Yeah, so I one of the problems of A lot of people were having was, was correlation between online activity and offline activity. And so what we tried to understand was how much of an effect was what we were doing. From a tactical point of view, you know, Justin beads and SEO strategy content, how much were people being affected in their sales process for that offline activity? So we started to, to speak to the dealerships and say, Okay, how do we get that information out? And then how do we then analyze that those two different data sets?

Dale McGeorge 17:41 And so what we we looked at was, could we do some trend analysis? Could we try and understand whether the, the overall effect of increasing website sessions was having effect on car sites, and what part of that website

Dale McGeorge 18:00 So if we spent extra money on Google ads, if we spend more money on Facebook was that having effect compared to offline activities like mail drops or, you know, outdoor advertising or radio. And so we started to put together a bit of a calendar of the activity that was going on, and then building up a trend analysis for what the yearly seasonality looks like for those businesses. And those particular companies, car dealers run the same campaigns in the same time of month, every year. It's not doesn't really change too much. So we have a good foundation to be able to compare the website traffic seasonality with the sales activity, and we were digging through it and we couldn't really find a correlation that at all between increase online sessions and goals, and that includes form submissions to actual car sales. So we sort of looked around and then looked at the data. And we actually found that there's a really good seasonal activity as to when people buy cars. So the online advertising that we were doing and even the offline advertising that the other agencies involved were doing, weren't necessarily focused at the right point in time at when people were buying cars. So the, the way that we looked at it from a tactical standpoint was shifting budget from periods where we know people aren't buying cars, which, as an example in the financial year, a lot of people are buying at the end of the financial, you know, at the start of the financial year. And so, a lot of businesses are buying vehicles at that point in time or or That's when most people feel that car dealers have their biggest sales. So we actually then just from a tactical standpoint, shifted budget out of certain parts of the year. So things like Christmas, New Year, and then after the end of financial year, and push it in to the end of financial media, and then that made a massive impact because when people are searching for a deal, during that period, we were far more visible because we had extra budget to compete with, you know, with their competitors. So one one question that comes up is for someone that wants to start off with this, almost like predictive forecasting with seasonality. What Where do you get started? How much historical data do you actually need for you to be able to accurately predict something like this? Yeah, great questions. I am. I'm by no means a computer science guru. So What we have found is that a few years of data is great. Like if you've got less than that, you obviously need to understand that the way that the algorithm works. So if we dive into this particular algorithm, so, profit is a open source predictive algorithm released by the core data science team from Facebook, widely used amongst the analysts and the data science community.

Dale McGeorge 21:36 This allows you to be able to get that in, you know, fairly few steps, there's probably, if you use a tool like exploratory.io, you could do this in in probably two minutes. It doesn't take a lot of understanding.

Dale McGeorge 21:55 The vantage of exploratory.io that I found is that they have a connected with Google Analytics, and then you can start to play around with the data in there if you need to manipulate it. But these algorithms adopted into a lot of tools. And anybody can now access these types of algorithms very, very easily.

Jordan Choo 22:17 Cool. That's really cool. Because what's what's what I find really interesting is that you no longer need or no longer necessarily need a data warehouse to start crunching these numbers, you can, you know, tap directly into Google Analytics or AdWords API and start crunching.

Dale McGeorge 22:36 Yeah, and a lot of the data sets that are needed, they don't do a big data set, right. So the way the profit works is that it takes a daytime column, and then it takes a metric on so you don't need to pull in your whole Google Analytics account. You just need to have a date time series with a single metric. So if that metric could be Google Analytics goals could be session.

Dale McGeorge 23:00 Users, it really just could be any metric that you want to be able to forecast. So the advantage of this particular algorithm is that it gives you a yearly and weekly seasonality as well as the predictive forecast. So the underlying algorithm, it's a regression algorithm. So it's, it's an additive model, so builds the yearly and weekly for you. And then that's how it comes up with that prediction. It takes into consideration things like missing data, and then what they call holidays. So one of the problems with predicting data is you have points in time that will affect your data. So if you have a public holiday, for example, if you had a big sale or you were closed during that time, that affects what It looks like from a forecasting point of view. And not only that is the fallout after that. So on Christmas day if you're closed on Christmas Day, and then you've got Boxing Day and those days after the algorithm learns that and understand that means the subsequent days on top of that

Dale McGeorge 23:12 We lost you for a second after Boxing Day. Like it. Yeah. consideration.

Dale McGeorge 24:33 Yeah, yep. So I apologize. This morning.

Noah Learner 24:37 Can you remind Can you remind us because we were talking yesterday about the G click ID, and how there was a, that was something that you needed to warehouse. Can you talk us through that? That was really fascinating.

Dale McGeorge 24:50 So when you visit a website after clicking on a Google ad, you have the G click ID, the GCLID in the URL.

Dale McGeorge 25:03 That particular g click ID is able to be queried against the AdWords API. And then it's able to give you back all of the information around that particular click. So it'll tell you the campaign, it will tell you the Search Network type, what the key word was a bunch of information, but it's only available for 90 days. So what the specific use case that we're looking at is that, in general speak, Google ads will give you a ability to say this particular campaign chose this particular goal. And these are the keywords that triggered it. But we wanted to bring that down to the individual user level. So we came up with a system where once the person landed on the website would use JavaScript to then pull the G click ID of the URL because it's also on the on the landing page. And then if that individual was to identify themselves in some form, so logging in, a web form submission, we would then pass that g click ID into the form submission as a hidden field. And then we were able to then link the so make the the API call in the background as a separate process, and then join the data sets together. So then we get the individual user, and then the individual users search history of what ads they're clicking on. So what they're exposed to, and then make some determination against what they actually did versus their search term. So we could understand that the search term that a person is using shows intent, but their actions speak louder than that. And that's the bit that we wanted to understand is that if we were putting certain budgets towards keywords, was that driving that service directly related to that keyword, or where people coming in off one particular search query and then converting on another service or product. And what we found is that some of the low CPC keywords were driving conversions. And so we looked at shifting budget away from some of the highest CPC keywords and knowing that a lot of people are converting to the most profitable service or products for the clients of those those side services.

Dale McGeorge 27:55 And that was a really good way to understand to the client that the way that we set the budget out is not just, you know, you advertise x widget and then you sell x widget, you can advertise something else. And then I end up, you know, purchasing the most popular product, because that's normally what you get known for and what seems to be resonating with with customers.

Jordan Choo 28:18 And I'm curious, how did you take into account the various attribution models that you know ga and Google ads have into your model?

Dale McGeorge 28:30 So we basically bypassed all that. And we didn't look at attribution in that sense, mainly because we could tie the individual user back to the search query. So we knew that he was biased in the sense that whatever attribution model we were using, ads is like I was going to have a focus there because of the different models of last click or not. So we, we kind of knew that those people were probably. Depending on this vertical or a point where they're looking for brand related terms or or generic terms, so we can make some general assumptions there. But we left attribution to the side because the data that we had was more on an individual level and not necessarily channel focused.

Jordan Choo 29:25 Interesting. So how did you I guess take into consideration zero click search results, like within your model, I guess?

Dale McGeorge 29:34 um, well, so this analysis is really at the end of day like a giant spreadsheet. So all I had was the CRM data on one side, and then we would match the the query to the individual on the other. So there is a correlation that we tried to establish between which particular purchases and customers versus what the actual sale was. So we then we initially create a model per se, it was really just a bunch of data analysis to have a look at the keywords that were driving the sales themselves.

Noah Learner 30:15 Okay, so I want to reframe to make sure that I'm hearing you correctly. So let's say that we're doing a lead gen job, where we drove the user with a Google ad to a page, the G click IDs in the URL string. We use JavaScript to grab it added to a hidden form field, which pushes the G click ID into the CRM, which is the unique ID that we can then join later. Right. That's, that's the magic. Okay, yeah. You gotta push it to the CRM in the first place.

Dale McGeorge 30:47 Yeah.

Noah Learner 30:48 Cool. I just want to make sure that for anyone who wasn't putting two and two together, they weren't carrying three or five.

Dale McGeorge 30:54 Yeah, and then you can use a tool like that. awql.me to be able to schedule, those API calls. So you don't need a developer. You don't necessarily need to get into the nitty gritty of code to be able to get that data out. So you could just say schedule that to the export to the CSV once a week.

Noah Learner 31:16 Jordan, I could see a real use for your clients for for this type of technology.

Jordan Choo 31:21 Yeah.

Noah Learner 31:22 Chatting after

Jordan Choo 31:24 I'm getting. I'm getting giddy about it. I like I like this a lot.

Noah Learner 31:28 Yeah. So I was just telling Jordan to say, hey, say well Wil Reynold said. So we're at MozCon last week, real Reynolds was on stage and and his whole presentation was basically like you people are a bunch of idiots. And you're all wasting hundreds of thousands of dollars of your clients ad spend. And what you really need to do is go after any search term that AdWords defines as 00 searches per month. Cuz he was seeing tons of conversions, and no one else was targeting it. And and that's kind of parallel to with what you're saying. And it's almost like you learn that by accident at some point, right? You're like holy shit. And his he had all kinds of cool things that he was sharing but anything that we missed on this side the data exploration side,

Dale McGeorge 32:21 yes, it this is just a quick visualization of what the seasonality looks like from the algorithm. So this is from the Facebook profit algorithm. So, as one of the I guess a byproduct of the forecasting is that they make available the weekly and yearly seasonality charts. So you can see for this individual customer on a yearly basis, they actually have six weeks sales cycles, so they have a sale, a new sale every single six weeks. So the peaks and troughs there relate to when the campaign launches and then obviously as it goes on, it becomes less popular because the the bulk of the messaging is always on at the start of the campaign. So this is using, as it says their Google Analytics sessions. So this is what this customers website traffic from a seasonality point of view looks like. And and this was really good information for us to really understand about when people are visiting. And then you can overlay this with your sales data, to really start to understand how the two correlate. If you are an online business, those two are going to be tightly coupled. But if you're an offline business, then applying that yearly seasonality over the website seasonality helps you to understand also some of the the length of the buying cycle. So how long does it take someone from first visit over to actual sale? And when do people start to think about it. So you can make a few different assumptions from the data and then run some individual tests to help you confirm those hypotheses.

Noah Learner 34:11 When we were talking about the car dealership example, yesterday, my brain went straight to if everybody sinking their budget into the time when the dealerships are the least profitable, I either have their cars selling at the lowest prices, why don't we instead push budget to win their have the best margin? We didn't really get the answer to that question that I remember, but it sounds earlier today. I mean, would you say that the answer is you, you put the budget where the people are versus where the margin is?

Dale McGeorge 34:46 I think that there's a little bit more to it than that. Like it's it's ideal that obviously, if you could, but I think that you have to understand more about the buying behavior and what I kind of want In the data analysis of a lot of different customers was, you really need to go back to the individual, like the marketing, art of digital marketing. And I think that that's a part where a lot of people don't focus on, they don't understand a lot about consumer behavior and the way that people make purchases. And so it's very easy to get fixated by data and say, you know, put this here and do that. And then I'm going to have a big win. But there's a lot more complexity to the way that people actually purchase. And that's the bit that was kind of the missing piece of the puzzle. So I would say that, ideally, yeah, you could definitely do that as an experiment. But if it didn't work, then you need to go back to the drawing board. And that's the things you need to to run a test, and then record that test and then be ruthless with the outcome and say, either it worked or it didn't. And that's what I think customers really appreciated is the fact that you're trying something different. You're giving the results And then if it if it doesn't work, then you've got another idea, rather than just hoping that someday it's going to get better by keep adding the adjustments and, you know, keep releasing blog articles hope that one day that that's all going to pay off. Yeah, yeah.

Noah Learner 36:17 I'm anomaly detection. This is one of my favorite things. I love to focus on this. And it all comes back to Dana DiTomaso scolding me at lunch last year, saying, she, I showed her one of my Data Studio reports, and she kind of chuckled at me. She's like, you know, this is kind of crappy. Like, this isn't a report. This is this is like internal, like, there's reporting. And then there's, like monitoring. And so anyway, I figured I learned quickly that anomalies and monitoring have their place and reporting is somewhere else. But can you share with us what your take is on anomaly detection and why it's important? Yes, sir.

Dale McGeorge 37:00 I found this important because we had faced what I think a lot of agencies have where they have a goal that gets misconfigured. And then you've got no goal data. And so going back to the customer after two weeks, and send me hey, yeah, something happened to your Google Analytics. And we've lost our we don't have any tracking and kind of tell you how performance when, and then your whole report falls down.

Dale McGeorge 37:29 Because someone made some small mistake somewhere along the line. And we kind of looked at this as a way to try and give us back that peace of mind that as you know, an agency scales, the ability to watch over metrics for each client diminishes rapidly. So I started to look at this and say, well, let's look at ways to monitor them, but do it in a smarter way. So a lot of systems out there will look at thresholds And we tried a lot of different tools along the way to try and get that, you know, peace of mind back didn't bring them three in the morning, how's everything actually running the way it's supposed to be running? And the thresholds kept pinging. Right. So we started to see that 234 5am, you're going to get alerts because that's when there's no traffic. And so then you get over alerted, and then you kind of just lose hope that, you know, that's actually working the way it's supposed to. Because you get into this mindset that are just another false positive. So we looked around at the algorithms that were available. Again, like we're not data science gurus, we don't know how to build algorithms, we just want to apply those algorithms to the data. So Facebook Prophet is an option, but its preference is basically built for prediction. So we looked elsewhere and we found Twitter released a Open Source an algorithm called the season hybrid ESD.

Dale McGeorge 39:10 And as you can understand the companies are throwing

Noah Learner 39:14 we lost you right when you said seasonal hybrid ESD?

Dale McGeorge 39:18 Yeah. The I didn't say the message to the families are throwing massive amounts of money and resources to build these algorithms. And so we just want to take advantage of that and say, well, let's use that and apply it to our daughter and see what happens. And the algorithms are very robust. And you can find to them so these algorithms have inputs, and so they called hyper parameters. And that allows you to adjust the way that the algorithm understands and interprets that data.

Dale McGeorge 39:57 And so you can start to get an understanding of how often you want to be alerted. So if you have a high profile client that you really want to be aware of, if dips happen or increases happen, then you might adjust up the sensitivity. But if you've got a client that you just kind of want to know that everything's sort of running, and then if something drops it off for, or it looks kind of odd that you want to get alerted, you can back off some of the sensitivity and know that you're not necessarily going to action, something straight away for that particular client. Anyway.

Dale McGeorge 40:33 And so we again, we started with exploratory start of this algorithm is supported within within their tool. That's actually a screenshot from their tool. And we started to play around with that and work out well. How can we actually use this in a more automated fashion.

Dale McGeorge 40:52 I never got to finish that within the agency. But I'm actually building this out within within get to the outside. To make this a lot easier, so you just simply, you know, select the metric that you want, you can adjust the different hyper parameters. And then from the app side, we're just building slack notifications, emails, and in the UI populates these particular anomalies for you to take one of action you need to.

Noah Learner 41:21 I'm not really a fan of the negative alerts. I'm not I don't really appreciate those, like when you go through the pain, first celebration.

Dale McGeorge 41:35 I mean, if you it's, it's one of those things where if it goes wrong, you probably wish that you did care about it. And that's the point where it's too late. And that's that's the situation that we were in is that we, you know, we had an issue and the client understandably grilled is for that particular missile. So if we had that the time to ask changes completely. And if you're not getting alerted for those down periods or for the false positives, then that, you know, insurance that you're going to know when something goes wrong or goes right. Like, again, you can choose obviously, how you want to be alerted, then I think that's a pretty powerful thing.

Noah Learner 42:20 Yeah, I mean, I'm a massive fan. I think it's super cool. Um, can you tell us a little bit about what you've learned?

Noah Learner 42:30 Yeah, I learned a lot. I mean, obviously, being within the industry, for a little while, sort of opened my eyes didn't show me everything. Because we were one company and you know, we did things the way that we thought and having no agency experience prior to that was difficult for me to get my head around the way things were normally done and what the way convention says. You know, it really should be done.

Dale McGeorge 42:59 But the main one Like these buzzwords of machine learning AI, can always throw blockchain or whatever else into that mix. There's, I think, a lack of understanding of how to apply those new technologies. at a lower level, a lot of examples and talks at events focus on very niche edge cases that you can train things like a machine learning model to be able to recognize objects and you know, self driving cars. But it's not really relatable back to an individual agency, what they do in day in, day out, and it's kind of a pie in the sky dream. And I think that's where I'm trying to tackle that head on. And so there is valid ways to apply this and it doesn't take a data science team. It just takes a curious individual that wants to play around with it and and there's no real sort of shortcuts to learning it other than getting in there and sound says the first one. And it kind of touches on the second one. So everyone wants to do more. So I hear a lot of people go, yeah, I want to do this, I want to do that. But they don't have time clients don't want to pay for it. They're busy running agency, and they don't want to be able to spend time on the weekends or personal time to be able to do this stuff. But it's coming, whether we like it or not, the industry is going to be heavily affected by the way that these types of technologies are going to be applied. Obviously, Google's been on this for years, and the way that their algorithms are optimizing your ads campaigns for you is showing that and you know, there's still a few people that want to run stuff manually.

Dale McGeorge 44:52 I think that's probably a wrong move in the long term. I'm trying to beat the amount of data that they just companies have on individuals and in buying behaviors, you know, I think you're going to be, you know, fighting a losing battle.

Dale McGeorge 45:08 And I think really that the last point here is that I spoken to a lot of agencies, you know, building the tool and you're trying to get some direction for the tool was that there's a lack of type of data analysis. And that's really where I think a lot of value is going to be had and where agencies can set themselves apart is the focus on what other people aren't necessarily doing. So everyone's focused on execution, everyone's going I can do my ads better than his ads. But you know, the, the main part is the strategy and then how the data or the not just the ads data, but the business data, so talking to customers about business goals, and understanding more about them from a business sense, rather than just focusing solely in what happens inside the Google Ads interface.

Dale McGeorge 46:03 And I think that's really an important part for agencies as margins is becoming smaller and smaller as well. As you know, there's a, I think, an overall shift towards client education as to why, you know, why the fees that you're charging, you know, should be charged and, you know, definitely warranted, and if you provide value outside of just the execution, I think clients will love that.

Noah Learner 46:30 What are your thoughts about the challenges to build a tool around to build a tool around establishing a true north metric for any client? C, how hard is it to build an algorithm around? Is that is that a problem you're trying to solve?

Dale McGeorge 46:52 I'm not not directly. I think at the end of the day you need to track something. I think there's a lot of companies this Don't track anything, I think that they don't set internal or external KPIs.

Dale McGeorge 47:05 So the way there's a lot of different obviously, thoughts around and methodologies around how you come up with that metric, but the first one is tracking it. And then the second one is talking about that to the client and going, Okay, we're seeing a positive increase in this particular KPI and then matching their satisfaction. I think, at the end of the day, if they're happy, and that that metric is going up, then you've chosen the right one. If you've chosen a metric where it's going up and your clients are happy, then you need to rely on

Noah Learner 47:42 Right, so it's like realigning to their business goals.

Dale McGeorge 47:46 Yeah, I mean, the expectation at the end of the day is that you meet a certain objective. So as long as you define that objective at the start with the with the client and say, this is how we're going to judge for performance. This is how you should judge our performance, then you both sharing a common goal for understanding what the output of your efforts are.

Noah Learner 48:13 I guess I'm lucky in that revenue is is just easy for me to focus on and it aligns pretty well with what they care about.

Dale McGeorge 48:21 Yeah, I think most people do. I think a lot of people are complicated. They want to know how many search queries into the top 10 and all these other random stuff that doesn't affect the business. Like it's good in the fact that things are moving forward. But businesses, you know, they have cash flow issues, they have, you know, bills to pay. They want to see that you're driving business effects and not just, you know, fancy metrics across a dashboard somewhere.

Noah Learner 48:50 Patrick had an interesting question, he said, I suppose I'm curious as to whether this type of automation can be integrated into the newer messenger bots technology.

Dale McGeorge 49:01 Yeah, totally. I think that the messenger bot craze is far from over. There's a lot of learning still to do around NLP, natural language processing and NLU natural language understanding, to be able to get that intent. And that's where I think people trying to understand not only what the individuals asking, but then, you know, the overused term of the 360 view of the customer, of how they make that personalized and personalization is definitely an area where people are spending a lot of money and that and they seeing the benefits from and that's why they're spending the money.

Noah Learner 49:47 Got it. It looks like we've got about 10 minutes left. It's not going to take me super long to show my teamwork automation, which I think is kind of the cherry on top. Jordan, what are your thoughts after all of this? I'm sure you've got some questions anything boiling up,

Jordan Choo 50:06 I, I have a lot of work to do now. I'm not sure if, if if this was good or this was bad for my productivity, but we will find out over the next few weeks and months. But this it's really interesting because, you know, using machine learning, you know, both on the predictive sense and for anomalies is something that I've been really interested in. And you know, I know you're working on that over at Jepto so I'm quite interested to see how that pans out and how it can be, you know, further incorporate into my practice because, you know, I know for myself, there have been times in the past and even recently where you know, shit hits the fan and you know, rankings will drop or budget will go to zero for one day for for no apparent reason. So, being able to stay on top of these and take a proactive approach to this rather than, like a reactive is is going to be super important not just like now but especially in the future as a you know, agencies become more and more commoditized.

Dale McGeorge 51:12 Yeah. And you want to be the one telling the customer that they got an issue rather than telling you.

Dale McGeorge 51:18 Yes, that's exactly I

Noah Learner 51:20 totally. So we What do you think? Should we do the automation?

Dale McGeorge 51:26 Yeah, let's it let's do it.

Noah Learner 51:28 This This one's pretty sick. So I my little agency has a team. Our team came to me oops. And of course, I can't spell very well. They looked at my task list since I teamwork and they said, Hey, we can't tell what the hell's do when there's there's deadlines and due dates right on the right hand side of each task and the task list view. The task list view for people who don't use teamwork looks like oh boy, I guess I got a login.

Noah Learner 51:59 It looks like Like this, it'll have like a Task name, you can have more information. You can have a deadline. Anytime you see this little recurring thing, it tells you that it's due and future dates. The downside to this is that if I click this as complete, and I refresh, the team member still sees the task, even though it says it's due August 9, or whatever.

Noah Learner 52:24 You know, so they found that that to be visually confusing, and my initial thought was, well, can't they see the deadline? And then I started to change my perspective as an agency and I said, Well, you know, I want to make their job as easy as possible. So we came up with a solution that does a couple different things. Number one, we wanted to improve the visual clarity inside teamwork, we wanted it. We wanted people to know that when they marked a task is done, you know, their task lists got to be more and more empty overtime so that at the end of the month, their task lists are just plain cleaning.

Noah Learner 53:00 Easy, and it'll be super easy.

Noah Learner 53:03 We wanted to automate the process of task list creation. And we wanted to automate the process of task assignments based on both the project category, and the roles that are assigned inside each project in the team members that are assigned to those roles in advance. And I want to show you what I mean. So if I look at at my project view, on the left hand side, for people who use teamwork all the time, they're used to seeing all of these these project categories. Each one of those categories has an idea associated with it. And so we wanted to grab those, we wanted to grab those categories. And we also wanted to grab the people who are assigned to each of the different assigned to each of those tasks. And so we do three different API calls. One is we grab a list of all of our projects that live inside of our SEO retainer list. And then we also grab all of the roles for each project and the user IDs associated with the role.

Noah Learner 54:12 We then do a post for each project that pushes to teamworks API. So rebuilds a new task list based on the new month. And if I was to run it, we it's a node based app that lives on on one of our servers. And we have a cron job that runs at midnight on the first day of every month. And so we've built some logging into it as well. So we can see if the task is running. And we can see if it's if it's applying. And so if I let it run, this is just the command line or terminal inside of Mac. Right now my whole script is running. It's making that first API call the second and right now we're watching it post, and it's building out all my new task list. And when we go back into teamwork, and I refresh that that client task list, what we're going to see is a task list associated with the current month name. So when it runs on August 1, it's going to it's going to create a new task list that says, August SEO tasks.

Noah Learner 55:18 And the dependencies are making sure that you have the categories set up inside teamwork, the role setup inside teamwork, and then you reference both the categories and the roles when you're creating your your post. Okay, cool. So when I refresh, we're going to see that I have a brand new monthly task list. And then we're also going to see that it's properly applied to the proper roles inside the project.

Noah Learner 55:49 And then after the fact, on the YouTube video will put in the comments. The link to this deck and the deck has a video Of me running the tool. And if you have questions about this, or if you use teamwork in your agency and you're interested in and having it kind of work for you, instead of spending hours and hours, just reach out to me I'm, I'd be stoked to help you with it. The bottom line is that if you're in an agency that has whatever 100 accounts, or 100 live projects or 200 projects, and you're the project manager each month on the first you have to pull it a new task list. And you have to assign all the tasks. This runs in 10 seconds and does the work that took you hours and hours each month.

Noah Learner 56:40 I don't know. And I was kind of inspired by your script that you shared with us and board view. When you showed it to us in March. I was like, holy crap. And then we started polling team members and they started to complain about the view. And I was like, Oh, you're killing me kidding me.

Noah Learner 56:59 You see, it's a recurring date. And then we pulled the more teamwork, more team people in there like, Hey, I hate this. Like, I can't tell when a things do you have, everything is recurring. So it's totally confusing. I can't see many things do and then I was like ah, ah, and I knew is going to take me a while to figure out the API calls. So there's a couple of things that I learned in the process of doing this number one.

Noah Learner 57:26 So whenever you're doing an automation project, it's like, what are the challenges? What's the proper execution? My mind went first to Oh, no problem. I'll just do I'll just do a an execution in Zapier and I had problems with the node programming inside Zapier. So I abandon that. And I figured what I would do is I build it all in vanilla jazz, and then I would get as far as I could to make a note app out of it and then pull in my dev resource to help me with that piece of it, which is what I ended up doing. One of the things that

Noah Learner 58:00 I learned along the way. And this is how dumb I was about it. I built a vanilla JS code that lived on my website. And I figured, oh, no problem, I'll just post to it. And then I'll create something in my script that reacts to that post by firing the function. Well, that didn't work. And that's what made me build out a node based server to do it. The other thing that I learned along the way, and this is something that we need to stress every single time we have a show automation is brake. Automation is brake automation is brake automation is break. And one thing that we ran into as we were doing this is that we ran up to teamworks a rate limiting, you know, like, we were hitting their API too fast. And so we had to make sure that we throttled the timing to make sure that we didn't get out of that problem. Did you have the same issue when you were building your API or thing with them? Yeah. And they're generous. I mean, you read the API docs and it's like, I don't remember what it is 150 requests per

Noah Learner 59:00 Second or something like that. So I don't understand how I was because I was only hitting 10 projects, one API call to pull in all all the projects, and then one API call per project, and then one post per project. And when you add those all up, that's less than 50 per second, because it took like three or four seconds for the whole script to run.

Noah Learner 59:24 So I think this is super useful. I know it's going to be super useful for my team. I feel like other people are probably going to want to do it. Jordan and I have talked about it, we're going to build out or I'll probably build it because like turns busy. I'm going to build out a long blog post about how to execute it step by step on our agency automated com website. It'll be our first blog and the blog and the blog section. And I feel like this will be really useful for people.

Noah Learner 59:51 Dale, is this cool or is it a piece of dog poo?

Dale McGeorge 59:54 No, it's definitely cool. I think the danger is that his life is addictive and be careful as to what you automate. I think, you know, the process for automation that you guys are covered for what to automate is good. And I think that that that's a solid approach and I think that you're right, things break, things need to be maintained. And that you just need to consider how much time you actually savingyou guys.

Noah Learner 1:00:22 Yes, exactly. And you've been just an incredible supporter of our of our Hangout, and it's been really cool to touch base with you after a lot of the episodes. And I always cringe like if I feel like I didn't bring anything good to the show. I always when I'm on Okay, did you watch the Hangout? I'm like, Oh, God, but uh, I'm glad that that you've been a guest. This has been pretty amazing.

Noah Learner 1:00:50 Are we at that time when they are at the time? Yep. Okay, so for everyone who doesn't know Dale isn't American. I know it. Surprisingly, not from South Carolina. He's actually from another country. It's called New Zealand. I'm just kidding. He's down in Australia, and he started with us today at 6am. And he's the most impressive human being I've ever seen who hasn't had coffee yet? Right? It's now 7am on a Saturday. So yeah, my hat's proverbially off to you. And I'm really excited for our our automation that's going to our episode that's going to be in two weeks. What is that is that in August 4, fifth or sixth? What's the date that two weeks from now?

Noah Learner 1:01:43 Let me check. I should know this but clearly I don't.

Noah Learner 1:01:48 Well then. As you're checking the date and the guest if you can pull that up because I shouldn't have this I felt like such a tool on last week's show.

Jordan Choo 1:02:00 And we will be having Sam Marsden from deep crawl.

Jordan Choo 1:02:04 Oh, he owes us big time cuz he bagged on the first episode. Well, the other thing of note is that I'm going to be in a public library, and lovely Stanley, Idaho. And so if anybody not or any listeners to this to this Hangout are watchers like fly fishing, if you want to come up to Stanley, Idaho with me and go fly fishing. It's pretty phenomenal up there. And as long as they're in a forest fires within 40 miles, knock on wood will have a grand old time. But I'll be in a public library with a bunch of sketches over my shoulder knowing asking why I'm talking and that's legit. That's legitimately The only way to get good Wi Fi within a 60 mile radius. So I'm really excited to join you then. Anything else guys, you want to share? Any last thoughts?

Jordan Choo 1:02:55 I think that's it.

Noah Learner 1:02:57 Dale. Amazing. I'm so stoked. You're with us today.

Jordan Choo 1:03:00 This is really Yeah, everybody. Take care. See in two weeks

Noah Learner 1:03:06 take care of one thanks a ton Dale. Keep automating we love it. Ciao.



Join the official Agency Automators Slack Group

Share and learn about automating your digital marketing agency.

AgencyAutomators is a Bike Shop SEO and Kogneta company devoted to helping your agency become more profitable with automation.

AgencyAutomators © 2019
Sitemap
Terms of Service
Privacy Policy

2019-07-262019-11-10