Automate Actions You Thought Only Humans Could Do
You can think of all sorts of really complex ways of talking about automations, but in the end of the day, it's really that. It's something that we can set up so we don't have to think about it ever again. It will just run, it's scalable, it could be free or very cheap, and doesn't rely on a human being. And as far as I'm concerned, any process that is repetitive is just one that has not been automated yet, quite honestly. So, triggers and actions. There's a couple tools which are worth mentioning. One is IFTTT, which is what this one is showing. The other one is Zapier, which I'll show you a little screen shot of as well, which is much more complex. But with that first step of optimize, we're gonna identify the process, and that's where we really have to start. The second part of automate, that's where we start to look at, where can we actually automate these things. So even though a process itself might be repetitive, it doesn't mean that the whole process can or should be automated,...
but there's almost definitely elements within it that can be. So going back to the example I gave before, of the paying a bill. If part of that process, which it was part of my process at the time, to take the PDF bill or whatever it was, the e-bill. If it wasn't an e-bill, I had to scan it. Everything's an e-bill now. I would have to then send it to my accountant and also I would add it to Evernote. Those three steps right there can be automated with very little effort. There's actually a tool called Wappwolf, for example, which I don't have up here on the slides. But Wappwolf is a really esoteric one that automates actions within Dropbox, for the most part. So you could save a PDF to a specific folder on Dropbox, and then Wappwolf would see that, it would email it to your accountant or whatever, and it would add it to Evernote, just that simple. So you're just taking those steps out of it because, again, we start to just naturally miss the things that are automatable. That's why we have to go through this process. So something like IFTTT is a really great place for people to start with looking at what they can automate. It is not hugely powerful, but it's a really great intro to thinking about what we can automate. Zapier, which again, I'll show you on the next slide, is more powerful, but if you just wanna get in this mindset of thinking about, what can I have as a trigger and what can I have as an action, then IFTTT is the place to start. So the way to work with something like IFTTT is, you go to it, and you look at the services that are included in it, so it probably connects with over 600 services at this point. And I guarantee you that you use at least a third of them. Every day. Everything from Instagram to Twitter to PayPal, MailChimp, Highrise, Internet of Things kind of items, there's just all these things in your life that you definitely use that are on this platform. So you would go to one of them, and you would pick whatever that channel is, let's say it's iOS photos, so that would be that second one there. And you look at the available triggers and actions. Some of the triggers that are available to you with the iPhone are that if you have a new screenshot, that's a trigger. Any new screenshot, or any new photo in a particular area, that can be a trigger. The actions are really, the sky is kind of the limit. So you could say, if I take a new photo on my iPhone, then it can email those pictures to myself. That's pretty basic, so trigger and action. If this, then that. You can sort of tie these things together to have more complex processes, so you can have one trigger results in multiple actions. You could say, if I take a photo on my iPhone, I want to tweet it with this particular hashtag, I wanna email it to myself, I wanna post it to Slack, to my social media channels to that my team can see it, and I wanna add it to Dropbox. So all those four things can happen from one single action. The way to think about this is not necessarily the huge, big processes in you business, but if you think of the things that you're doing every day that take you 30, 45, maybe 60 seconds to do, but you're doing them dozens or hundreds of times every day, that's where IFTTT is really powerful. Which requires a little context to that as well. We as human beings really cannot multitask. I don't know if that's a surprise to you guys, but, (laughs) we really cannot multitask. It doesn't actually exist. The neurological definition is context switching, and that's what it is. We're rapidly switching back and forth between tasks, all day long, and it is exhausting on our brains. Women are marginally better at it than men, like one or two percent, but it's not really a competitive advantage honestly, we're just not good at it. So I'm always telling people, you actually have to treat yourself a little bit more like you're stupid. (laughs) Because we just don't have the working memory to do all the things that we think that we can do. And the feelings of overwhelm that a lot of people experience are really because they get to a level where they just throw their hands up and then the tide washes over them, and that's it, and they can't really do anything else. So anything in this regard that we can get out of our heads, any of those little decisions, they really add up. So that's why I point that out, because some of the actions that you might be able to do with IFTTT could seem kinda trivial, such as if I update my profile picture on Twitter, then I also wanna update my profile picture on Facebook. These are things that are probably important if you're doing marketing, you wanna have the consistent brand image and whatnot. But it's also not something that you should be thinking about, and it's also something that you could make a mistake and forget to do. Now something like Zapier is more complex. Zapier will work with a lot of the more business-minded sort of apps, so you're gonna see, all the CRMs are in here. SalesForce will be in there, PipeDrive, Pipefy, all those kinds of things. You'll see InfusionSoft, you'll see QuickBooks, a lot of those more business-focused ones, and there's no end to the automations you can do there. The other thing is that with Zapier, we can do what they call a multi-step zap, so you can have multiple steps that can happen off of one action. So there's some really cool things that we can do here, and again, you start with IFTTT, which is free by the way, and if you can do it there, great. If you can't, then we can move on to something like Zapier. And going through the motions of this really starts to train you in what is possible. A lot of this stuff is just, people don't know what they don't know. So if you start to look at the different systems that are in here that you use on a regular basis, and start to see what the available triggers and actions are, then you can start to think through what might be automatable. In fact, a lot of times now when choosing a new software for our company, or for me even personal use, I will first see if it actually has a tie-in with something like Zapier. Because that's really what makes us sort of future-proof, and what means it can integrate into our systems in a much bigger way, so that we can do that better. So two big things that I wanna talk about that we've automated in our business and for other clients, and again, you could do these exact things that I'm talking about, but I want this to be examples to you of what is possible. So the first one is our hiring process. At some point if you wanna scale your business, you're gonna have to hire people, and you're gonna have to hire good people. There was a really great study done recently that workplace flexibility is one of the most competitive advantages that you can have as a business. People being able to choose when and where they wanna work. And enabling them to change where they are and when they wanna work if so desired. (laughs) Because the truth is that we as human beings have different rhythms to the way that we do everything that we do. There are better times of the day and worse times of the day for any individual to do any different kind of activity. So to assume that you could ever find two people in the same place that would just be at their best, and have the best graphic designer, the best web developer, in the same zip code, that could both work on a nine-to-five schedule together... It's just not realistic. So, we need to have hiring processes that are scalable, and that work really well. So the way that ours works is like this. We have a online form, which we use Wufoo now. Again, I have to just point out, as I said, the tools are not that important. You could use Gravity forms, you could use Jotform, you could use a Google sheet, I don't really care. This is just what we use, and how they tie together with this duct tape called Zapier. So Wufoo form asks some interesting questions. We have a pretty standard hiring process application. We ask their basic information, where they're from, resume, all that kind of thing. We add in two components that I think are fairly unique. One is a video self-interview, and the second one is a psychometric test, I think is the best way to describe it. So the basic information is important and we have all that. It's sort of a baseline, you have to collect that information. The video one is a fascinating one for us. So it's, ask them to upload a YouTube video themselves, basically just describe their experience, describe who they are. Very simple. So what we found is, probably about 10% of the people who apply are not able to figure out how to do a YouTube video. So that's a fairly good litmus test for us, and they might be really great at something else, but we kinda need them to be able to have that technical ability, or to at least follow those instructions. So sometimes we'll see a private video uploaded so we can't watch it, or it's uploaded to a different platform. And that's a really good screener, honestly. If you think about game theory, and having people play into that or not. So that's the first one. The second thing then is the video. And this is fascinating, because we give very little direction. So we always ask, just tell us about yourself and your experience. What we've found, first of all, is that people who do a video that's less than a minute are probably not gonna work out. Because less than a minute usually means they just didn't really care, and didn't put a big effort in. And it's not like that's a hard cutoff for us, that's just something that we've found. The other thing is that if they go over three minutes, that tends to be a bad thing too. Not because they were too verbose, but because they were actually really uncomfortable, and there's a lot of pauses and ums and uhs. And if they're not comfortable by themselves on a video, they're probably not gonna be comfortable on a video with a client when it really counts.
Now that you've told us this process, are you gonna have to change it?
No, I'd love to see more videos in that range, you know? And still gonna see their personality. It's still gonna be genuine. It's very genuine, that's the best way I can put it, is super authentic. I was thinking about that, I was getting to the second part actually, but no, I won't have to change it. So we've actually seen people who were reading off the screen, like and those are usually over three minutes, and those are not very good. It's not a training opportunity or a learning moment, it's really just seeing how somebody is at that moment. We do a lot of training in the company, but there are two things that I really don't believe are trainable, and this is live (mumbles), so people may disagree with me, but I don't believe that you can train productivity. I just don't think that you can train somebody to go that extra mile and make sure that that extra step is taken. And I also don't think that you can train attention to detail. And I don't have it, by the way, to be fair. I am not very good at attention to detail. But it's something I don't think you can train. I think that that's something that's sort of in people's wiring, and it's just the way they are. And I certainly will gloss over things to move quickly. I'm much more nimble, I wanna move quickly rather than get bogged down in details, and there's a place for that, and there's also cons for that. However, with the work that we do, we really need people to be proactive and attentive to detail. So in the video, you kinda see that, honestly. It's a really good way to do that. And again, the less direction, the better. The second part of that is, like I said, the psychometric test. So I just have to see how I can present this to you without giving you the right answer. So what it is is, it's a paragraph that is from an actual task that we were given by a client. A year ago or so. So the task was, I'm staying at a hotel, and they name the hotel, and I have 1,000 stamped envelopes that I need to mail. Can you find the nearest post box or place I can mail them. Hopefully it's within walking distance. That's it, it's a very short paragraph. Now this was a real task, and it's also the sort of task we refer to as a one-and-done, meaning that the assistant could pick it up, get the answer, give them the answer, and then they're done. There's no real back-and-forth with the client. So we give them the paragraph, and we say, "What would you say? "What would you do?" So we get quite a big range of answers there. And a bad answer would definitely be, "I'll check into that and get right back to you." That would be a bad answer, and we've seen that. Another bad answer is, "The nearest post box is 3.5 miles away. "It's a 45-minute walk. "Have a good day." (laughs) Also not a good answer. The right answer is that they actually called the hotel to find out from a concierge where the nearest place is, and it's a lot closer than that post box might be. So I'm not gonna give you the actual answer. (laughs) And I'm not gonna give you the actual answer, not because I don't want somebody to get it, but because it occurred to me after several months of doing this that we actually should call the hotel and apologize to the concierge staff for all the calls that they were getting. Because people were starting to name the concierge that they had spoken to. It's like, "Oh, I might wanna call Robert "and say why they're getting all these calls." So anyway, it works out really well, because it's a real thing, it shows that they took five minutes of their time to actually take that extra step, they didn't just google it. Because that's a big one too, there are services that you can pay to google things for you, but that's not what we do. And that's not what a lot of services are looking for. So both of those things are in this Wufoo form. Now, a couple things happen that kick off this process. When somebody fills out this form, and now, by the way, we have four different tracks, we have the generalists, we have design, developers, and I think we have graphic designers as well. Or design, rather, sorry. So they fill that out, and any automation where you can set up where a branch of things happen, is really kinda cool, so multiple things can happen at the same time, concurrently. You don't have to have things sequential, they can be parallel. So first of all it posted into a Slack channel for our teams, because it pops up with their name, their video, and who referred them, and 90% of the time that's it. But sometimes there might be a little bit of discussion about that particular candidate, because somebody knows who they are, or because they might just have a comment on their video with their application or whatnot. Otherwise, they just sort of keep going through the process. The other thing that happens is that they get loaded up into Process Street. As I showed you before, Process Street has that hiring application. I can go back to it to show you guys for better reference. So there you go, review YouTube, US Postal Service question, their resume, everything. The way Process Street works is that you have a template. So we have a hiring process template, and then it will run an instance of that template for each person. So this is automated, so when they fill out that form, it automatically creates a new checklist, and that checklist will have the person's name, their first name and last name. It'll have actually the YouTube video link. Everything'll be pre-populated in there. And the nice thing about this is that anybody can go in there and do this hiring process. So the other day I was giving a talk and I was showing people this, and I don't manage this process anymore. I actually have never managed this process in its current form. And I pulled it up in front of this audience and I went through a process, and I brought somebody on board. (laughs) Hopefully. So it'll prepopulate that information, and then we can go through and, if we approve them, and again this takes you through each step, so the postal service question is good, the YouTube video's fine, great. So if they move through all these steps, then it will automatically take them forward through a background check, and there's a number of services now that allow us to do those checks, not manually, automatically, with an API. If they pass the background check, it can greenlight the application and proceed further, and it can keep going that way, and it can take them through a personality profile, if you wanna use something like Kolbe or Meyers-Briggs or whatever your business requires. And then it can even put them into a testing environment, and take them through that entire process. So it's almost like an assembly line. As it were. And all of this stuff is based on automations. So as I was saying, actually the background check is a really good example. So if you optimize the process, you identify, then you start to see the places for automation. So when you first do the process, you might say, "OK, well now we're gonna have the personal information, "and we're gonna go to our background check company "and give them this information, "and then when they give us a green light, "then we'll come back and move them along. It's a really good opportunity to be like, "Huh, I wonder if there's any background check companies "that have some sort of automation we can integrate into." And then you just plug and play and get that in, and that's what we did. The first process required manually sending in information to the background check, and manually moving it forward. And eventually we ended up switching providers to a company now we use called Onfido. And they can automate the process into ours.
Kind of a technical question, what if you have a really important step you don't wanna accidentally automate, like you wanna maybe have two people look at it for it to move forward?
Great question, so I would add that into the process itself, that step one is that you look at it, step two is another person looks at it, whoever that may be. And the nice thing with Process Street is you can assign individual steps to individual people. So what you can say is that it's assigned to you first, if you want, and when you approve it, the next step is then assigned to whoever that second person might be. That's a really good question. And that's what I was saying before, about how you can build in the checks and balances as necessary. And again, it's very fluid. So you can decide, after a week or two, "we don't really need the second one," that's fine, so just remove it. And then the process continues the way it should. So I was saying before how we can do something like machine learning and artificial intelligence for free, OK. So this is a very specific use that I'm doing this for, but this hopefully will, again, open up your mind to the possibilities that we can do. So there is a tool now called MonkeyLearn, and again, I have to keep saying this, but the tool is not the most important thing, this just happens to be one way of doing it. MonkeyLearn is a artificial intelligence, natural language processing platform, that essentially, there's a couple different forms of machine learning. And the one that's most relevant here is where it's literally learning by watching. That's what some forms of machine learning are. In fact, there's a robotic chef that's under development right now. It's really cool, with like an arm and everything. And it watches professional chefs slicing, dicing, sauteing, and that's how it learns to do the things it does. Because a lot of times, human beings can't program an algorithm, but they can show it, and the computer can make the algorithm based on that. It's kinda like the idea that it's easier to act your way into a way of thinking than to think your way into a way of acting. So it's a lot easier to smile, physically, and try to make yourself be happier from that, than just be like, "I'm gonna be happy, I'm gonna be happy." So that's kinda like machine learning, machine vision, it's all related. So the way that MonkeyLearn works is that you give it a lot of information, whatever kind of information you want. I'll give you some examples, but any kind of information you want, and then it can categorize it based on what you teach it. So for example, you could show it a bunch of pictures of cars, and you could say, "This one's Italian, "this one is German, and this one is American." And go through that, and eventually it should be able to be shown a picture and say that it believes, with a certain percentage of accuracy, that this is one of those three things. It can also do sentiment analysis, meaning that it can look at an email and say, "This is a negative email," or, "This is a positive one." Which is great for customer service. If you're getting 1,000 customer service emails every day, and you just wanna look at the ones that are negative, it can categorize them for you and bring those to your attention. All these things are possible. Another thing that's relevant to our business, to Leverage, is that when a task is issued, one of our value props is that we can do anything, and we really push that to the limit a lot of times, as long as it's legal. When a task comes in, currently, a human does have to look at it and say, "Well, generally speaking this is a marketing task," or, "this is a design task," or, "this is a research task," for example. We're gonna be able to very shortly use machine learning to do that, and the benefit there is that there's less mistakes of course, it happens instantly, 'cause right now a person has to look at it and get to it. And we don't show anybody to actually be able to work on it until it's categorized correctly. So the categorization is a big thing. So I had an idea for how I could use this. And I wasn't sure it was gonna work, but it did. There was a service called Help a Reporter Out. Which, that's one service that, yes, that service matters. It's a really good service, and everybody should know about it. It's pretty incredible. What it is is that, I think seven or eight years ago, this guy created this service that basically connects reporters looking for sources, with sources that can provide that kind of information. So you might on a given day have somebody from Forbes who's writing an article about stay-at-home dads who have businesses in the toy industry, and then somebody else who's writing an article for some obscure medical journal who needs people who have had hair transplants and live in Arizona. Whatever, there's all these very specific things. And there's three emails a day, and there's a hundred items usually in each of those emails, and it's the kind of thing where, it's like, the term fomo was invented for this, the fear of missing out. Because if you don't open it that one day, that's when you miss that story that somebody could do. And it's really cool, 'cause it's totally free, they're real journalists looking for sources, and you can then write back to them and say, "I'm that kind of person. "Write about me!" So over the years I've gotten over 60 pieces of press from responding to HARO requests, some really good ones, and everything in between. And I've had a lot of clients use this over the years and also look at how you could make it more efficient. Now the problem is, this is a really tricky one, and why I like to use this as an example. And also, I do think this is relevant to every business, 'cause what I'm eventually gonna show you here is how to create an automated PR machine, basically. So over the years people will always try to automate this, and optimize it, and I've done that as well, but it's tricky, because it's a hard thing to spend money on, and have someone look at, because the hit rate is really low, quite honestly. So in a week you might see one or two things that are even interesting enough to pitch, and then if you pitch, you might not even get it. So it's really hard to justify paying an assistant to read through these, and if you wanna try to automate it, keywords don't really work. Because, for example, let's say there's an article in Men's Health coming out about entrepreneurs who overcome chronic illnesses. Great, clearly I would fit the bill for that. However, an article for me in Men's Health would really just be for vanity purposes. It wouldn't necessarily serve a greater mission. However, if that article is in Entrepreneur magazine, about how entrepreneurs with chronic illnesses deal with stress, these kind of things have come up. That's great, I would wanna go for that. But how would you describe that as a keyword? Right, it's really challenging. So that's where the machine learning stuff comes in really handy, they're the things that you can't quite describe, you kinda know the answers, but you can't really describe them. It's like trying to define is. So, I had this theory that I could show MonkeyLearn HARO requests and tell it what was interesting or not interesting, and have it learn from that. I'm gonna tell you conceptually how that happens, and again, all this is to open up your minds to how we can automate crazy things in our businesses. Because quite honestly, that's the kind of thing that somebody would easily hire a full-time person for, or a publicist, or PR. And at some point in your business, you probably need to do that. But when you're starting out, you definitely don't. It's not a good use of your money. But you really can get a big bang for your buck. The first part of this was getting the data into a form that MonkeyLearn could actually read. MonkeyLearn is actually pretty flexible. It can look at Facebook posts, or Gmail messages, or an RSS feed, which is like from a blog, or a spreadsheet. There are a couple others too. The HARO email is very heavy just text, it's all text, with little links for emails. And it's structured in that there's a query name, and a paragraph, and an email, but it's just a whole list. And there could be 140 in one email, there could be 70 in the next email. And the length is all over the place. So there's really, it's a challenge. So there's sort of two ways to do it with Zapier. One is a little bit more accurate, but required a lot more time. So the easier way is to have it come through as an RSS feed. So there's actually web sites out there that have already done this. There's one called HARO2RSS for example, which gives you the RSS feed of the daily ones, and it seems to miss some for some reason, but it's good enough. The better way to do it, which is a really cool thing you can do in Zapier, is Zapier has a built-in email parser. So what that means is you can show it an email, and you can actually highlight parts of the email and say, "Hey, this part right here, that's the query. "This part is the query description, "and this is the email of the reporter," and you actually identify it that way. You have to do that for about 100 of them, but then from then on, every time it sees an email like that, it will know what that information is, and it can actually pull it out into a structured format. So that's fine. So once we have the structure, great. So we start with Zapier, and this is not the recipe for this, but I can explain. So we have a trigger, as you can see there, trigger, action, action, right? So the trigger in this case would be new RSS item. If it's the HARO RSS or the one that you created, that's fine. Then it shows it to MonkeyLearn. That's the second step. Now on the side of this, you have to teach MonkeyLearn how to do this, and because again, I said before that something in Men's Health might not be interesting to me, but something in Entrepreneur might be, it's very individual to you and what your needs are. So you feed this into MonkeyLearn, it'll show you the two of them, and I wanted to do something very simple. I just wanted interesting or not interesting. Those were my categories. And the only issue there is that when you have something where the categories are very skewed, so in that kind of situation, out of 100, I might have two or three that are interesting. So it's not a lot of information to show a machine what's good, so the more, the better. So I did 69 entries with it, and after that it was 87% accurate based on that. So the more you show it, the better. But I was just saying, "OK, interesting, not interesting, "interesting, not interesting," just going through it, and it kept showing me a sample, and I kept categorizing it. That was it. So then I have the box that it can go into, so Zapier takes the new entry, shows it to MonkeyLearn, and then I have a filter that says only go forward if it's interesting. And then after that I have a couple of choices. So what I ended up doing was just having it send me a Slack message, with the interesting inquiry only. So now, once a week or so, once or twice a week, that's all I see. I get a Slack message with the interesting inquiry, and it's really good. And it's, again, like an automated PR machine. Now if you wanna take it even a step further, you could say that based on those certain keywords, you will actually email the reporter with your standard pitch, and then it's fully automated. So I haven't taken it that far yet, but it's really easy to do, depending on what you want to express. So if the keyword is entrepreneur, maybe you have your pitch about your life as an entrepreneur, who you are. If it's about social media or it's about whatever your product might be, then you can have a standard thing that goes to the reporter and says, "Hey, I'm this person, this is my history, "saw you're writing an article, "happy to contribute," kind of thing. So I know that was kinda fast, I went through that pretty, does that make sense? Yeah, OK.
You could have it do an email draft, as a safeguard, 'cause you wouldn't want this thing to go crazy if there was some tech issue.
So one of the available triggers for Gmail, I'm, sorry, one of the available actions for Gmail in Zapier, is either send new Gmail message, or create draft. So you can absolutely do that. Or you could just say, send it to an assistant or something, or someone else in your team, just have them read it for 30 seconds and decide if it should go, but yeah. So that is what's possible with automation nowadays.