Designing an Enterprise ML Platform That Drives Real Decisions

Enterprise supply chain platforms feel stuck

Forecasting platforms have been around for decades, with established players like SAP, Oracle, and o9 shaping the category. But the experience often lags—heavy UI, steep onboarding, and slow workflows that rely on customer success to bridge gaps.

Stemly’s opportunity was to build a modern product: a clear, fast interface that enterprise teams can learn quickly—without sacrificing the accuracy and depth that drive real savings.

Rapid prototyping to reduce risk

In complex products, the fastest way to learn is to prototype and test early. We built interactive prototypes across planning, forecasting, and optimisation workflows to validate assumptions before engineering investment.

Prototypes aligned product, data science, and commercial teams on what mattered—and helped sales teams tell a clearer story while gathering feedback from real enterprise users.

This approach let us iterate quickly, keep scope realistic, and design workflows that feel simple even when the underlying models are not.

Platform structure and shared language

Early on, teams had different mental models of how the platform should work—and even used the same words to mean different things. That’s a fast way to create confusing UI and inconsistent product decisions.

We introduced a term dictionary to define domain concepts and UI naming. For data-heavy products, clarity in language is part of the product experience.

We then shaped the platform structure by benchmarking best-in-class SaaS patterns and adapting them to Stemly’s needs. An organisation → workspace → module model gave us a scalable foundation.

A design system built for shipping

To scale quality and speed, we built a design system informed by Google's Material UI 2 and the Vuetify 2 UI framework, tuned for a data-dense enterprise product.

The system was developed alongside the brand refresh—first inside the product, then across the marketing site and supporting interfaces for a consistent, recognisable experience.

We focused on reusable, build-ready components mapped to the front-end library, reducing one-off UI and helping engineering ship faster without quality drift.

Designing buildable products

In data-heavy products, precision is non-negotiable. Naming, behaviour, states, and edge cases need to be explicit—otherwise engineering and users interpret the experience differently.

Over time, prototypes shifted from “ideal” to “shippable.” To get there, the team continuously learned across data science, forecasting, and visualisation—so product choices stayed grounded in how the system actually works.

Enterprise-ready capabilities

Stemly serves enterprise teams with large datasets and high upside from better forecasting accuracy. We prioritised a set of capabilities needed to operate confidently at scale.

Multi-tenancy was core: an organisation/workspace structure designed for data isolation, security, and scalable administration.

We then tackled Identity and Access Management (IAM)—users, groups, roles, and data permissions—simplifying a complex area without losing control or auditability.

Finally, we designed “Connectors” for ERP and data integrations via APIs, supporting large volumes and reliable ingestion for real-world enterprise environments.

Search makes data usable

In data-first products, search is not a feature—it's a workflow. Users need to find the right time series, hierarchy, or range quickly, then turn it into charts and tables without friction.

We designed a next-gen search experience that blends speed with control: type-to-search with guided properties across time series, frequency, date ranges, and hierarchy—so users can refine without getting lost.

A bold brand built for trust

As the product matured, Stemly needed a brand that matched the value it delivered—credible for enterprise, distinctive in a crowded category, and consistent across product and go-to-market.

We refined the story and visual identity to signal precision and momentum. The calming blue and energising lime palette supports a clear, modern feel, backed by a brand book and asset library for consistent use.

Valuable takeaways

Designing for technical experts still requires radical clarity—terminology, behaviour, and edge cases are part of the UX. You can’t fake it without learning how the system works.

Feature work needs ruthless focus. Aligning on trade-offs early prevents an endless backlog and keeps delivery tied to outcomes, not opinions.

Early-stage teams move faster with constraints. Use proven building blocks for non-core features and invest your energy where you actually differentiate.

Data visualisation looks simple but rarely is—great charts and tables require clear models, consistent logic, and careful interaction design.

Clickable prototypes can close the gap in enterprise sales, especially when procurement cycles are long and stakeholders need to feel confidence early.