- SaaS / Accounting softwareIndustry
- Online accounting platformClient
- Machine LearningCategory
- Phased rolloutTime to impact
The challenge
- By the time churn shows up in the numbers, it is usually too late to act.
- The team needed an early signal for who is at risk, and just as importantly, what to actually do about each account.
Two models: one predicts at-risk customers, the other recommends the next action to take.
A predictive system built on two models. One scores churn risk and forecasts key metrics to inform decisions; the other recommends the next action to take on each at-risk account. So the team knows not just who is leaving, but what to do about it.
What we built
- 01One model scores churn risk and forecasts the metrics that drive decisions.
- 02A second model recommends the next action to take on each account.
- 03At-risk accounts surface early, each with a reason.
- 04The team sees who is leaving and what to do about it.
How it works
What it can do
Decision model
Scores churn risk and forecasts metrics to inform decisions.
Action model
Recommends the next action for each at-risk account.
Risk scoring
Ranks customers by likelihood to churn.
At-risk alerts
Flags accounts that need attention now.
Who stays in the loop
The call the system deliberately does not take, and what the person is given to take it with.
A predictive system built on two models
The retention or customer-success owner
Decideswhether to act on an account, and whether the recommended action is the right one
Seesthe risk score, the reason it moved, and the next action the second model proposed
The payoff
What is different once it is running. No invented numbers: these are the changes the work was built to make.
You see who is about to leave and what to do about it, while there is still time to act.
Not just who is leaving, but what to do about it.
At-risk accounts flagged early, with a reason.
Retention effort goes where it actually matters.
Built with
The layers this runs on, from what comes in to what it plugs into.
- Model layer
- ML models
- Forecasting
- Runs on
- Agentic workflow
Common questions
The same answers as above, written out.
A predictive system built on two models. One scores churn risk and forecasts key metrics to inform decisions; the other recommends the next action to take on each at-risk account. So the team knows not just who is leaving, but what to do about it.
By the time churn shows up in the numbers, it is usually too late to act. The team needed an early signal for who is at risk, and just as importantly, what to actually do about each account.
Feed in account and usage signals. One model scores risk and forecasts; another picks the next action. At-risk accounts surface with a recommended action.
The retention or customer-success owner decides whether to act on an account, and whether the recommended action is the right one. They see the risk score, the reason it moved, and the next action the second model proposed.
Model layer: ML models, Forecasting. Runs on: Agentic workflow.
Not just who is leaving, but what to do about it. At-risk accounts flagged early, with a reason. Retention effort goes where it actually matters.
Where this fits
The practice it belongs to, how an engagement like it runs, and what we have written about the subject.
Data AnnotationData Labeling: The Overlooked Bottleneck in AI and Machine Learning
Model architectures often get the spotlight, but real-world performance in AI depends heavily on data labeling quality. Learn why annotation workflows, human-in-the-loop systems, and synthetic data strategies are what make an ML model hold up in production.
Frederico VicenteSep 20259 min read
Business & OperationsOptimising your operations with AI: start with the process, not the model
A practical playbook for optimising operations with AI: find the one slow step, automate it, keep a person in the loop, and measure the before and after.
Miguel Vicente JrJun 202612 min read
Machine LearningChurn Prediction: From Logistic Regression to Foundation Models
Customer churn costs businesses billions annually. This technical deep-dive compares statistical methods, gradient boosting, and transformer models like TimesFM 2.5 and Chronos 2 for churn prediction - with benchmarks, architecture diagrams, and implementation insights.
Frederico VicenteDec 202512 min read
Other projects in this line.
Bring us the problem nobody has cracked yet.
We are a small team of senior specialists. We pick the right model and the right layer, and we build the least machinery that does the job. You get a call with an engineer, not a sales deck.