How To Use Excel To Unlock AdWords Secrets

Editor's Notes

• #2: Hello everyone and thank you Michael. Today I’m talking about Excel and how you can use it to unlock AdWords secrets.
• #3: A quick introduction. Hi I’m Sophie and I’m a director at Brainlabs. Brainlabs is the fastest growing digital agency in the UK. And we do all things biddable. That means we get involved in paid search, paid social, programmatic, analytics and we event build some really cool tech. I originally studied maths at Oxford and then went on to working in consultancy. Even though my short time in the corporate world was one I’d rather forget, I learnt a lot of important skills, in particular I learnt a lot of Excel.
• #4: So today I’m going to be talking a lot about Excel. Now for me, Excel is a bit like marmite – there are those who love it and those who hate it and indeed those who simultaneously love and hate it. I’m definitely in the love camp, although there are times like any relationship were I can go a bit crazy at it. So I’ll tell you a bit about all the things I love and hate about it.
• #5: My love of Excel partly stems from one thing: shortcuts. They make your life so much more simple and you feel like you’re part of some Excel cult by knowing all the secrets one. So from F4 which fixes formulae with dollar signs, to CTL SHF DOWN to highlight a row of data to my favourite and more obscure ALT +E+S, which allows you to paste special and do wondrous things like transpose and paste values and number formatting at the same time! So I really love shortcuts. On top of shortcuts, Excel is also useful for data visualisation and quick results to show your clients and bosses. And as a mathematician, I can’t ignore the formulae. In fact I like formulae so much, it’s become an interview question at Brainlabs. So I do a lot of the recruitment at Brainlabs and I always ask candidates what their favourite Excel formula is. There is no consensus yet, although I do get some weird looks. On the other side, there are a few limitations to Excel that are frustrating. The limit on rows can cause problems as can the speed of formula calculation. But overall Excel is a really good thing.
• #6: There are further limitations of Excel that can be solved by running macros, which can automate your tasks. To run a macro you need to write with a language called VBA. VBA is cool because it allows you to loop through data and perform lengthy tasks quickly. The best part is that you can record your actions with the Record Macro button. When you click on this, your actions are coded up as you do them. Once you’ve recorded them, you can take a look at the code by opening the editor.
• #7: And this is what is looks like. You can find your code in Module 1 here. And then on the right of this image is what it looks like. Every Macro begins and ends with Sub Macro1() – or whatever you want to call it – and End Sub. It’s best practice to comment your code, so someone else can quickly understand what is going on. You can do this by preceding text with an apostrophe. And then you’re free to write whatever you like.
• #8: Without getting too much into VBA, the most important thing to know is how to define variables. And there are four components of this. Dim declares the variable. Then call your variable something, here it’s my_variable (remember no spaces), and the you have to declare the variable’s name and specify its type – is it a string or integer or another type.
• #9: And as a last point before I begin with the meat of my talk, all the data that you will see is random. And not only that, but it has been generated by a useful function called RAND(), which returns an evenly distributed random number between 0 and 1
• #10: So today I’m going to unlock some adwords secrets with Excel. And I’ll start at the beginning of any PPC journey, and that’s with campaign creation.
• #11: So when thinking about a campaign build, there are a number of questions you may ask – first and foremost, what do you want to advertise and what keywords do you want and how many? Next, a big challenge with automation is to still ensure that ad copy is as relevant as possible. And finally, what is the overall structure for the account – how do you want to split campaigns and how many campaigns and ad groups do you want and need
• #12: So let’s get started with keywords. Let’s imagine we’re an online shoe retailer. What keywords do we want to bid on? There are the obvious shorttail keywords, but we really want to make sure we have all the longtail covered in our template as well. So I like to think in terms of tokens: that is qualifying each word in a keyword. Here I have colour, material, type, noun. And we can use concatenate or & because no one actually uses concatenate, to make all the combinations. So our first combination is red canvas womens shoes and so on. It’s good to keep track of the number of possibilities you have as you go along so as not to max out the number of rows in Excel. So with this data here, we have 6 colours, 3 materials, 5 types, and 3 nouns which gives 270 possibilities. So you can see it can get very big very quickly.
• #13: Next think about how many keywords you want in each ad group. If we use keyword planner to find all combinations of keywords with high monthly searches, we can start to build ad groups. So here for the ad group, red leather unisex trainers we have 6 keywords as shown, which would give a total of 1,620 keywords in total. And remneber to delete low search volume keywords to declutter your account
• #14: Now for ad copy. This is the part I like best and thinking about hierarchies is one approach. Create a template of all possible headlines, DL1s, DL2s etc and list them from most relevant. You count number of characters with LEN. Remember to use substitute to replaces spaces with dashes or plusses in urls
• #15: To pick out the most relevant, longest possible headline we use index matching. Index matching is one of the most powerful tools in Excel. I’m sure a lot of you have used vlookups and hlookups but index match takes this to the next level and is much more flexible to use. In my opinion, there are two schools of thought: those who use index match, and those who don’t know about index match and thus use vlookup. At Brainlabs I even go as far as to ban vlookups. Index match combines two formulae INDEX and MATCH. If we look at this example here, I’ve used an IF condition in the Binary column. This outputs a 1 if the character length is less than 25 and if no row above is already less than 25. So you can see that ‘Buy Red Unisex Trainers’ is the first such cell, and has a 1, whereas at the bottom ‘Buy Red Leather Trainers’ is 25 characters but has a 0 because there is already a 1 above. We use this information with index match. The first input is the array that you want your output to come from – so in this case it is the headline column. The other input is the row number and for this we use the match formula. This takes the match cell – in this case we input 1, and a match array – in this case the binary column. So index match works as follows: match finds the 1 in the binary column and then outputs the adjacent cell in the headline column. Simple eh.
• #16: So now we’re almost done. Right back at the beginning we could have used excel to product all the different combinations of keywords, but actually there’s a much quicker way to do this and that is with VBA. VBA is great because it allows you to loop through sets of data rather than having to rely on formulae. So here we have four loops, one for each token – shoeColour, shoeMaterial, shoeType and shoeNoun. You notice that I’ve already defined four variables with are the shoeColour, shoeMaterial, shoeType and shoeNoun to make it simple to write out. The syntax for a For loop in VBA is For i (or j etc) = 1 (or any number) To another number Then and ending with Next You can also create a macro to delete all duplicates at the end of the campaign creation.
• #17: On a final note, this is crucial to remember at the end. When you’re automating a campaign creation it’s important to check for errors. Obviously you need to make some human checks but you can also check for error messages with countif function and a simple Macro. Countif takes a range, and counts based a condition – here it’s if a cell equals ‘#n/a’. I combined this with a message box that pops out if the count if >0. A simple way to check for errors in the matching
• #18: The second part of campaign management is budget modelling.
• #19: So I’m sure you’ve all used excel to put together a basic budget before. Typically we put together a budget monthly or annually for a client and model how we should spend in order to maximise return. Impact factors to consider are any YoY or MoM changes – for example CPC or conversion rate You also will want to segment your data based on device, location and day of week This is how we produce our budgets. Plot budgeted spend and actual spend together. As well as lines for how we should spend to meet budget and how we will spend if we continue at the current rate
• #20: But what about the million dollar question. If someone asks, if the budget increases by x%, what is the return? What is the answer – Is it A – maths argh!. B – I have no idea. C – X%. Or D – let’s use bid simulator. Anyone know the correct answer? Anyone want to phone a friend? Well the correct answer is D and I’m going to explore bid simulator more now.
• #21: You can find bid simulator at campaign, ad group at keyword level. You get data for all important performance metrics based on what would happen at different CPCs. This data allows you to model the potential impact of a budget change on your conversions. However an important caveat, is that bid simulator gives you data for the last 7 days and this does not always predict future performance
• #22: This is what bid simulator looks like at campaign level. It shows where we are currently as well as performance numbers for an increase or decrease in bids. You can see this data plotted on the rights with clicks and cost. The gradient of the chart is cost divided by click so CPC and you can see the line getting steeper, which is expected as CPC increases to get more clicks
• #23: So how do we look at bid simulator in practice. Well at keyword level, you can add columns for estimated clicks and cost with different bid changes. If you download this report you now have all the data you need. You can filter data however you like, but the aim is to sum the estimated click and cost and get a table that looks something like this. The left column has the bid multiplier, so if we were to multiply clicks by 50%, 100%, 150%, or 400% we can expect clicks and cost to increase like this. You can also see CPA increasing as well which we’d expect as we become much more aggressive with our bids.
• #24: Now we’ve got the table, we want to find some relationship between cost and conversions. I like to use a polynomial model, and here is a formula to analyse your data. Linest is a really powerful formula, which takes in an array for data for a y axis, and data for an x axis. Here I’ve specified a quadratic polynomial by the {1,2} but you can use any polynomial up to order 6. It’s also optional to remove the constant, and include stats. And this is the output. So one thing that you may have noticed it that the output is not in one cell it’s in an array. You’ve actually got 5 rows and 3 columns. And you can also see this with the curly brackets around the linest formula. To output an array with a formula you need to highlight the array area and do control shift return. The data that you get in row one is most important: it’s the coefficients for your model. The other data is statistics which tell you how accurate your model is.
• #25: Linest is actually most useful when combined with INDEX, just like INDEX MATCH before. We can use INDEX LINEST to find coefficients in our quadratic polynomial. When I say coefficients, I mean the a, b, c that you see here in this equation. How does it work. So again linest is the ouput array, and we want row 1, and input the column number to output the correct coefficient. So to output a, we put 1 for column 1.
• #26: Once we have coefficients we can build a cost versus conversions table. Use Cost as your X value, and the coefficients to write a polynomial formula to give you conversion numbers. You can also plot this to see visually where you are and what the diminishing returns are.
• #27: And now we put it all together. We have the basic model from the very beginning. We have our bid simulator model. Together we can output different budget scenarios to give us an idea of what the return will be. Instead of manually changing numbers each time you want a different budget scenario we use a macro. It uses Goal Seek to automatically output different scenarios. So here we input three budget scenarios, hit run simulations and we get an output straight away. Great right.
• #28: The third part of a PPC toolkit is bidding. There are a number of ways to use Excel for bidding and I’m going to speak about one of these, namely ad scheduling.
• #29: Ad scheduling allows you to bid up or down at times of the time that you convert well or poorly. However there are restrictions in AdWords – you only have a maximum of 6 bidding windows a day but what happens if you want to make changes more often this. Well 3rd party tool can’t help either – they can only make up to 4 changes a day. We have a tool which takes an Excel input and allows you to bid differently for each hour of each day. And I’m going to explain how we put together the Excel model now.
• #30: So when we started working on this challenge, we identified two problems to avoid. One – at times of lower volume, conversion rate fluctuated a lot and we needed a model to take that into account And two – the challenge is a massive maximisation problem.
• #31: Maximisation may seem obvious but it’s not as easy as it looks! Take this example here of DVD A and DVD B and let’s say we have a $20 CPA target. You can see how the stats differ for A and B in different positions. Now if we combine these together and our aim is a $20 CPA what is the best combination? You may think it’s combination with DVD A in position 1, and DVD B in position 2 but actually that’s the wrong answer. Actually by scaling back on A and pushing B we get a larger overall number of orders for the same CPA. And this is our aim with ad scheduling but this time looking at every hour of every day which means a lot of possibilities!
• #32: When putting together an Excel template, I always separate of any inputs from any calculations. So I have an input sheet where you can dump data just like here. In this case download from Dimensions tab with Day of Week and Hour of day segments. Then use pivot tables to get clicks and converted clicks.
• #33: Quickly for those of you who haven’t come across pivot tables before, they look a bit like this and there are pros and cons. For me the biggest pro is that it’s easy to get quick analysis done which can give you some really great results. There are some added bonuses like using pivot with the SQL server, and calculated metrics which come in handy. However there are also some cons: charts are inflexible with pivot tables calculating additional formulae off a pivot table doesn’t work – you have to copy the values out first. Errors can easily slip in if you’re not careful – if someone doesn’t understand that a metric should be calculated then that can cause a lot of problems. And lastly, with a lot of data, pivot tables are very slow and indeed have caused my Excel to crash many times.
• #34: So going back to ad scheduling – we have our inputs and the first problem we have is that conversion rate can be unstable from hour to hour. This doesn’t really make sense as the conversion rate of a site shouldn’t vary from 4:59 to 5 o clock. Simple exponential smoothing not good enough so we use 5-wide weighted smoothing window. The weightings for the average can be set as above with symmetric or asymmeteric past and future smoothing. And you can see the effect here – so the overall distribution looks similar but the statistical stability is much greater in the image on the right.
• #35: We output the smoothed values with a heatmap to give us an idea of trends. Make sure you smooth clicks and converted clicks to find conversion rate, rather than smoothing conversions rate which is incorrect as you’re smoothing averages.
• #36: So far, so easy. Now comes the most technical part of the model. And the great thing is you don’t have to understand a single thing of this page to actually use the model, it does it all automatically for you. The main thing is we are now taking into account evidence to determine the probability of our hypothesis. It also incorporates greater uncertainty over the rate at times of low volume. P(A|B) = posterior – probability hypothesis is true given evidence. P(B|A) = evidence – probability of getting this evidence if hypothesis were true. P(A) = prior – probability of A being true, before evidence. P(B) = prob of B over all possibilities
• #37: An important part of any model is the constraints. We set how much we’re willing to have our bid increased and decreased by so here +/- 35% and we also set how much our weekly clicks should vary and we typically set that to +/-5%. This is make sure we maximise conversions with roughly the same number of clicks. We also built how a feature to take into account advertising hours. So if you’re a call centre and you can only take calls during working hours, then you can set that here.
• #38: Now I don’t know if you’re ever plotted the distribution of conversion rates for the 168 hours in a week. If you have, you’d find it looks a bit like this chart. There are a few really low and really high conversion rates, and the majority are concentrated in the middle. So we therefore chose to model our bid multipliers with a hyperbolic tan function like this. And before you finish the most important part is to check you model has worked – calculate the number of conversions before and after to see the effect of you bid multipliers.
• #39: And this is what the output looks like. I now leave it to one of our tech team to implement our 24/7 bidding tool and you’re sorted with ad scheduling.
• #40: Lastly I’d like to talk about competitor tracking.
• #41: So once you’ve got your campaign set up, you’ve sorted out budgets and you’ve thought about bidding, you want to consider competitors. There are some 3rd part tools out there but they will never be as accurate as Auction Insights data
• #42: Auctions insights data is really cool but can be somewhat limited. Here are some questions you may be asked about your competitors – who are my toughest adwords competitors and on what terms? Now that relatively simple on auction sights by just selecting the campaigns or keywords you’re interested in and looking at the stats. The next question is slightly harder - what is the impact of a competitor’s change in strategy in the last 3 months. Again you can pull a couple of auction insights reports are see how the numbers have changed. You can also tie it back you CPC. And lastly is the hardest question, and much more open. How has the competitive landscape changed over the last year. This is tough as auction insights doesn’t make it easy for you to track competitors over time. So that’s where Excel comes in and that’s what I’m going to speak about now.
• #43: So this is what auction insights looks like. You get stats for things like impression share and average position. Rather annoyingly auction insights is not available via the API which means there is no option to code up the data, we have to rely on spreadsheets.
• #44: Start by downloading the auction insights report with a time segment – this can be day, week or month depending on how far back you want to look. In the auction insights template, there’s a setting sheet to configure everything. Here you can see I’ve selected days and day interval 7 which means I’m looking at data on a weekly basis. I use a simple macro which creates a list of all unique competitors. This uses the delete duplicates formula and combines it with some simple VBA. A final part of the set up is to decide whether new competitors or old competitors are more important to you. I usually think new competitors are more important and therefore I select backwards direction which means start at the most recent date and work backwards.
• #45: Now here’s the Excel challenge. For each time period you will find there are a different number of competitors. This makes thing hard from an Excel point of view when setting up a template because you never know how many competitors you’ll have. There are in fact two approaches – firstly set an upper limit of competitors, say 40. Or don’t set an upper limit – and I’ll talk through how to do both in Excel.
• #46: Firstly I’d like to say that I tend to set an upper limit. This is because no client that I’ve ever worked with has more than 40 important competitors and therefore it’s over the top to consider more. If you make sure you sort your competitors by impression share over time you’ll always have your top 40. So assuming we have a limit, we can set out our data as below. I use the offset formula to sort the data. If’s really useful in a situation like this. Offset takes a cell, and moves you across and down with rows and columns just like this. It’s a simple and powerful function to use. To order the data I use median first and then rank to order. Rank function works by comparing a cell with an array. So for example competitor 1 here is ranked 8.
• #47: The other option is to set no upper limit. In this situation you need to use index match match, and here is how it works. Your output array is now make up of multiple rows and columns, all you data. To find your row number, take the cell you want to match and compare it to the array, and do the same with your column number. And then your output is where you row and column meet.
• #48: So now you’re almost done. It’s nice to add functionability to input different competitors so you can change up what you see. Then chart all this over time and add in your CPCs to see how they’ve been affected by competitor behaviour. And you’re ready to send straight to your client or boss.
• #49: So today I’ve talked about using Excel during the PPC lifecycle. From Campaign creation, budget modeling, bidding through to competitor tracking.
• #50: And I’ll leave you with my favourite motto: keep calm and use excel!