Bookings & revenue
Forecast commercial outcomes, pipeline conversion and period-close performance. Align the prediction horizon with how your business plans.
MLAIA Prediction & Causal Analysis
Predict what’s next.
Understand what changes it.
Forecast bookings, revenue, churn and demand. Use causal analysis to evaluate which actions can change outcomes, with explicit assumptions and evidence.
A specialist practice of MLAIA Data ScienceLed by Dr. Yochai Edlitz · Ph.D., Weizmann Institute
Where we help
Depth where it matters.
Focused expertise, with the research and engineering support to move your project forward.
Churn & retention
Model customer risk and behavior to support retention priorities. Distinguish predictive risk from the measured effect of an intervention.
Demand & inventory
Forecast item-level demand, including slow and intermittent patterns. Connect predictions to inventory and replenishment decisions.
Warehouse predictions
Model relevant warehouse demand, volumes or workloads using your operational history and the planning horizon you need.
Lead scoring & sales
Prioritize opportunities using customer and historical interaction data. Evaluate ranking quality against the capacity of your commercial team.
Operational integration
Bring predictions into existing reporting and workflows, with backtesting, data-quality checks and model monitoring.
Try an interactive example
Put a prediction under the microscope.
Choose a synthetic customer, explore what-if scenarios, inspect SHAP explanations and adjust a decision threshold. See how individual predictions connect to model performance.
Open the prediction explorer →Synthetic demonstration data · No client information
The Ladder of Causation
Three levels.
Three different questions.
Climb from predicting an outcome to understanding an action—and reasoning about what might have been.
Each step requires additional evidence and assumptions.
01 / Association
Which customers are likely to churn?
Use customer history to estimate risk and prioritize attention.
Evidence needed
Try a prediction & what-if →Historical data and honest out-of-sample evaluation.
02 / Intervention
How much would a retention offer reduce churn?
Estimate the effect of making the offer, rather than comparing recipients with everyone else.
Evidence needed
Experiments or credible causal designs, with explicit assumptions.
03 / Counterfactuals
Would this customer have stayed without the offer?
Reason about the same customer who received an offer and stayed, under an alternative history.
Evidence needed
A causal model with additional structural assumptions. Individual alternative outcomes are not directly observable.
01 / Association
What revenue should we expect next quarter?
Forecast revenue from the patterns and conditions observed in historical data.
Evidence needed
Try a prediction & what-if →Historical data and honest out-of-sample evaluation.
02 / Intervention
How would launching a campaign change revenue?
Estimate incremental revenue relative to a defined no-campaign policy.
Evidence needed
Experiments or credible causal designs, with explicit assumptions.
03 / Counterfactuals
What would this market’s revenue have been without the campaign?
Given the campaign and revenue actually observed in this market, reason about the alternative outcome.
Evidence needed
A causal model with additional structural assumptions. Individual alternative outcomes are not directly observable.
01 / Association
Which items are likely to run out?
Predict demand and stockout risk under the operating conditions represented in the data.
Evidence needed
Try a prediction & what-if →Historical data and honest out-of-sample evaluation.
02 / Intervention
How would an earlier reorder policy change stockout risk?
Estimate the effect of changing the policy, accounting for demand and supply conditions.
Evidence needed
Experiments or credible causal designs, with explicit assumptions.
03 / Counterfactuals
Would ordering earlier have prevented last month’s shortage?
For an item that actually stocked out, reason about what would have happened under an earlier order.
Evidence needed
A causal model with additional structural assumptions. Individual alternative outcomes are not directly observable.
The step up is in the evidence. Prediction alone does not establish the effect of an action. Causal conclusions depend on study design, data and assumptions; counterfactual conclusions can require stronger assumptions still. We make those assumptions and limitations explicit.
Inspired by Judea Pearl’s Ladder of Causation, presented in The Book of Why by Judea Pearl and Dana Mackenzie. Read Pearl’s explanation ↗
Experiment design
Define the intervention, comparison, outcomes and analysis plan for an A/B test or other randomized study.
Causal inference
Assess whether observational data and a defensible study design can support the causal question. Examine confounding, assumptions and sensitivity.
Uplift modeling
Estimate how intervention effects vary across customer groups, to inform targeting where the evidence supports it.
From our project archive
Customer conversion modeling.
Explore MLAIA’s published account of using customer data and machine learning to support sales prioritization in the communications sector.
Read the project account →How we work
A clear path from question to evidence.
Frame the problem
Agree on the decision, available data, operating constraints and success criteria.
Establish a baseline
Inspect data quality, design the evaluation and test what current approaches can do.
Develop & evaluate
Build in bounded milestones. Review results, failure modes and limitations together.
Integrate & hand over
Plan deployment, monitoring, documentation and ownership around the agreed scope.
From our knowledge base
Explore the thinking behind the work.
Before we start
Good questions. Straight answers.
Is this the same as NextDemand?
No. MLAIA develops bespoke forecasting and predictive models across business use cases. NextDemand is a separate MLAIA product focused on spare-parts demand and inventory planning.
How do you evaluate whether a model is useful?
We agree on decision-relevant measures and compare against current practice. Forecasts need time-aware backtesting; churn and lead-scoring models need evaluation that reflects how teams will use them.
Can a predictive model tell us what caused an outcome?
Predictive accuracy or feature importance alone does not establish causation. Causal analysis starts with a defined intervention or counterfactual question, a suitable study design and explicit assumptions. Some questions cannot be reliably answered with the available evidence.
Can we start with the data we already have?
Yes. A feasibility review can begin with your existing CRM, ERP or historical exports. Data quality, coverage and the decision horizon determine what is realistic.
Prediction & Causal Analysis
Tell us what
you’re working on.
Share the problem, the data you have and what success would look like. We’ll discuss a practical next step.
Please don’t include confidential datasets, credentials or patient information.