6. Creating Your Diminishing
Returns Graph
www.click.co.uk
Budget
Revenue
Logarithmic
Line of Best Fit
Use Excel to find your line of best fit
7. Channel Budget Revenue ROI
SEO £1,000 £4,000 4.0
PPC £1,000 £3,000 3.0
www.click.co.uk
Example
Target ROI: 2.5 to 1
Budget Budget Budget£1000£1000£1000
Revenue
Revenue
Revenue
SEO PPC Overlay
£4000 £4000
£3000 £3000
“Which channel would you rather allocate an additional £2000 to?”
“Now which channel would you rather allocate an additional £2000 to?”
8. Example (cont)
www.click.co.uk
Channel
Old
Budget
Old
Revenue Old ROI
New
Budget
New
Revenue New ROI
Inc.
Budget
Inc.
Revenue Inc. ROI
SEO £1,000 £4,000 4 £3,000 £7,000 2.33 £2,000 £3,000 1.5
PPC £1,000 £3,000 3 £3,000 £8,000 2.67 £2,000 £5,000 2.5
Budget BudgetBudget
Revenue
Revenue
Revenue
£1000 £1000 £1000£3000 £3000 £3000
£3000 £3000
£4000 £4000
£8000
£8000
£7000£7000
SEO PPC Overlay
Additional
£3000
Revenue
Additional
£5000
Revenue
9. Transforming your Revenue vs Cost
Graph into Profit vs Cost
www.click.co.uk
Budget
Revenue
Budget
Profit
It’s much easier to see the point of diminishing returns on the profit graph
12. Referral
Domain Sessions Conversions
Assisted
Conversions
Referral 1 16 0 0
Referral 2 2 0 0
Referral 3 32 0 0
Referral 4 60 0 0
Referral 5 51 0 0
Referral 6 21 0 0
Referral 7 10 0 0
Referral 8 55 0 0
Referral 9 15 0 1
Referral 10 17 0 0
Referral 11 19 0 0
Referral 12 6 0 0
…………… … … …
Low Volume Data Sets
www.click.co.uk
“With the data so diluted, how can you be sure which of these assisting
referral domains is performing the best?”
13.
14. www.click.co.uk
Pass vs Drop
PPC SEO Social Display
Display SEO Email
PASS PASS PASS PASS
PASS PASS DROP
In this case we count Email as a ‘drop’ for not leading to another interaction
17. With vs Without Analysis
www.click.co.uk
PPC SEO
PPC SEOSocial
+1%
2% Conversion Rate
3% Conversion Rate
The presence of a Social channel in the path has
increased conversion rate by 1 percentage point.
18. Example
www.click.co.uk
Maximising Last Click Channel Profit
Channel Budget Last Click Channel Profit Attributed Profit Total Profit
Channel A £7,000 £10,000 £7,600 £17,600
Channel B £2,000 £6,000 £6,500 £12,500
Channel C £10,000 £11,000 £4,800 £15,800
Totals £19,000 £27,000 £18,900 £45,900
Budget
Profit
Profit
Profit
Budget Budget£7000 £2000 £10,000
£10,000
£6000
£11,000
Channel A Channel B Channel C
Maximising each silo individually leads to £45,900 total profit
19. Example (cont)
www.click.co.uk
Maximising Last Click Channel Profit
Channel Budget Last Click Channel Profit Attributed Profit Total Profit
Channel A £7,000 £10,000 £7,600 £17,600
Channel B £2,000 £6,000 £6,500 £12,500
Channel C £10,000 £11,000 £4,800 £15,800
Totals £19,000 £27,000 £18,900 £45,900
Maximising Total Profit
Channel Budget Last Click Channel Profit Attributed Profit Total Profit
Channel A £6,415 £9,664 £9,463 £19,126
Channel B £3,163 £5,240 £6,034 £11,274
Channel C £9,422 £10,638 £6,195 £16,833
Totals £19,000 £25,542 £21,692 £47,234
Attribution Rules
• For every £1 spent on Channel A:
Extra 50p profit for Channel B
Extra 30p profit for Channel C
• For every £1 spent on Channel B:
Extra £1.80 profit for Channel A
Extra £1.35 profit for Channel C
• For every £1 spent on Channel C:
Extra 40p profit for Channel A
Extra 30p profit for Channel B
Maximising based on the Attribution Rules leads to an extra £1334 total profit
20. Recap
1. Find out your profit vs cost graphs for each of your channels in isolation
2. Look at up-weighting high engagement channels alongside traditional attribution models
3. Review metrics such as ‘pass completion’ when final conversion data is diluted
4. Up-weight further any ‘passes’ that move visitors to different stages in
Awareness>Consideration>Conversion funnel
5. Use with vs without analysis when data is significantly large
6. Combine your attribution rules with profit vs cost calculation for each channel in
isolation in Excel
7. Maximise total profit after attribution rules are applied using Solver
8. Regularly review if your budget allocations actually are driving the expected bottom line
profits
9. Adjust your attribution rules, and add more in where necessary
www.click.co.uk
21. Thank You
Dave Karellen Head of Paid Search
Email david.karellen@click.co.uk
Website www.click.co.uk
Further info www.click.co.uk/blog/why-everybodys-doing-attribution-analysis-wrong
Any Questions?
Notas do Editor
Hi everyone, I’m Dave Karellen, Head of Paid Search at Click Consult. Today though, we‘re not going to be talking just about Paid Search specifically, but instead about data and measurement in general. See, I’ve always been interested and fascinated by data, but what’s always fascinated me most is how little people know about how to correctly use the data available. Now, there’s so much data available these days that it can be overwhelming to decide how to make sense of it, so how can we actually gain insight, and moreover actionable insights from all of this.
The purpose of this session is to answer a very simple question, which is how best to use the data available to decide on how to allocate budget to achieve the maximal revenue. So often, I would get asked by clients “Which channels should I put more budget towards in order to drive the highest profits?” And at this point I have to resist the temptation to just say..
So to answer this question, we are going to look at attribution with a fresh perspective, and actually go some way in answering some of the important questions that are often avoided in this field- such as how to decide which model works best for your business. However, we will begin by considering these channels in isolated silos, as the basic principles need to be understood before progressing to a point where we consider how each of these channels interact together, otherwise we risk running before we can walk.
Fundamental to understanding the reasons for ‘pushing and pulling’ certain marketing channels is understanding the law of diminishing returns. Ad lib
Ad lib, then: So the critical question, is where does your ‘point of diminishing returns’ hit? Two advertisers may have exactly the same revenue vs cost graph, but each business’ point will be different based on their respective business margins, different lifetime values, etc. It’s all about looking at gradients as in previous example to decide upon incremental ROI, the big question is at what point does the incremental ROI become break-even. Now anyone who is particularly good at calculus would be able to take the curve’s equation, differentiate, and solve for the cost that results in the equation’s differential equalling the break-even incremental ROI. However, and this always surprises me, not everyone loves calculus.. I know!
So instead, you can use an alternative method that starts by transforming our revenue vs cost graph into a profit by cost graph. This takes into account all the margins, other costs, and the budget itself into consideration. Now, here the diminishing returns point is much more clear. We want to maximise the returns, so to find the cost needed to do this, we could again turn to calculus to find the differential of this curve, and solve for the cost of the stationary point, i.e where the differential is equal to zero- making sure of course you differentiate again to ensure that the stationary point is indeed a maximum. Nobody fancy that method? I don’t know what’s wrong with you all!. Okay, there’s a tool that can help us out here. Ad lib
Ad lib, then: This all works perfectly if we assume that all channels work in silos. However, this is not the case. Channels work together to create an end to end customer journey. Pushing one in favour of another may lead to cutting off the source of assisting conversions, if our measurement is focussed solely upon last click results.
Looking at how channels work together is the realm of cross-channel attribution. Most people when they think of attribution think of the popular models. Ad lib However, this tends to be all people think of when they consider attribution- these positional models. More valuable is looking at engagement metrics from each channel. For example, the time on site, the number of pages viewed, and the type of pages viewed- e.g blog vs product pages vs basket. Each of these is then correctly up-weighted, for positive engagement metrics. To bring you back though, I promised a fresh look on attribution. You’ll notice that the positional method and even the engagement based methods are all built on paths to conversion. Everyone has become engrained in the idea that the best way to decide where to allocate budget is through measurement of current success. However, this viewpoint only lets you see things as they currently stand.
If you only have a couple of channels to analyse that is fine, but when you are dealing with, say hundreds of referral domains, only getting about 50 clicks each per month with a 1% average conversion rate then it can be quite difficult to decide how best to correctly attribute due to the scarcity and dilution of the data. It may be that some of these domains could be hidden gems, but as a lot of budget isn’t already being funneled into it, we can’t see how good they actually are. In cases like these, how do you decide which channels are performing the best? To use a football analogy, if you’ve got a team full of misfiring strikers..
then you need a way of discerning which of your midfielders (assisters) are actually putting in a good shift, even when the end goals aren’t coming in. The best way to look at players like these is to look at stats such as their ‘pass completion rate’ to decide how effective these assisters actually are when the end goals don’t come. After all, to what extent can you really blame the midfielder who makes a great pass only for the striker to consistently fire wide?
So rather than looking solely at conversion paths, it makes sense to look at other attribution metrics, such as the ratio between a channel leading to a different interaction compared to when it was the last interaction in a chain without a conversion. Ad lib This would be the attribution equivalent of a pass completion rate, which can be a great metric to use when conversion paths are in short supply. You wouldn’t expect a football manager to not be able to tell how any of their players performed, even if the game ended nil nil. One of the biggest questions you must now be asking yourself is exactly how you can measure this ‘pass completion rate metric’. It is difficult, considering most Analytics tracking tools are naturally focussed on conversion paths. To achieve this, you generally have to employ a couple of ‘hacks’. One such way would be to set up a goal in Analytics that sets every interaction as a tracked goal In this way, you can view all paths, not just conversion paths and find out how often each channel results in another interaction, or if it is the last one in the path with no conversion, i.e they have given up or gone elsewhere.
This type of method is so important as it ensures that everything fits correctly into the awareness>consideration>conversion model. We have to always remember that behind the data are actual people following this basic funnel towards an end goal. Ad lib
Ad lib, then: The important thing to remember here is to test if the actual uplifts meet your expectations based on the rules, go back and look deeper into the data and refine more where necessary. Remember, you’re never going to find a ‘holy grail’ of attribution, there’s always going to be insights you haven’t yet uncovered, or data that can’t be measured 100% accurately such as offline attribution. But by digging out insights one by one, and applying them on top of your channels’ cost vs profit curves, then you can start to make incremental improvements to your bottom line.
Overall, I hope this has given you a taster of the different ways in which we need to re-examine the attribution question, and give you some understanding on how to actually put this into practice to maximise your overall returns, which you should never forget is what measurement should always be about. Thank you all for listening. I’d love to hear any questions you might have.