All industries have ups and downs, and marketers are constantly tasked with finding the best audience-offer fit in a world that is far from linear. At Tinyclues, we are often asked how our algorithms work, and more recently, how they adapt to changes in the market and the associated consumer behaviors. Can this technology really “understand” what is happening? How does it make sense of all the data points and give us accurate and up-to-date predictions? Find out more below.
Predictions in changing situations
Algorithms are deliberately built to adapt to the world in which they inhabit, taking into account changes in the market and providing contextualized results. Tinyclues works on first-party data, that is, your customer data. Data from the last 30 days is especially important, as it helps us make optimal predictions. This recent data, combined with your transactions and other data points from the past 24 months, help create robust consumer behavior predictions.
The good news is that our Deep Learning technology is autonomous and always on the lookout for new behaviors from your refreshed data. It can confidently predict buyer propensity scores with only 30 transactions, as part of your larger dataset. So, when you have just changed your offers, or when your operations resume, you can have accurate predictions in a very short amount of time.
Just a tiny reminder
- In order to provide you with the most up-to-date predictions, our technology is using your data that is refreshed on a daily basis.
- If you have temporarily shut down operations, any predictions made on older data risk being suboptimal.
How Tinyclues can help you adapt, right now
Many of our customers have varying business objectives at different times throughout the year.
Updating your success criteria with your CSM is essential, as this helps Tinyclues find the best path to meet these objectives. Just as when you went through the initial set-up phase, we can also continually adapt your settings to better define what type of behavior you’re looking for (repeat buyers, clickers, viewers, etc.).
For example, models can be optimized for clicks (business objective: engagement) compared to purchases (business objective: increase revenue), as well as adapted to prioritize online buyers over store buyers. In light of the current crisis, many customers are opting to optimize for clicks as part of their branding efforts, which also helps you ensure high deliverability rates.
Contact your Tinyclues CSM for more information on how we can work to help you meet your current business objectives.
The effects of market changes on audiences
Data science can sometimes seem abstract, we know! Let’s take a look at a large jug of water as an example.
Sixth floor walk-up? No problem.
Who is the buyer for this jug?
“Normal” buyer profile: Buyers located in rural areas who do a large monthly shopping trip in a larger city, with a car to transport it back home.
Recent buyer profile: Rural buyers are still in this profile, but now we also find urban buyers without a car.
Why are these city folk suddenly appearing in the audience?
Rural buyers still need water and are going to continue doing their monthly trip to stock up on supplies, no matter what. But the newcomers are urban buyers who are perhaps a bit panicked about water availability—in a health crisis, for example. Not having a car and living in a city are no longer factors that stop them from buying giant jugs of water. They want this jug, even if it means schlepping it up 6 flights of stairs!
The product itself hasn’t changed. But the type of buyer has evolved. Location is just one of the many factors that influence the audience, but it is the most easily recognizable to the human eye. Therefore, as a marketer, when you decide to promote this jug of water in your next campaign, your audience will be filled with an – updated – more varied buyer profile—both rural and urban buyers.
A quick overview of Tinyclues’ technology
Recent buyers are used to define the behavior you want to encourage with your marketing campaign. But the magic comes from the finesse and quality of the latent variables found by our algorithms in your deep history of data. Essentially, these latent variables are all the implicit signals, or hidden “tiny clues” that can’t be directly observed, but that really help predict the right audiences for your offers.
Unlike many other data solutions, we do not aggregate or simplify your data to make it easier to process. At Tinyclues, we do not pre-filter or pre-format beforehand—we want to leverage the millions of raw data points you have about your customers because this helps us create a deep understanding of their complex behavior.
To do so, we have built unsupervised and multi-layer algorithms that learn from large-scale patterns in customer relational databases. They can process extremely large amounts of raw data and autonomously extract very powerful latent variables from them, without any preconceived notions about the source or industry which the data comes from. This is the only way to capture the complexity of a changing product catalog or fluctuating customer behavior patterns.
Long story short, the last 30 days of data will tell you who is currently buying your offer, and the core of the solution will help you find your audience by using latent variables. Thanks to these, our platform will search across your entire customer database to find similar profiles that match the buyers you have, right now. These buyers will never be the same from day to day. Just as the world is constantly changing around us, so is our behavior and propensity to buy. The good news is—so are our algorithms.
One last example for the road
We had an interesting case last year with a popular video game sold by one of our customers.
Was it a gamer or a grandma who just bought FIFA?
For the soccer-obsessed, FIFA comes out every September, much to fans’ delight. So, who are the buyers for this game? Well, this depends on when we’re looking.
Release day: Generally, hardcore gamers who are willing to shell out the top price for a new video game. These buyers typically have a larger disposable income, so they are probably not teenagers.
3 months later, December: Generally, parents and grandparents who want to give this game to their children and grandchildren for the holidays. They have a higher income level and purchases tend to drop off sharply after Dec 23rd.
6 months later, March: Generally, gamers who don’t want to pay full price for a variety of reasons: perhaps they have a less disposable income (teenagers for example), or they are not giant fans and can patiently wait for a more reasonable price, etc. They will almost certainly buy this FIFA video game at a discounted price or second-hand.
Just as in the first example, the product didn’t change, but the type of buyer changed over time. This is based on many factors (seasonality, external market conditions, trends, etc.), with the most recognizable factor in this example being the proximity to the release date. Tinyclues’ technology is able to recognize all these patterns from the latent variables processed by our algorithms.
Our technology is built to continually adapt, which is great news for companies and brands who go through changes in their environment. Perhaps the change is a sudden must-have item in fashion, a new trendy destination, or challenging market conditions for example.
Whatever the changes are, whenever your campaign date is and whatever the offer is, our technology can find the best audience in minutes, and is constantly taking into account last minute changes. Our Deep Learning algorithms are working to keep up—not only with the market trends and consumer behaviors, but also with your business objectives.
We’re here to help! Contact your Tinyclues CSM for more information.