Guides
  1. Best Practices

Cost Optimization

Best practices for controlling storage, transcoding, and AI costs in Nomad Media without sacrificing the features users actually need.

Cost Optimization

Nomad Media costs usually come from four places:

  1. storage of masters, proxies, and metadata
  2. transcoding for proxies and delivery renditions
  3. AI processing for transcription and enrichment
  4. retroactive catch-up runs when new rules are applied to a large back catalog

The most effective cost strategy is not to disable useful features. It is to enable the right features for the right parts of the catalog, in the right order.

Start With the Highest-Value Features

For AI specifically, do not treat every processor as a Day 1 requirement.

The most cost-effective rollout order is usually:

  1. Transcription for speech-first audio and video
  2. Image enrichment / image search
  3. Richer LLM-enhanced search
  4. Deep video visual search

This avoids paying for advanced features before users have proven they need them.

Scope Processing by Folder, Year, and Media Type

The Rules Engine is the main tool for cost control.

Use it to limit expensive processing to content that benefits from it most.

Common examples:

  • enable transcription only for content/podcasts/, content/interviews/, or published video folders
  • exclude music-heavy folders from speech transcription
  • enable deep video search only for curated or public-facing libraries
  • roll AI out by year so recent content is enriched first
  • exclude work-in-progress folders that do not need user-facing discovery yet

The more intentional the scope, the better the cost-to-value ratio.

Use a Test Bed Before Broad Rollout

Before enabling richer AI features across the whole library, create a controlled test set.

This helps answer the real cost question: which AI tier materially changes user outcomes?

A good test bed includes:

  • representative audio files
  • representative video files
  • representative image files
  • clear and difficult examples
  • different languages where applicable

Run the same benchmark searches across each AI tier and compare the difference in results before scaling up. See Phased AI Rollout.

Watch for Catch-Up Spikes

One of the easiest ways to surprise yourself on cost is enabling a new processor on a large historical library.

Because Nomad can add missing outputs retroactively, a newly enabled processor can trigger a significant backfill wave.

This is often the right decision, but plan for it.

Good ways to reduce the spike

  • roll out by folder instead of whole-catalog
  • roll out by year, department, or content priority
  • process recent or most-used assets first
  • validate value in a test environment before enabling in production

See Reprocessing Assets for how additive reprocessing works.

Align Proxy Strategy With AI Strategy

Some AI costs are indirectly driven by proxy settings and dependent workflows.

  • audio transcription depends on MP3 proxy generation
  • video delivery may create extra renditions that increase transcoding cost
  • poorly chosen proxy rules can generate outputs in folders that do not need them

Review Proxy Generation Overview and Transcoding alongside AI rollout planning.

Archive Cold Content Deliberately

If large sections of the library are rarely accessed, storage lifecycle policies matter more than shaving a small amount off processing spend.

Typical pattern:

  • keep masters in cheaper archival tiers when appropriate
  • keep proxies and metadata hot so the user experience remains responsive
  • enrich only the content that needs searchability and active user access

See Media Archiving & Storage Lifecycle.

Practical Cost Checklist

  • Start with transcription before deeper AI features
  • Scope expensive processors by folder, type, year, or priority
  • Exclude low-value or unsuitable content such as music-heavy speech workflows
  • Build a representative AI test bed before full rollout
  • Plan for backfill spikes when enabling new processors later
  • Review proxy and transcoding rules alongside AI rules
  • Archive cold masters while preserving hot proxies and metadata

Related Pages