ML Cost Attribution by Feature: Make Spend Visible

Most ML platforms hide where the spend goes. The attribution layer that lets product owners see which features cost what, and the conversations it unlocks.

Why product owners need this

Most ML platforms emit one big monthly bill. Product owners cannot tell which feature drove the spend; finance cannot prioritise; engineers cannot optimise.

Attribution turns the bill into a per-feature line item. Now the product team sees that the recommendation widget cost $42k while the search box cost $8k.

The conversation changes. Instead of arguing about whether ML is 'too expensive,' the team argues about whether the recommendation widget is paying its keep.

How to attribute

Tag every inference request at the feature boundary. The tag is a small string: 'reco-v3', 'search-typeahead', 'classifier-spam'.

Roll up tags daily. Sum tokens, GPU seconds, and dollar costs per tag. The roll-up lives in a small table that any dashboard can read.

Joins to product analytics: feature usage by user. Now you have cost-per-user for each feature; the comparison is direct.

What the dashboard shows

Per-feature spend over time. Trends matter; week-over-week changes catch silent regressions.

Cost-per-user per feature. The most useful number for product decisions. A feature with $0.50 cost-per-user is healthy; one with $5 cost-per-user needs attention.

Top 10 most expensive features, ranked. The list usually surprises product; the surprise is the value of the dashboard.

The policy that closes the loop

Each feature has a cost owner. The owner sees the dashboard weekly. They are accountable for explaining cost spikes.

Feature launches require a cost projection. The projection is compared to actuals after 30 days; gaps drive the next iteration.

Quarterly rightsizing review: the top 5 most expensive features get an optimisation pass. Most yield 30-50% savings without quality loss.