AddMaple Multi-Type Adaptive Discovery Framework
Find the relationships you didn't know to look for.
AddMaple turns mixed research data into a space you can explore.
Instead of deciding every statistical method in advance, AddMaple understands the kinds of variables in your study, applies methods appropriate to the question, and lets you move rapidly between three forms of discovery:
What is related? What explains the outcome? What groups exist in the data?
The result is analysis that is broader than a hand-built plan, more defensible than generic AI analysis, and fast enough to follow the research wherever it leads.
Built on established statistical and machine-learning methods. The adaptation happens in how they are selected, combined and explored.
The payoff
One study. Hundreds of questions you could have asked.
- Overall satisfactionLikert
- Likelihood to recommend0–10 scale
- Purchase channelCategorical
- TenureContinuous
- Job levelOrdered category
- Benefits usedMulti-response
Traditional analysis starts by deciding which analyses to run.
Multi-Type Adaptive Discovery starts with the question.
Your study
Mixed measurements, one file
What is related?
Association discovery
What explains this?
Driver discovery
Who groups together?
Segment discovery
Methods adapt to the structure of the data
A new way to explore research data
Multi-Type Adaptive Discovery
Most analytical workflows make the researcher choose the technique first.
Correlation. Crosstab. Regression. Cluster analysis.
The researcher rarely starts there. They start with a question about the study.
- What seems to be driving satisfaction?
- What distinguishes the people who churn?
- What else in this study is connected to this result?
- Are there meaningful groups hidden in the sample?
Multi-Type Adaptive Discovery reverses the workflow.
The researcher defines the question. AddMaple examines the measurement structure of the variables involved and routes the analysis through appropriate statistical methods.
Numeric variables can remain numeric. Categories can remain categories. Ordered responses retain their order. Multi-response questions can remain multi-response.
The researcher gets to think about the study rather than continually translating it into statistical machinery.
Three modes of discovery
01 — Association discovery
Find what matters before deciding what should matter.
Pick any variable in the study.
AddMaple can scan across the surrounding dataset and surface the strongest relationships — including relationships the original analysis plan never anticipated.
Rather than applying one universal statistic, the comparison adapts to the measurements involved. A numeric-to-numeric relationship is not treated like category-to-category. An ordered rating is not flattened into an unordered label. The statistical route changes with the pair.
The researcher sees the finding, not just the test.
A result can go further than “purchase channel is associated with satisfaction.”
Satisfaction is disproportionately lower among customers purchasing through assisted channels.
That gives the research somewhere to go next.
02 — Driver discovery
What still matters when everything is considered together?
Association discovery broadens the field. Driver discovery narrows it.
Choose an outcome and AddMaple evaluates the explanatory variables together, using modelling appropriate to the outcome and the structure of the predictors.
The result is not a list of correlations. It answers a harder question:
Which signals continue to carry information about this outcome when the other signals are present?
One answer is not always enough.
An important part of Multi-Type Adaptive Discovery is that the first plausible model does not have to become the final story. AddMaple can explore alternative formulations.
An AddMaple agent can investigate several defensible formulations in parallel, compare their findings, and bring the researcher the differences that matter.
AI expands the search. The statistical framework constrains it.
Outcome
Likelihood to recommend
| Candidate signal | Measurement | Role in the result |
|---|---|---|
| Support quality | Ordered rating | Strong positive signal |
| Wait time | Numeric | Negative relationship |
| Purchase channel | Categorical | Differing outcome profile |
| Benefits used | Multi-response | Benefit-specific effects |
Focused model
8 variables
Broad model
24 variables
Alternative framing
Ordered outcome
Compare the evidence
03 — Segment discovery
Discover groups without pretending every respondent is a row of continuous numbers.
Segmentation is often where mixed research data becomes most awkward.
A study may contain ratings, behaviours, demographics, ordered responses and selections from a list. Those variables describe respondents in different ways. Multi-Type Adaptive Discovery allows segmentation to begin with that reality.
AddMaple can explore different defensible ways of describing structure in the population, rather than committing immediately to a single clustering assumption.
The useful question becomes:
Which grouping produces a stable, interpretable description of the people in this study?
Segmentation becomes a research decision supported by evidence, rather than the output of whichever algorithm happened to be selected first.
Same respondents
Solution A
3 groups
Clear story
Solution B
4 groups
Strongest separation
Solution C
5 groups
Fragmented
Profile and compare
Defensible segments
The adaptive layer
AI can explore. Statistics set the boundaries.
This is what separates AddMaple from asking a language model about a spreadsheet.
A language model is good at proposing questions, exploring alternatives and interpreting results. It should not invent statistical evidence.
The Multi-Type Adaptive Discovery Framework separates those responsibilities.
The agent does not manufacture the analysis.
It can:
- Investigate hypotheses
- Launch alternative driver formulations
- Compare segmentation solutions
- Follow unexpected associations
- Ask for deeper cuts
- Explain what changed
The evidence comes from the analytical framework. That distinction is what makes agentic research useful rather than arbitrary.
Researcher or agent
Asks, explores, reframes
Multi-Type Adaptive Discovery Framework
- Understand the measurement
- Select valid routes
- Execute the analysis
- Evaluate alternatives
Statistical evidence
Explanation, then the next question
Interactive by design
Discovery works differently when analysis is interactive.
A traditional advanced-analysis workflow carries friction at every step. Each iteration waits.
Multi-Type Adaptive Discovery is designed around a shorter loop. When the engine can answer quickly, researchers behave differently. They test the unexpected result. They change the target. They remove a questionable variable. They compare another segmentation. They ask an agent to explore three alternatives at once.
Speed does not merely save time. It changes how much of the research space can be explored.
A traditional pass
- Question
- Request
- Setup
- Run
- Export
- Interpret
- Next question
Multi-Type Adaptive Discovery
01
Ask
02
Run
03
See
04
Refine
Refine returns to the next question.
The name
What we mean by Multi-Type Adaptive Discovery
Multi-Type Adaptive Discovery is an analytical approach in which the statistical route adapts to both the research question and the measurement structure of the data, allowing multiple forms of discovery to operate across the same heterogeneous study.
Multi-Type
The study can contain different forms of measurement without requiring everything to be transformed into one representation first.
Adaptive
The analytical route is selected according to the problem rather than fixed in advance.
Discovery
The system is designed for exploration: testing relationships, explanatory structure and population structure beyond a predefined analysis plan.
Framework
The same methodological layer governs human exploration, the interactive product, and agent-driven analysis.
AddMaple developed the Multi-Type Adaptive Discovery Framework to make this approach practical inside a modern research workflow.
Your study contains more questions than your analysis plan.
You already paid to collect the data.
Multi-Type Adaptive Discovery helps you find more of what is inside it.
Find the unexpected relationship.
Work out which signals survive multivariate analysis.
Compare competing descriptions of your audience.
Follow the next question while it is still useful.
Do it inside one analytical framework that can be inspected, explained and defended.
Use AddMaple Multi-Type Adaptive Discovery on your next study.