We’re building a data company. Not only are we in the business of collecting and analysing data, we’re also using it to inform how we operate.
The mission for CampaignDeus is to build the largest, richest and most accurate database on social influencer marketing campaigns. By giving brands and agencies access to industry benchmarks, data-driven recommendations and unparalleled accuracy, we're also bringing transparency to the influencer marketing industry as a whole.
We’re using data and statistical models internally to optimise the decisions we make today and ensure we achieve our goals further down the road. This post gives readers a sneak peek of what we’re doing at CampaignDeus.
Data: the world’s most valuable resource
Data and statistics have been used for decades to improve the way organisations work. In the 1940s, Ford Motor Company employed the so-called ‘Whiz Kids’ to turn around their ailing administration using the tools of management science. Later in the 1960s, prominent Whiz Kid, Robert McNamara, applied the same techniques to improve the efficiency Kennedy’s war machine in Vietnam, as Secretary of Defence.
More recently, rapid improvements in computer processing power and the massive increase in the quantity of data brought about by the advent of the internet have revolutionised the way organisations use data. The Economist recently claimed ‘The world’s most valuable resource is no longer oil, but data’.
There has been a lot of hype surrounding data science and big data over the past 5 years.
While much of what has been said is exaggerated, it is undeniably true that data can, if used effectively, profoundly improve how we work.
Using data effectively
It’s tempting to simply acquire as much data as possible, create a lot of charts and hope some knowledge pours out. In many cases, small amounts of the right data are infinitely more useful than bucket loads of junk.
Our three principles for analysing data effectively are:
1. Ask specific questions
Decide specifically what it is you want to know and formulate it as a testable hypotheses.
For example, ‘Do branded posts in the Beauty vertical perform worse than non-branded posts in terms of relative engagement?’, or ‘Do branded posts shot in bedrooms perform better than posts shot in bathrooms?’
2. Think in terms of models
Models are an abstract representation of some real life thing. We use models to describe and understand how real life processes work.
A model might describe how Instagram likes convert into incremental sales, or how much website traffic to expect given spend on different media.
3. Collect the right data in the right quantity
Statistics derived from the wrong data can very misleading.
For example, many influencer platforms report engagement rates to help brands and agencies decide which influencers they should be working with. But these are typically just the mean average of the engagement rates of their past 6 - 12 posts. What if their last posts weren’t branded and their branded posts perform worse than their non-branded? Are 6 - 12 posts really enough to make a fair comparison? What if one went viral, is the reported mean average still representative of what to expect from future posts?
What we’re doing
We’ve developed technology that ingests thousands of data points on social media posts. We use machine learning to derive certain attributes and a team of human taggers to manually pull off the other, more esoteric qualities. To guarantee data accuracy, we also have a team of approvers, who check each post has been tagged up correctly.
As our database grows we’re able to automate more and more. We’re constantly trainingmachine learning algorithms to recognise the post attributes we’re interested in, increasing the efficiency with which we collect data over time.
Our system looks like this:
It takes time to train taggers and approvers. Once they’re trained, they tag and approve at varying speeds. Likewise, it takes time to find influencers, who post sponsored content at an uncertain rate.
We’re using data and statistical modelling to answer questions like: ‘how many influencers do we need to load into our system per week to ensure we have an 80% chance of having at least X thousand tagged and approved sponsored posts in our system by this time next year?’ and ‘how many new taggers and approvers do we need to employ and train per month to ensure we’re keeping up with new data ingests?’.
A computer simulation uses our assumptions to spin out thousands of examples of what could happen. With thousands of examples of possible futures, we can deduce which are most likely, and alter our input decisions - the number of taggers to employ for example, until we have projected outputs that meet our requirements.
For example, influencers post sponsored content with varying frequencies. Some many times a week; some only every month.
The shape of this chart captures that uncertainty: most publish sponsored content only very occasionally (the peak on the left-hand side), a few publish very often (the long tail).
Similarly, the number of posts tagged per day has a (thankfully reducing!) spread. We can incorporate these data into our simulation model. We then let it loop over and over again, giving us thousands of possible future scenarios for the total number of sponsored posts tagged and approved in our database.
Using this logic we can take a snapshot of how our business will look 12 months from now and, crucially, what the likelihood of each scenario is. Our model tells us we’ll achieve our target of 90% coverage of the UK influencer market by this time next year with a better than 80% chance.
What are doing with this data?
Our data is unparalleled in its depth and accuracy. We are now using it to help brands and agencies make better investment decisions in influencer marketing:
We have created a central repository of brand-influencer campaign data, which helps brands get a better idea of who’s worked with who, and how well influencers have performed in each vertical.
We provide industry benchmarks, helping our clients set campaign targets and allowing them to compare their campaign performance against the market. This is split by verticals, sub-verticals, type of posts, locations shot at, and many more.
Our detailed reports provide data-driven recommendations on how to work with different influencers across different verticals. We answer questions like ‘which are the best post types for this product?’, ‘what emotional tone should influencers be using in their posts?’ and ‘what time of day is best to publish sponsored travel posts?’
CampaignDeus is the leading independent provider of influencer marketing campaign data for Instagram and YouTube. Our platform identifies and classifies brand sponsored influencer campaign performance metrics, tracking hundreds of thousands of posts.
We use this data to provide Brands & Agencies with industry insights across verticals, benchmark campaigns against vertical & competitor averages, and equip clients with in-depth reporting and recommendations on how to make campaigns more effective. Get in touch for more details.
Duncan Stoddard is the co-Founder and Chief Data Officer of CampaignDeus. Previously a data scientist with AlphaSights and media agency Mindshare, he specializes in predictive modelling, forecasting, optimisation and data visualisation.