MACT INTELLIGENCE™
Service
- Scientifically analyze results and replicate them.
- MACT INTELLIGENCE™ is an integrated intelligence system where AI learns "why results happen" and guides "next steps." It combines three capabilities: customer behavior learning via GNN (Graph Neural Network), ROI expectation maximization algorithms, and natural language strategy proposals through RAG (Retrieval-Augmented Generation). By merging these three, it achieves "autonomous marketing optimization" that transcends the realm of data analysis.
Sensitivity Matrix Algorithm
Convert customers, channels, and behavioral events into a directed graph and calculate sensitivity α_i. Using the proprietary formula w_ij = α_i × F_i × exp(-β × T_i), we integrate and evaluate time elapsed, contact frequency, and contribution. This quantitatively identifies “what factors drove the results.”
ROI Expectation Maximization Engine
LTV prediction and sensitivity matrices are linked to calculate expected ROI in real time using the structure: ROI_i = (Predicted LTV Increment × Realization Probability) ÷ Campaign Cost. AI guides maximizing results based on cost efficiency.
RAG Generative AI (Proposal Text Engine)
AI searches and references a database of past success stories to automatically generate optimal strategies for each customer segment in natural language. By outputting proposal text, generative AI provides consistent support from strategy to execution.
Continuous Optimization via Time-Series Learning
The AI relearns implementation results and self-updates model accuracy. It monitors RMSE (error threshold) and performs automatic retraining only when accuracy declines. As an “AI that evolves its performance,” it maintains constant optimization.
LTV Prediction × Behavioral Data Integration
Integrate heterogeneous data such as advertising, CRM, and purchasing at the customer level, then analyze the causal structure of “customer × channel × time” using GNN models. Predict future LTV (Lifetime Value).”Learning customer behavior as a temporal directed graph,” “quantitatively estimating causal influence (α_i),” and “learning the behavioral pathways leading to campaign results”—these capabilities are beyond the scope of conventional LTV models. They form the core of MACT INTELLIGENCE™’s uniqueness and patented technology.
Visualization and Proposal Dashboard
The “RAG Generation Results Screen” integrates all analysis results into a single view. ROI, sensitivity, and actionable recommendations are instantly visible. MACT’s unique interface makes decision-making intuitively visual. *This interface is customized to the optimal form based on the client’s BI environment, metric design, and decision-making layer.
Method
- MACT's Unique Theory: The Science of Reproducibility
- The design philosophy behind MACT INTELLIGENCE™ is not merely data analysis or AI automation, but "intelligence that makes results reproducible." It learns past success factors as causal structures and probabilistically reproduces future actions. This "mathematical modeling of reproducibility" is the core value MACT delivers. Furthermore, the uniqueness of MACT INTELLIGENCE™ lies in its "cyclical AI structure": LTV prediction → sensitivity matrix → ROI optimization → RAG-generating AI → output → relearning.
Step
- step
- 1
Data Integration and Structuring
Integrate advertising, CRM, and purchasing data to convert customer behavior sequences into time-series graphs. The GNN model learns the causal relationships from "customer → channel → outcome".
- step
- 2
LTV Forecasting and ROI Calculation
AI predicts each customer's future lifetime value (LTV) and calculates expected ROI. By selecting the strategy with the highest ROI, it automates efficient strategic decision-making.
- step
- 3
Generation of the sensitivity matrix
Calculate the contribution α_i for each channel and initiative, and have the AI identify which initiatives are most effective in driving results. Extract the "replicable factors" of the strategy.
- step
- 4
RAG-Generated AI Strategy Proposals
AI searches past cases and proposes optimal strategies in natural language. Proposals (such as emails and campaign copy) include evidence-based rationale backed by performance data.
- step
- 5
Output, Verification, and Re-training
After generating RAG text, the AI retrains on the results. Accuracy is managed using an RMSE threshold, ensuring the model consistently maintains a high-precision state.
