Can you provide more insights into how Brand Affinity is calculated? I understand it's based on the user’s email/mobile engagement, types of messaging users are engaging with. However, does this include app engagement? How are the different message types weighted?
We have a disproportionate number of negative affinities compared to loyal and positive so far. Can we expect that to even out over time as more data is collected on our contacts?
To kick off our first AMA of the year, we received multiple questions about how we measure Brand Affinity, and to explain a little bit more, I'm going to hand it off to my colleagues to explain a little bit more.
@Anthony Chiulli go for it!
Will there be a way to track how users move from one Brand Affinity score to another, in either direction to see trends across our user base?
@Tiffany Cornelison
Question: Can you provide more insights into how Brand Affinity is calculated? I understand it's based on the user’s email/mobile engagement, types of messaging users are engaging with. However, does this include app engagement? How are the different message types weighted?
Answer: Thanks for kicking us off! We received a handful of questions around how Brand Affinity is calculated.
To generate Brand Affinity
labels, we look at email, mobile push, and in-app engagement across all of your messaging. More recent engagements are weighted more heavily, with newer engagements weighted more heavily than older ones. We also dynamically determine how far back in time to look based on how frequency contacts in a project are messaged.
Our AI looks at the percentage of messages a contact has interacted with; we do not want to ""penalize"" anyone because they aren't being messaged. Maybe your contacts want to hear from you! However, there is a slight boost for contacts that interact with more messages overall.
We include transactional messages as well, because eCommerce receipts and even password resets are positive signals for a brand, but they are less important to the AI than marketing messages.
For anyone interested in more detail, we have hundreds of individual datapoints (features) that we include as inputs ot the AI, and output a single number (scalar) that predicts a propensity for future engagement. We then turn this number into the Brand Affinity label that you see. The boundaries of where we draw the line between labels—for example, between positive and loyal—is determined dynamically for each project. So the results you see are unique to your business and the way you message your contacts in a given project.
@Ryan Glanzer
Question: We have a disproportionate number of negative affinities compared to loyal and positive so far. Can we expect that to even out over time as more data is collected on our contacts?
Answer: Thanks for your question. In short yes. The more data we have on your customers interacting with your messaging, the more thorough our AI can score and predict their affinities. It also may depend on your frequency of messaging and available data we have on your customers to score.
@Jessica Balis
Question: Will there be a way to track how users move from one Brand Affinity score to another, in either direction to see trends across our user base?
Answer: That's a great question and is something we are looking into now. I can't give you any timelines, but we're actively looking into how we might provide this type of reporting.
"we have hundreds of individual datapoints as inputs to the AI"
Like what? Like more pieces of data that revolve around and provide context to the recency and frequency on engagement, or what? Could you provide any more details here?
@Anthony Chiulli
Any plans to incorporate SMS into the affinity scores?
@Davida Gaffney
Question: "we have hundreds of individual datapoints as inputs to the AI"
Like what? Like more pieces of data that revolve around and provide context to the recency and frequency on engagement, or what? Could you provide any more details here?
Answer: Not to get too far into the weeds, but part of feature engineering for AI products is to determine all of the possible ways raw data can be rolled up and used as inputs for the models. Examples might include the number of opens over the last 7 days, over the last 28 days, over the last 90 days, etc. We also have features that are the outputs of other machine learning algorithms.
Our exact set of features and the models themselves are part of our secret sauce, but I will say this is something our data science team has been working on and improving for quite some time.
I am using a formula similar to NPS to track movement week-to-week. Where total ((loyal+positive)/scored total)-(negative/scored total)=brand affinity score.
My question is more of an opinion on whether this is a good use of the scoring field.
We do not (yet) use your mobile SDK for messaging, rather the API solution. Does that impact this feature?
@Marc Mancuso
Question: Any plans to incorporate SMS into the affinity scores?
Answer: Currently, we cannot track SMS interactions because our tracking links are too long. And without that feedback (connecting the prediction that someone would click with whether or not they actually clicked), we cannot build or improve models. I will say this is an area that we are looking to improve, so while I can't give any timelines I will say stay tuned.
@Marc Mancuso
Question: We do not (yet) use your mobile SDK for messaging, rather the API solution. Does that impact this feature?
Answer: You're good. Our system was built so that we could support both API and SDK customers on mobile. We wanted to make things as flexible and powerful as possible and not lock you into a specific implementation.
@Cora Martin
Question: I am using a formula similar to NPS to track movement week-to-week. Where total ((loyal+positive)/scored total)-(negative/scored total)=brand affinity score.My question is more of an opinion on whether this is a good use of the scoring field.
Answer: This is an interesting use-case, one that I have not heard of to date. I would caution trying to leverage your users' Brand Affinity score distribution across your customer base to map to a NPS score. They are not 1-1 per se. Brand Affinity is more designed to measure how your audience is engaging with your messaging rather than an aggregate promoter score.
@Davida Gaffney
Thanks for pre-submitting your question!
Question: Can we customize this score further, say with internal data we have about customers? - How are your customers using Brand Affinity?
Answer: "Brand Affinity
is currently calculated based on opens and clicks for email messages (transactional and marketing) and opens for both mobile push messages and mobile in-app messages. Customizing how Brand Affinity is calculated with additional data points is being considered for future iterations.
The versatility of Brand Affinity being a contact property, allows customer's to utilize Brand Affinity labels across the Iterable platform an in numerous use cases. For example:
Loyal and Positive
Offer exclusive VIP deals to brand champions
Triggered a reward/perk workflow when a user changes to a loyal/positive affinity label(s)
Encouraging loyal/positive users to become Brand Ambassadors, send NPS surveys
Measuring ROI on loyal/positive users to identify their Lifetime Value
Accelerate the IP warm-up process with positive and loyal affinity customers by boosting positive engagement
Neutral and Negative
Send automated re-engagement campaigns to negative affinity customers at risk of churn or unsubscribing.
Assist in repairing deliverability issues by suppressing messages to negative or neutral affinity users to improve IP and domain reputation.
Triggered a reactivation workflow when a user changes to a negative affinity
General Affinity Label Use Cases:
Gain insights into distribution of your audience based on engagement across channels.
Dynamically populate content in templates using Handlebar language, Catalog, and Data Feeds to match each customer with personalized products, offers, and content based on affinity labels.
Nurture customers to move into the next step of a customer lifecycle with targeted and meaningful messaging.
Export an audience’s affinity labels to use for retargeting campaigns through external platforms such as Social
"
@Maxi Schicho
Thanks for pre-submitting your question:
Question: What are best practices/ use case for an app?
Answer: During our beta, we had one customer use Brand Affinity in testing cart abandonment push messages in their app. They were seeing relatively high uninstall rates form push marketing - aroiund 3% - and wanted to leverage Brand Affinity to try and reduce uninstalls. They began to only sending their cart abandonment push campaigns to users who are either Positive and Loyal - while holding out all other affinity labels - and saw a 30% lift in order rate and 0 uninstalls!
@Tiffany Cornelison
Thank you for pre-submitting your question!
Question: Also, how frequency is defined for each brand level.
Answer: The AI looks at, on average, how frequently contacts in a given project are messaged and uses that to determine how far back in time to look and decide how important older messages are. So for example, the AI might look back a few weeks for a brand that messages their contacts every day or a few months if a brand messages every week.
@"Meli Musson"
Thanks for pre-submitting your question!
Question 1: Are the definitions of loyal vs positive Brand Affinity based on averages specific to my company or an aggregate of all/or similar Iterable customers?
Answer 1: Brand Affinity
scores and associated labels are based on engagement behaviors of your users specific to your project. They are not influenced or derived from other Iterable customer's user behavior.
Question 2: Are we able to edit the criteria for affinity if we would like them to be more or less strict?
Answer 2: Not currently, but this is a great idea and something we will keep in mind for the future.
@Davida Gaffney
Thank you for pre-submitting your question!
Question: Assuming BA is calculated based on engagement, do you think it is reasonable to view BA scores as positive vs negative when we aren't measuring disposition but some measure of frequency of engagement? Is is reasonable enough to think that more likely to engage = positive "affinity"? I am not saying it isn't, just would like some other marketers take on this.
Answer: At its core, Brand Affinity looks at previous interactions with a brand to predict how likely someone is to interact in the future. We could have easily used the term "Most Likely" instead of "Loyal", but at the same time your Loyal Brand Affinity contacts are also the most likely to interact with your messaging, so conceptually they're very similar.
@Cora Martin
Question: I am using a formula similar to NPS to track movement week-to-week. Where total ((loyal+positive)/scored total)-(negative/scored total)=brand affinity score. My question is more of an opinion on whether this is a good use of the scoring field.
Comment: While I agree with
@Anthony Chiulli, this is giving me some great ideas that I'm going to take back and discuss with the data science team. Thank you for your question and feedback!
@Netti Worms
Thank you for pre-submitting your question!
Question: How are will the affinity for the different channels (especially SMS and push) be measured?
Answer: Right now, Brand Affinity takes into account email, push, and in-app interactions and blends them into a single, cross channel score that predicts how likely someone will engage with future messaging. We currently do not include SMS in the calculation because we have no good way to track opens or clicks in a text, but hopefully, we will be able to address this soon.
@Ryan Glanzer
Question: Are there benchmarks for all organizations as a whole? I want to know how our org's scores stack up against others.
Answer: Unfortunately not currently, but there are some data privacy concerns around sharing benchmark data across customers.
Will there be a way that we as customers or at least our account managers can see more details about how users are being placed in the brand affinity category on our particular project? (timeframes you're looking at for engagement, percentages that place users in a category, how clicks are weighted different than opens, etc.).
Not knowing this information for us makes brand affinity difficult to use or explain at an executive level. Every company differs of course and has different customer lifecycles so having more detailed clarity on how your technology is working would enable us to understand and utilize it.
Erin Watkins
Thank you for pre-submitting your question!
Question: I'd like to leverage this for reporting. Will there be a view where I can see the growth/decline of each group and how they are trending over time?
Answer: Not currently, but we agree this would be extremely useful and it is something we are looking into now