Creating deterministic and probabilistic Next Best Action (NBA) metrics.
This Treasure Workflow allows for rule-based (deterministic) models and ML-based (probabilistic) models for building Next Best Action recommendation systems in real-time.
The deterministic logic is executed using SQL code. The SQL code adds specific NBA attributes to users’ profiles through workflows in real-time, using page-views data coming from our JS SDK. Examples of deterministic segments include:
- cart_abandon (retargeting)
- new_user_visit (awareness)
The workflow also adds additional custom web-behavior attributes to the user_master table. For example:
- channel_ad_response_percentile
- activity_period_engagement_percentile
These percentile attributes are used to train ML models and create Probabilistic NBA Segments to layer on top of the deterministic rules. They help in the execution of specific marketing strategies and help target users wherever and whenever they are most likely to respond to an ad.
Examples of the Probabilistic Segments can include web behavior for a propensity to engage with:
- Channel Affinity -- a certain marketing channel. For example:
Facebook, Email, Display - High Activity Period -- an ad during a certain time of day. For example:
morning, afternoon, evening, overnight
Why is Next Best Action important and how can it add value to marketing efforts, optimize campaign performance and improve KPI?
Using these recommendations can ultimately guide marketing strategies and campaign optimization techniques by allowing marketers to target users more strategically and dynamically in real-time.
By knowing How, When, and Where might be the best way to target different segments of users, marketers can ensure that the frequency and spend for different marketing campaigns is aligned with the specific web behavior that their users have exhibited in the past.
This way marketing budgets can be spent strategically and KPIs such as CPA, ROAS, ROMS etc. can be improved and ad $$$ waste can be reduced. Think of this like a trading algorithm that would tell you which stocks to buy next and when/where to buy them, so you could optimize your ROI expectations given historical data on the stock price action and statistics to predict future performance. More info and visual illustrations on how an example Next-Best-Action driven campaign orchestration can be executed and activated within Treasure Data's CDP.
Our NBA approach aims to answer three main questions concerning the next best marketing action that a marketer might want to activate against a specific user - What? Where? When?
NBA: What?
The deterministic models typically answer the question - "What?" (what is our next best marketing action to take against this particular user?) For example, if a user has added something to a cart recently, but has not purchased yet, our system adds the user to the "abandoned_cart" segment within the CDP in real time.
Our Next Best Action recommendation system will then recommend that we retarget this user more aggressively because they have recently shown a high intent to purchase. The same could be done for many different marketing rules that can be predefined and customized by you within our workflow and attached as attributes to the user_master table.
NBA: Where?
After the "What" is determined for a particular user by a deterministic model, we then layer an ML predictive algorithm for Next Best Action, which typically answers the question - "Where?". What marketing channel might be the most effective for this user to drive them further down the funnel.
Users are scored on a daily basis by our ML algorithm, based on their browsing and ad-response data, and assigned a propensity score to engage with a certain channel - Social, Mobile, Email, Display etc.
Our Next Best Action Recommendation System will tell you "What" specific marketing action to take against different segments of users in real time, and "Where" that marketing action is most likely to generate the best results.
NBA: When?
The "When?" works similar to the “Where”, but looks at what time of day a marketing message might be the most effective for this user to drive them further down the funnel.
Users are scored on a daily basis by our ML algorithm, based on their browsing and timestamp of event data, and assigned a propensity to engage score during a certain time period - morning, afternoon, evening, overnight etc.
Our Next Best Action Recommendation System tells you what specific marketing action to take against different segments of users in real time, where and when that marketing action is most likely to generate best results.
This allows you to spend your marketing budget strategically and efficiently at scale in real-time. The algorithm logic and rules can be customized within the workflow per your request to best fit your specific business goals and use cases.
Use Cases
- Audience Segmentation
- Campaign Orchestration
- Dynamic Marketing Budget Allocation and Targeting
- KPI (ROAS/ ROMS) Optimization