Retail & CPG · Databricks Lakebase

Manuka TwinOS: A Digital Retail and CPG Twin on Databricks Lakebase

Retail and CPG leaders do not have a data problem. They have a coordination problem.

Inventory data lives in one system, supplier updates in another, promotions in a third, and the business is forced to make decisions from lagging snapshots instead of live operational truth. By the time a regional stock imbalance or a slipping supplier shows up in a weekly report, the window to act on it has usually closed. Manuka TwinOS changes that. It creates a living digital twin of the retail network on Databricks Lakebase, connecting operational state, analytical intelligence, and AI-driven action in one governed environment.

A twin is not a dashboard with a faster refresh. It is a continuously updated model of how the business actually runs: what is in stock, what is moving, what is late, and what that means for revenue and margin right now. That model is what lets teams stop reacting to yesterday and start steering today.

Inventory, supply, promotions, and demand stop living in separate systems and resolve into one continuously updated TwinOS digital twin. — Inventory, supply, promotions, and demand stop living in separate systems and resolve into one continuously updated twin.

Why this matters

Retail performance is won or lost in the gap between signal and action. A delayed shipment, an underperforming promotion, or a regional stock imbalance can erode revenue and margin long before most teams see it in reporting. The cost is rarely one large failure. It is a steady leak of markdowns, missed sales, expedited freight, and working capital tied up in the wrong places. TwinOS closes that gap by giving operators, merchandisers, and commercial teams a real-time system of understanding rather than a rear-view dashboard.

What TwinOS is

TwinOS is a Databricks-native digital twin that mirrors the moving parts of a modern retail and CPG ecosystem: stores, fulfillment nodes, suppliers, SKUs, promotions, inventory positions, orders, and demand signals. Lakebase acts as the low-latency operational database for the twin, holding the current state of the network so applications and agents can read and write it in milliseconds. Delta Lake and the broader Databricks platform provide the analytical and AI foundation around it, and Unity Catalog governs both as a single estate.

That architecture matters because it removes the usual seam between the systems that record what is happening and the systems that reason about it. The current state of the business is directly connected to forecasting, BI, machine learning, and agent workflows, instead of being isolated inside transaction systems and copied out on a nightly schedule.

This digital twin brings together demand forecasting, supply risk modeling, supply network and distribution planning, and Sales & Operations Planning into a unified decision intelligence platform. It showcases how Databricks-powered AI modeling can help retailers and CPG organizations anticipate disruptions, optimize inventory, improve service levels, and drive more resilient supply chain decisions.

How it works under the hood

Lakebase is a managed, serverless Postgres database that runs directly on the Databricks platform, with its data held in the same open cloud storage as the lakehouse. That design is what makes the twin practical to operate:

Two-way sync, no fragile pipelines. Lakehouse tables sync into Lakebase so apps can read them at low latency, and operational data flows back into the lakehouse for analytics and machine learning. Teams do not have to build and babysit a separate integration layer to keep the two in step.
One governance plane. Lakebase registers in Unity Catalog, so the access controls, auditing, and lineage that already apply to analytical data apply to the operational twin as well. There is no second security model to maintain.
Elastic and economical. Compute and storage scale independently, and idle compute scales to zero. The twin can absorb peak-season load and cost very little when the network is quiet.

The result is a single foundation for transactional apps, analytics, and AI agents, rather than three stacks stitched together at the edges.

Manuka TwinOS reference architecture on Databricks: sources and Lakeflow ingestion into the lakehouse, a Bronze to Gold digital twin, and BI, apps and agents acting on it. — The TwinOS reference architecture: sources and Lakeflow ingestion, a Bronze to Gold digital twin on the Data Intelligence Platform, and BI, apps and agents acting on it.

A short walkthrough of TwinOS, a living retail decision system on Databricks.

See TwinOS on your data →

What makes it different

This is not just another visibility layer. TwinOS turns operational data into a decision surface where teams can ask questions in natural language, investigate root causes, and test actions before they execute them. With Lakebase branching, teams can spin up an isolated, full-fidelity copy of the twin in seconds and simulate changes to lead times, supplier reliability, replenishment logic, or promotional plans, then discard the branch or carry the decision forward. Because branches use copy-on-write storage, that experiment costs almost nothing and never touches production.

TwinOS enables teams to:

Detect stockout and fulfillment risk earlier using a live operational twin.
Explore root causes across demand, supply, logistics, and commercial actions in one view.
Run scenario simulations on a branch without disrupting production workflows.
Support conversational decision-making through Genie-style interfaces and AI agents.
Move from insight to action on a single Databricks-native architecture.

Why Lakebase is the unlock

Lakebase gives TwinOS something most digital twins lack: a transactional backbone built for real-time apps and AI agents. Instead of stitching together a separate operational database, an analytics stack, and an AI app layer, TwinOS uses Lakebase, Unity Catalog governance, and the lakehouse as a unified foundation. Agents can hold their state and memory in the same governed database they reason over, so a monitoring agent that flags a fulfillment risk is working from the same source of truth as the BI report and the replenishment model.

The result is faster time to value, fewer integration bottlenecks, and a cleaner path to enterprise-scale retail agents that can monitor, explain, and recommend action continuously.

Where it creates value

TwinOS is designed for the teams that sit closest to revenue, margin, and service outcomes.

Team	How TwinOS helps
Supply chain	Detects disruptions earlier, quantifies downstream risk, and prioritizes corrective actions.
Merchandising	Evaluates promotion and assortment impact against live operational conditions.
Commercial	Connects execution risk to sales and margin outcomes in a business-friendly interface.
Data & AI teams	Delivers one governed architecture for apps, analytics, and agent workflows.

What a first engagement looks like

Manuka is a pure-play Databricks partner, so TwinOS is delivered on the platform your data team already runs, not as a separate product to procure and integrate. A typical engagement starts narrow and proves value fast. We model the slice of the network where the pain is sharpest, connect the operational and analytical sources that feed it, and stand up the twin with a working decision surface on top. From there, the same foundation extends to more categories, more regions, and progressively more autonomous agents, without re-platforming.

The bigger idea

The next generation of retail systems will not be defined by static dashboards or isolated AI copilots. They will be defined by living operational models that can understand the network, reason over current conditions, and help the business act faster. Manuka TwinOS is that model: a Databricks-native decision OS for retail and CPG, built on Lakebase.

Want to see TwinOS running on your retail data? Let’s talk.

Manuka AI | We build AI on Databricks