Offline Attribution Demystified

Offline Attribution is the most recent holy grail that the marketing community across all media channels – digital, mobile, out-of-home, TV and even print are chasing at large. With the availability of human mobility data at a reasonable scale, it is believed that it may be possible to answer a very simple question for the modern-day marketer.

Offline Attribution is the metric that aims at measuring the effect of data-driven marketing campaigns across any channel (digital-online, mobile, OOH, print, TV) in driving foot traffic to offline brick and mortar locations of interest. The simple thesis is that one needs to be able to measure the effectiveness of campaigns in the offline world. The idea is motivated by the notion of being to measure the lift in online page visits to an on ecommerce or website from online campaigns. Can we do the same thing offline? (now that human mobility is available via apps, beacon systems, Wi-Fi hotspots and more)

As companies start working on this, it is important to recall the history of online digital attribution. It is still not a solved problem. Much ad hocery exists with notions of first and last touch points. Baseline metrics do not exist – as to what is good versus what is bad even in the online world – Is a .0001 % lift in online page visit rate good for a 10000K $ campaign? A lot of endorsement is anecdotal. Given this state of online attribution, one needs to be sanguine about how the notion of offline attribution will evolve. We have a long journey to traverse as a community.

Current Scenario and the Challenges

There are a number of vendors offering offline attribution. Some are truly global whereas many operate in more mature markets, primarily the US. Some key factors to consider when choosing a vendor:

  • What is their primary data source of consumer visits to a geo-spatial location? How is a location defined/tagged? What is the resolution of a location fix from a consumer? Precision? How is it anonymized? Are any privacy requirements violated? What is the base level margin of error? Sources of noise? Does one use app data? Wifi Hotspot data? Beacon data?
  • ID resolution – how does one ensure that the consumer identifiers used in ad delivery can be realistically linked to the identifiers on visit data? There are many combinations to be dealt with here – cookies, mobile ad identifiers, beacon ids, hashed mac ids and more. Given ID matching limitations, scale of actual data can vary considerably.
  • Do they work with a pre-defined panel (a pre-defined set of consumers as a baseline) or work with data from all consumers? Each approach has its pros and cons in terms of brand coverage, spatial coverage, data quality etc.
  • How is a visit to a location defined? confirmed? validated? Does the vendor own the acquisition of the ad exposure data? How long should a consumer stay at a location within a margin of error? Is it pass-by/trajectory data? How many location fixes should one get from the region of interest? Are time intervals considered – for example only pings during store working hours ? (or should you consider only the active hours).
  • Methodologically – assume we clean the data and there is enough scale to actually do a reliable measurement – how are baselines computed? What is the lift relative to? How long after the campaign is offline attribution computed? How do you account for other concurrent campaigns in other channels? How do you account for other brands also running campaigns to the same set of locations?
  • Finally, reporting of attribution results is another key step – Do any absolute numbers make sense? Are they stable? If indices are reported, do you understand the index baseline – the reference audience used? If your target audience is skewed with respect to the reference, how do you expect your results to skew? What real follow up decisions can you make based on this data? If multiple locations are being analyzed – how do you want to understand each location? Does an aggregated metric make sense? How does a marketer run multiple tests to evaluate the stability of metric?

Each vendor has an approach – data, methodology and reporting – that covers the basic questions discussed above at different levels of quality, scale and more. It is also important to know how they extrapolate the data, what are they transparent about, are they willing to share the raw data for the marketer to triangulate with another. Claims are made on accuracy and much more but it is early days for Offline Attribution.

What are the best practices for a customer?

Overall, there are a few additional conceptual issues to consider from a marketer’s viewpoint:

  • Considering that human mobility data is sparse and also exhibits power law behaviors (that is some locations are far more active than others) – what is the best way to measure offline attribution? Should I consider all stores of a brand or only some stores? What are the flagship locations that act as reliable indicators of campaign effectiveness?
  • How should I triangulate across multiple vendors? How do you even compare results reported?
  • When you run campaigns in different global markets – are the human mobility behaviors, really reflecting your target consumer in each market? How do you validate?
  • When campaigns are hyper-local – local effects become important – such as neighbourhood characteristics – spatial layouts, store locations, active traffic areas, lifestyles and more – Does the raw data capture any of these?

Contrasting with online attribution

Offline Attribution is a much harder problem than Online Attribution for a few obvious reasons. These include –

  • Addition of the spatial dimension – a physical visit is much harder to register than a visit to a landing page.
  • Combining two different worlds of data – the online world where exposures happen and the offline world where it may be potentially measured – There are many sources of noise.
  • Scale issues along with the quality-quantity trade-off exacerbates reliable measurements.
  • Finally, the human behavior that leads to a visit to a store may be for many more reasons – nothing related to an ad that popped up on their mobile phone while they were visiting a website – so the causality problem is quite acute. In the online world, most page visits are part of a click thru reaction from the consumer – actively seeking information or comparative information for a product or service. So assuming that your ad caused the page visit is far more valid than an offline visit.

However, much can still be done and offline attribution metrics utilized effectively as the world goes mobile.

What does Near do for Offline Attribution?

Given the various nuances to consider while estimating Offline Attribution, we at Near have been developing a rigorous approach to develop a reliable metric for the industry. We have analyzed different kinds of data from cradle to grave across different channels in different markets. Allspark, Near’s enterprise SaaS product offers its customers the results of our learnings. Our R&D work on this continues at pace. We continue to improve our data and methods and benchmark with the best in the industry.

If you have any questions or would like to know more about Allspark and its offerings, please contact us here.