An ML based mostly strategy to proactive advertiser churn prevention | by Pinterest Engineering | Pinterest Engineering Weblog | Could, 2023

Pinterest Engineering
Pinterest Engineering Blog

Erika Solar ML Engineer | Advertiser Development Modeling Staff; Ogheneovo Dibie Engineering Supervisor | Advertiser Development Modeling Staff

Old, rustic boat sinking in ocean — Photo by Jason Blackeye on Unsplash
Photograph by Jason Blackeye on Unsplash

On this weblog put up, we describe a Machine Studying (ML) powered proactive churn prevention resolution that was prototyped with our small & medium enterprise (SMB) advertisers. Outcomes from our preliminary experiment recommend that we are able to detect future churn with a excessive diploma of predictive energy and consequently empower our gross sales companions in mitigating churn. ML-powered proactive churn prevention can obtain higher outcomes than conventional reactive guide effort.

Like many ads-based companies, at Pinterest, we’re intently targeted on minimizing advertiser churn on our platform. Historically, advertiser churn is addressed reactively. Particularly, a gross sales particular person reaches out to an advertiser solely after they’ve churned. This strategy is difficult as a result of it’s extremely troublesome to “resurrect” a buyer as soon as they go away the platform. To handle the challenges with addressing churn reactively, we current a ML-powered proactive strategy to advertiser churn discount. Particularly, we developed a mannequin that may predict the probability of advertiser churn within the close to future and empowered our gross sales workforce with insights from this mannequin to forestall in danger accounts from churning.

On this weblog, we cowl the:

  • Churn prediction mannequin’s design and implementation
  • Experimentation within the managed North America SMB phase

Our workforce constructed a ML mannequin to foretell advertiser’s churn probability within the subsequent 14 days. We use the Shapely Additive Clarification (SHAP) package deal to estimate the mannequin’s options’ contribution to the churn prediction. We offer the mannequin churn prediction together with high contributing options to gross sales. Gross sales makes use of this data to prioritize their effort to mitigate churn for advertisers in danger. We’ll discuss every element in additional element within the following subsections.

Mannequin Structure

The preliminary model of our mannequin relies on a snapshot Gradient Boosting Choice Tree (GBDT) structure. We selected GBDT for the next causes:

  1. GBDT is a broadly used mannequin with good efficiency on small to medium sized tabular information* (our information matches on this description).
  2. SHAP works effectively with GBDT to estimate options’ contributions.
  3. Mannequin characteristic significance is simple to generate with GBDT.
  4. It could additionally function a superb baseline mannequin for future mannequin enhancements, e.g. a sequential mannequin.

*Snapshot means we use all the data out there as much as a given timestamp to foretell the churn chance within the subsequent 14 days with respect to that timestamp.

Goal Variable

After thorough evaluation and session on the enterprise wants, we determined to make use of the next goal variable definition (see Determine 1).

7/01 to 07/07 is 7 day spend >0. 07/07 to 07/21 is 14 days. 07/21 to 07/27 is 7 day spend >0 ? If yes, then Label 0: active. If no, then Label 1: churn.
Determine 1: Goal Variable Definition

For our use case, we distinguish between an lively and churned advertiser as follows:

  • Lively advertiser: spent within the final 7 days
  • Churned advertiser: no spend within the final 7 days

We solely predict the churn probability for lively advertisers. Particularly, we predict if they are going to churn within the subsequent 14 days.


There are over 200 options used within the mannequin. These options are aggregated throughout totally different statistical measures–e.g. min, avg, max and so on — over a spread of time home windows such because the previous week / month previous to the inference dates. We additionally embody week over week and month over month change options to replicate latest developments. These options may be grouped within the following classes:

  • Efficiency: impressions**, clicks, conversions, conversion values, spend, value per 1000 impressions, value per click on, clickthrough price
  • Purpose: purpose attainment ratio, distance to purpose
  • Funds: price range and utilization
  • Adverts supervisor actions: creates, edits, archives, customized experiences
  • Property: gross sales channel, nation, trade, tenure, dimension, spend historical past
  • Marketing campaign configuration: focusing on, bid technique, goal sort, marketing campaign finish date

**View greater than 1 second.

Characteristic Contribution

We use the SHAP library to estimate the characteristic contribution to mannequin chance output. Sigmoid of the sum of the options’ SHAP contribution is the same as mannequin chance. From SHAP characteristic contribution, we are able to know what the important thing drivers are of excessive churn chance. We then spotlight them for the Gross sales workforce to forestall churn.

We use an offline skilled mannequin to deduce lively advertisers’ churn chance each day.

Churn Threat Class

To assist the Gross sales workforce higher perceive the which means of the mannequin output, we classify accounts into three classes based mostly on their churn chance: excessive, medium, and low churn threat. Excessive churn threat captures the accounts which can be principally more likely to churn with excessive precision. Medium churn threat captures the accounts which have a decrease probability of churn. Low churn threat incorporates the ‘wholesome’ accounts which can be unlikely to churn within the subsequent 14 days. We choose the thresholds to outline totally different churn threat classes in line with the Gross sales workforce’s request of desired precision and recall. Extra particulars may be present in Experiment End result.

Our first experiment was targeted on SMB accounts in North America which can be managed by Gross sales Account Managers (AMs). We cut up the advertisers randomly into remedy and management teams throughout the experiment inhabitants. For the management group, we don’t make any modifications to the prevailing Gross sales workforce procedures. For the remedy group, we supported the Gross sales workforce to forestall churn with the next data:

  1. Churn Threat Class: Excessive / medium / low churn threat
  2. Churn Purpose Class. We categorized the detailed churn causes into coarse churn classes to ease understanding. The Gross sales workforce carried out investigations utilizing churn classes as instructions.
14 Day Churn Prediction Model — Overall Churn Risk High. Churn Category is Performance and Campaign Setup / Best Practices. Absolute Change in 14d Churn Risk % D/D is -11% down.
Determine 2: Churn Data Widget

Experiment Success Metrics

Our experiment was evaluated based mostly on the next standards:

  1. Mannequin predictive energy, i.e. how effectively our mannequin is ready to determine advertisers which can be more likely to churn
  2. Efficacy of churn prediction in churn discount

Mannequin Predictive Energy

In an effort to decide the mannequin’s predictive energy, we in contrast its on-line efficiency on the management group (i.e. AMs who didn’t have entry to the churn predictions) to what we had noticed offline throughout improvement (i.e. our out-of-sample analysis). Particularly, we measured mannequin efficiency based mostly on:

  1. Mannequin high quality: We in contrast the AUC-ROC and AUC-PR noticed on-line to offline.
  2. Churn threat segmentation: In session with gross sales, we decided thresholds for prime, medium, and low churn threat classes in order that:
  3. Recall in excessive and medium churn threat must be above 70%.
  4. Precision in excessive churn threat must be round 70%.

This permits gross sales to seize most accounts liable to churning whereas additionally prioritizing the right way to work via them, i.e. excessive churn threat first (highest precision).

With respect to mannequin high quality, our outcomes point out that the AUC-ROC noticed on-line is inside 1% of the offline AUC-ROC and the web AUC-PR is inside 3% of the offline AUC-PR. This means that the mannequin’s predictive energy in figuring out at-risk accounts is similar to what we noticed offline.

When it comes to churn threat segmentation, our mannequin’s precision, recall, and proportion of the inhabitants captured throughout the excessive and medium threat churn classes have been constantly inside 2–3% of our offline analysis. This means that the segmentation of account threat based mostly on churn probability have been in keeping with our offline analysis and gross sales expectations.

Efficacy of Churn Prediction in Advertiser Churn Discount

We noticed a 24% (statistically vital) discount within the churn price of excessive tier pods*** in our experiment remedy group in comparison with the management. This means that accounts whose churn dangers have been uncovered to AMs have been much less more likely to churn than people who weren’t.

*** In excessive tier pods, AMs handle about 50–70 accounts on common.

On this weblog put up, we illustrated the event and implementation of an ML-based resolution for proactive churn prevention at Pinterest. We’re additionally actively investigating sequential mannequin architectures corresponding to Lengthy short-term reminiscence (LSTM) and Transformers, which can higher seize the utilization behaviors of advertisers and decrease the necessity for guide characteristic engineering corresponding to week-over-week or month-over-month characteristic aggregation utilized in our present mannequin.

Advertiser Development Modeling Staff

  • Engineering: Erika Solar, Ogheneovo Dibie, Keshava Subramanya, Mao Ye
  • Product: Shailini Pandya
  • Product Analytics/Knowledge Science: Alex Simons

Gross sales Staff

  • Product: Wesley Kwiecien, Grace Yun
  • Gross sales Managers: Abby (Fromm) Lubarsky

Salesforce Staff

  • Engineering: Gayathri Varadarangan (She Her), Murthy Tumuluri, Phani Chimata, Gabriela Mihaila, Richard Wu

Optimization Workbench Staff

  • Engineering: Phil Value, Jordan Boaz, Lucilla Chalmer
  • Product: Dan Marantz

[1] When and Why Tree-Based Models (Often) Outperform Neural Networks | by Andre Ye | Towards Data Science

To study extra about engineering at Pinterest, try the remainder of our Engineering Weblog and go to our Pinterest Labs website. To discover life at Pinterest, go to our Careers web page.