Put a model behind every member, marketing and channel decision
FMCG data is scattered across membership systems, e-commerce platforms, distributor inventory records and stores. With SAM at the core, we turn the key judgments in member management, campaign spend and channel management into models and scorecards you can review — and put them to work through VDC, DH S1 and the Enterprise Agent Platform.
Members · marketing · channels · supply6 core model typesOn-premisesTry online for up to 10,000 samples
The challenge
Three old problems in FMCG marketing
01
Lots of data, no common definitions
Membership, e-commerce, store POS and distributor records all define things differently — an “active member” means something different in every report, so modeling stalls.
02
Spend that cannot be attributed
Coupons and subsidies often go to people who would have bought anyway. Without control groups and incrementality, spend keeps growing with no clear view of the lift it buys.
03
Channel risk you cannot see
Distributor credit rests on experience and diversion or false claims surface in audits after the fact — usually once payments are late or the price structure has already been undermined.
Use cases
Six core models across members, marketing and channels
Every model runs the same SAM fourteen-step pipeline: agree the data contract, then variable analysis, binning, fitting and independent validation, and finally release signed by an accountable person.
01 · Members
Member churn scorecard
Business question
Which members will stop buying in the next 90 days?
Method
A churn scorecard built on purchase frequency, recency, category breadth and engagement (WOE binning + logistic regression).
Deliverable
Churn scores and member tiers, with score thresholds for win-back campaigns.
02 · Marketing
Response and uplift models
Business question
Who buys because of the offer — and who would have bought anyway?
Method
Response and uplift models that separate persuadable customers from those who would buy regardless.
Deliverable
Targeted lists by segment with expected incremental sales — less wasted subsidy.
03 · Channels
Distributor and store credit scorecards
Business question
Which distributors get credit terms, and how much?
Method
A credit scorecard built on purchasing, payment, inventory turnover and past delinquency, with limits and terms tiered by score.
Deliverable
Credit grades, recommended limits and payment-term rules, reviewed on a schedule.
04 · Marketing
Promotion and price-elasticity analysis
Business question
Which promotions actually drove incremental sales, and where can prices move?
Method
Control-group and difference-in-differences analysis plus price-elasticity estimates, all computed by deterministic code.
Deliverable
A promotion return review and recommendations for the next plan.
05 · Supply
New-product and regional demand forecasting
Business question
How much should each region and store stock and replenish?
Method
Hierarchical time-series and machine-learning forecasts, deployed and refreshed on DH S1.
Deliverable
Store × SKU × week forecasts and replenishment suggestions.
06 · Channels
Diversion and rebate-claim anomalies
Business question
Which orders and claims show diversion or false redemption?
Method
Our fraud-ring and link-analysis methods, carried over from anti-fraud, combined with rules and anomaly detection.
Deliverable
Risk scores and a review list for channel and audit teams.
Why scorecards
Models business teams can read and use
A scorecard turns a complex model into rules — how many points each feature earns and what to do above a cut-off — so marketing, sales and channel teams can understand and act on it. Every bin and weight traces back to data and code, and once live, SAM-07 monitors distributions and performance.
Explainable scores: every feature’s points are visible
Adjustable cut-offs: move the strategy line with budget and goals
Monitored in production: drift opens a ticket and people decide what to do
The more precise the marketing, the firmer the boundaries
Models make marketing more precise, but personal information and published marketing content must stay within the law. Our approach builds this in, from deployment to process design.
On-premises: member and transaction data stay in your own environment, and modeling runs in a controlled sandbox.
Consent and opt-out: personal information used for marketing must be lawfully authorized; under Article 24 of China’s Personal Information Protection Law, marketing through automated decision-making must also offer an option not based on personal characteristics, or an easy way to refuse.
Fair pricing: models are not used to apply unreasonable differential treatment in prices or other transaction terms.
Content review: LLM-generated marketing copy is reviewed by people before publication, in line with the Advertising Law.
Full audit trail: every run produces an immutable manifest, so data, code, parameters and sign-offs can be replayed in full.
Try online
Every SAM feature, online, for up to 10,000 samples
Upload up to 10,000 records and run the whole fourteen-step pipeline — data contract to packaged scorecard, human gates included — to see for yourself whether SAM fits your business.
Up to 10,000 samplesEvery feature enabledThe full fourteen-step pipeline
Start with one business question. Let SAM deliver a model you can sign off.
Whether you are in banking, FMCG and marketing, pharmaceuticals or another industry, try SAM online first — or talk to our modeling experts about your use case.