How to Use Tines Actions and Formulas for Enrichment at Scale

November 2, 2026

Tines scales on two things: how well you use formulas to reshape data and how carefully you handle loops and concurrency. A story that processes 10 indicators fine will break at 400 — usually on rate limits or a loop that runs sequentially when it should fan out. Both are fixable without leaving the visual builder.

This is the part that separates a demo story from one that runs unattended every day.

What are formulas in Tines, exactly?

Formulas are expressions evaluated against event data and action outputs. a Tines formula expression pulls a field. Functions do the work: split, join, filter, map, sort, uniq, size, regex_replace.

Two formulas worth memorising:

URLS = EVENT.urls.map('url => url.trim()').uniq()
HITS = ENRICH.result.filter('r => r.score > 60').size()

The first deduplicates and trims a URL list. The second counts enrichment results above a score threshold. Both replace what used to be a code block and are readable a year later.

How do you loop over hundreds of indicators?

Use a Loop action instead of chaining 400 individual actions. The loop maps over an array and runs its child actions per item, collecting results.

The scaling detail that matters: set loop concurrency to somewhere between 5 and 20 items in flight. Higher than the vendor’s rate limit and requests start failing with 429s; sequential execution makes a 400-item loop take minutes.

A practical pattern for a 400-URL list against an API limited to 10 requests per second: concurrency 8, plus a retry action with backoff of 1s, 4s, 9s. In practice this finishes in under a minute.

What should be a code action and what shouldn’t?

Code actions support Python and JavaScript, and they’re seductive. A policy that keeps stories maintainable:

  • Formula for reshaping: picking fields, filtering, counting, string ops.
  • Code when you need a library (parsing a PCAP header, decoding a JWT) or genuine logic over a list.

A 200-line code action nobody can read is worse than six formulas someone can audit. Security stories get reviewed by people who didn’t write them, and readability is a real operational requirement, not a preference.

How do you handle failures without losing the batch?

A single failed API call inside a loop can abort the whole story. Three options, in order of preference:

  1. Error branch per action — route failures to a side channel, keep the loop going.
  2. Retry with backoff on the action itself for transient errors (429, 502).

  3. Emit partial results and mark the story “degraded” in a Slack message rather than failing silently.

Option 3 is the one teams forget. A story that returns 380 of 400 enrichments is useful; a story that returns nothing because four rows threw is not.

Does Tines bill per action?

It bills by workflow actions and the plan tier, which is why lean stories matter commercially as well as for speed. Ten small formulas cost less than one heavy code action doing the same thing, and they’re faster to read.

Consolidating repeated sequences into reusable actions or a shared story cuts cost and maintenance at once. Tines supports passing actions between stories, which is how you avoid building the same enrichment block four times.

When does a story need splitting?

Four signals:

  • More than about 25 actions in one story.
  • Two different teams editing the same story.
  • A loop over 1,000+ items, where a failure anywhere invalidates the run.
  • Any story that takes longer than a few minutes, where partial results would still be useful.

Split by stage: ingest and enrich in one story, decide and act in another, with a webhook between them. Each half becomes independently testable, and a failure in enrichment doesn’t take the response actions down with it.

Where visual builders hit their ceiling

Complex stateful workflows — anything that needs to wait for a human for days, or accumulate state across runs — fit awkwardly in a visual story, even a good one. If you find yourself building a queue inside a loop, you want a service, not a workflow.

Similarly, high-volume streaming detection (per-second traffic analysis) is the wrong job. Tines handles event-driven automation in the hundreds-to-thousands per day range comfortably. Millions per hour is a stream processor’s problem.

Quick Summary

  • Formulas reshape data — map, filter, uniq, size replace most small code blocks.
  • Loop concurrency 5–20 is the practical band; match it to the API’s rate limit.
  • Use code only for libraries or real logic, and keep stories readable for reviewers.
  • Always emit partial results — 380 of 400 enrichments beats an aborted run.
  • Split stories past ~25 actions so each stage is independently testable.
Tags: