What this looks like in practice

Take a mid-sized software company or fintech team.

Their operational data lives across systems they use every day:

• Product logs in S3 or Databricks

• Customer interactions in Salesforce

• Support tickets in Zendesk

• Internal discussions in Slack

• Analytics dashboards in Snowflake or Tableau

To stay compliant and resilient, all of this is continuously backed up using systems like Veeam, Rubrik, or Duplicati. These backups are immutable, retained for years, and treated as a safety net.

Separately, the data team is trying to answer questions like:

• What behaviors predict churn?

• Which features actually drive retention?

• What patterns led to our best-performing cohorts?

To do this, they don’t go to backups. They build new pipelines.

They export subsets of data into Snowflake. They clean it, reshape it, and often realize it’s incomplete. Historical context is missing, schemas have changed, and important signals were never captured in the analytics layer.

So they compensate.

They buy external datasets.
They subscribe to enrichment APIs.
They pay for analytics tools that promise better visibility.

Now the company is paying twice:

• Once to store its full operational history for compliance

• Again to approximate that same history through external data and tooling

This is not a tooling problem. It is a structural one.

Why backups never enter the equation

Backups contain the most complete version of a company’s history.

They include not just the cleaned, structured data that makes it into dashboards, but the raw operational reality:

• Every version of a customer record

• Every iteration of a product event schema

• Every internal document, message, and decision artifact

• Every change over time

In theory, this is exactly what AI models need: longitudinal, high-fidelity data that reflects how the business actually operates.

In practice, backups are invisible.

They are optimized for recovery, not access.
They are stored in formats that don’t map cleanly to analytics workflows.
They require tickets, approvals, and manual extraction to even touch.

So teams ignore them.

Instead of using the data they already own, they rebuild partial versions of it elsewhere and fill in the gaps with external sources.

That is the second payment.

The hidden cost is not storage. It is lost context.

When a team buys external data or rebuilds pipelines, they are not just spending money. They are giving up something more important: continuity.

External datasets can tell you what happened in the market.
They cannot tell you how your organization responded to it.

Analytics pipelines can show current metrics.
They rarely capture the full path that led to those outcomes.

This is why so many AI initiatives feel shallow.

Models are trained on clean snapshots instead of messy reality.
They learn patterns that look good in isolation but miss the decisions, constraints, and behaviors that actually drove results.

Meanwhile, the richest dataset the company owns—its own history—is sitting untouched in backups.

As described in the Duplicati plan, firms often retain 10–20+ years of operational data for compliance, yet still spend heavily on external datasets because that internal data is not accessible for AI workflows .

That is the core inefficiency.

Collapsing the double payment

The shift is not about replacing data warehouses or stopping external data purchases entirely.

It is about changing what backups are allowed to do.

Instead of treating backups as a sealed archive, Duplicati turns them into a usable data layer.

The process is straightforward, but the impact is structural:

1. Extract and structure backup data
   Data from backup systems is continuously exported into formats like Parquet or Delta Lake on S3-compatible storage, making it usable alongside existing analytics stacks.

2. Make it queryable and version-aware
   Instead of a single snapshot, teams can access how data evolved over time. This is critical for understanding causality, not just correlation.

3. Connect it to existing workflows
   The same teams already using Snowflake, Databricks, or MLflow can now pull from backup-derived datasets without building fragile, one-off pipelines.

4. Enable model training on internal history
   AI models can train on actual company behavior: product usage patterns, support interactions, operational decisions.

Now the equation changes.

The company is no longer paying to store data it cannot use.
It is converting that storage into a training asset.

External data becomes a supplement, not a crutch.

A concrete example

Imagine a SaaS company trying to improve retention.

Today, they might:

• Pull recent product usage data into Snowflake

• Combine it with CRM data

• Buy third-party enrichment data

• Train a churn model

The model works, but only sees a narrow slice of reality.

With backup data activated:

• They can access every historical version of user behavior

• See how feature usage evolved across product iterations

• Incorporate support conversations and internal notes

• Reconstruct what actually happened before churn events

The model is no longer guessing from partial data.
It is learning from the full operational history of the company.

That is not just better accuracy. It is a different category of insight.

The real economic shift

Most discussions about data infrastructure focus on cost reduction.

This is not primarily about reducing storage spend.

It is about eliminating redundant spend and increasing the return on data that already exists.

As outlined in the Duplicati thesis, firms often allocate hundreds of thousands of dollars across backup storage, data engineering, and external datasets. Even replacing a fraction of external data with internal history can unlock significant savings while improving model quality .

More importantly, it changes how teams think about their data.

Backups stop being a liability on the balance sheet.
They become an asset that compounds over time.

The broader implication

Every company is trying to become “AI-driven,” but most are building on incomplete foundations.

They invest in models, tools, and datasets without realizing that the most valuable dataset they have already exists.

It is just locked away in a system designed for a different era.

The companies that win will not be the ones that buy the most data.
They will be the ones that can actually use their own.

Duplicati exists to collapse that gap.

To turn compliance storage into training data.
To connect history with decision-making.
To ensure that companies stop paying twice for something they already own.