# ModaML at a glance

> Turn business data into predictive models without writing code. Train on historical data, review the pipeline and use a saved model to predict outcomes for new rows.

## Who it is for

Built for customer success and revenue operations at software companies with historical account and sales outcomes. Browse examples by team: Operations, Customer Success, Sales, Finance, Marketing and Research. Operations includes supplier and inventory examples for logistics teams.

## Input → prediction → decision

Customer Success: account usage and past software subscription renewals → non-renewal risk → a customer check-in; Sales: trial engagement and completed software sales → conversion likelihood → a sales follow-up. Operations: past purchase orders and delivery outcomes → late-delivery risk → a procurement review of suppliers, volumes and fulfillment. Predictions inform the review; people choose the action. These are illustrative classification workflows, not customer results.

## A workflow you can repeat

A dashboard or spreadsheet can help you explore what happened. ModaML adds a saved prediction model, a pipeline you can inspect and adjust, and repeatable runs on new data. Compared with a one-off conversation about your data, a Service keeps the selected model version, data setup and run history together.

## Try the product story

Explore the public interactive playground with illustrative data and a read-only sample pipeline. It requires JavaScript but no login; it does not upload data or train models. Training and predictions on your own data require a workspace account; the current signup form asks for an invite code. No ModaML customer performance results are claimed here.

## Know the current limits

Available in the workspace: classification and regression on workspace datasets, pipeline review and editing, saved model versions and manual Services with workspace results. Scheduling and read-only connector previews depend on deployment and account setup. External imports, automatic source synchronization and external prediction writeback are unavailable; time-series forecasting and clustering are unsupported.

## From historical examples to the next decision

Start with one or more owned CSV exports or workspace datasets. For incomplete or scattered records, use Clean or combine multiple exports to map columns, normalize values, explicitly aggregate and join or append sources, and review full-data counts before saving a separate derived dataset. Confirm what one row represents and which existing answer the model should learn.

Use guided setup or build a pipeline manually. Review the proposed setup, train, and inspect held-out results and a simple baseline before using a saved model on new rows. Prepared models retain their approved recipe: map fresh exports to its source requirements and review the prepared prediction dataset. The answer column can be omitted.

A Service saves a repeatable setup: workspace CSV data, an exact saved model version and a workspace results destination. Add the next prediction dataset, run the Service and review its results and history. Quick Start also supports exploring a one-off model. Separate Service source bindings are scored independently; combine records through the reviewed preparation workflow first.

## Examples you can build with your own data

Retention, lead scoring, repeat purchases, inventory and collections have Service starting points. Supplier Management uses Custom service with a delivery-risk goal. These examples are not pre-trained business models or evidence of customer outcomes. Prepare historical labels and inputs in your CSV; review predictions before acting.

- Software / Customer success: active seats, recent usage and completed renewals → non-renewal risk → decide which accounts need a check-in.
- Software / Revenue operations: trial activity, demo requests and completed sales → conversion likelihood → prioritize follow-up after reviewing fit.
- Logistics / Operations / Supplier Management: historical order snapshots with quantities, quoted lead times and known late-delivery outcomes → late-delivery risk for current orders → compare suppliers and follow up at-risk deliveries. A planner checks price, capacity, minimum quantities and required dates before choosing suppliers and allocating volume. ModaML does not optimize order quantities or place orders.
- Marketing: order history and known repeat purchases → predicted repeat-purchase outcomes → review a potential campaign audience and its permissions.
- Inventory: historical inventory snapshots and known stockouts → predicted stockout outcomes → review items for replenishment. This is classification, not a demand forecast.
- Finance: historical invoice details and known payment outcomes → predicted late-payment outcomes → review which invoices need attention.

## See and adjust how the model is built

In the authenticated workspace, open the generated pipeline to inspect its steps. Supported controls let you adjust preprocessing, selected columns and model settings, then validate and save the changes before training again.

The editor works within ModaML’s supported classification and regression steps; it is not an arbitrary code editor. Training creates a saved model version. Editing a pipeline does not silently change a model already selected by a Service.

The public playground uses the same pipeline canvas to show a read-only sample. Its data and settings interactions illustrate the workflow; they do not establish model quality or execute a training run.

## What is available, conditional or planned

CSV upload, reviewed deterministic preparation of owned CSV/workspace data, classification and regression training, saved models, new-row prediction and manual Service runs are implemented. Preparation supports explicit normalization, schema-aligned append, exact-key joins and bounded aggregation. Services currently read workspace datasets and save results in the workspace.

Scheduling is optional and must be enabled by a deployment administrator. Scheduled runs use the configured workspace dataset; they do not automatically refresh it from another system.

Salesforce, GitHub, PostgreSQL, Snowflake and Google Drive offer read-only preview capabilities when configured and available to the account. These previews do not import a dataset, synchronize data or send predictions back to a provider. Upload authorized CSV exports to prepare, train or predict.

Other provider logos are planned catalog entries, not working connections or delivery promises. OpenAI-assisted setup and AI help depend on server configuration. Guided setup has a heuristic fallback; AI help is explanatory and does not run your workflow.

## Check the evidence before relying on a prediction

Useful models need representative historical examples with known outcomes. Include only inputs that would have been available when the decision was made. A future outcome accidentally included as an input can make evaluation misleading.

ModaML reports evaluation metrics; a score is not a guarantee of future business results. Review data quality, errors and whether the evaluation fits your decision. Current training uses random splits, not a dedicated chronological forecasting evaluation. Time-series forecasting and clustering are not supported.

The public examples and playground are illustrative. Published market research on the homepage describes its own study populations, not ModaML customer results. No customer uplift, accuracy or revenue claim is made here.

## Explore ModaML

- [Interactive product playground](https://modaml.com/overview#product-lab): Public, illustrative demo; JavaScript required, no login, no live training.
- [Request early access](https://modaml.com/waitlist): Read about access, then choose whether to load the hosted interest form. Optional booking follows the form.
- [Sign in to your workspace](https://modaml.com/login): An account is required to train and predict with your own data.

- [ModaML website](https://modaml.com/): Explore the product and its interactive examples.