Enrichments overview

What enrichments are, the sixteen types grouped by Transform, Guard, Watch and AI, the order they run in during a sync, and which plan each type needs.

7 min read

On this page

Enrichments are the processing steps BankSync runs on your data while a feed syncs, before anything is written to your sheet, database, or table. They fill in fields, keep junk rows out, split or pair records, and notify you when something matters — so the data that lands is already the data you wanted.

There are sixteen enrichment types. This page is the map: what each one does, the order they run in, and which plan carries them. Each type then has its own guide.

Where enrichments live

Open Enrichments in the left navigation for the workspace-wide grid, or open a feed and use Mapping → Enrichments to see and toggle the enrichments attached to that feed. Every enrichment is created from the same New Enrichment dialog and every one has an enable switch, a Dependencies tab for feed scope, and a Live Preview you can check before you sync.
The New Enrichment dialog: a search box above a category rail listing All, Transform data, Guard data, Watch and AI and learning with counts, beside a two-column grid of enrichment cards, each with its icon, name and one-line description, and a Don't see the enrichment you need? Request one link in the footer.
The New Enrichment dialog groups the sixteen types into four categories.

The four categories#

Transform data

Fill field values as records sync. Ruleset applies your When/Then rules; Lookup Table applies a big key-value mapping like vendor to GL code.

Guard data

Change which rows reach your destination. Filter excludes matching records, Split replaces one record with several, Transfer Matcher pairs internal transfers, Duplicate Screen flags look-alikes.

Watch

Notify without touching data. Alert, Feed Monitor, Budget, Recurring Registry, Fee Ledger, Anomaly Detector and Digest watch records, feeds and totals.

AI & learning

Fill what nothing else could. Memory learns from your corrections; the AI Categorizer and Merchant Cleanup work on still-empty fields, with low-confidence results parked in the Review Queue.

Every type at a glance#

Transform data — set field values.

TypeWhat it doesRuns on
RulesetWhen these conditions match, set this field to this value. Rules run in order.Transactions, trades, loans, receipts
Lookup TableMaps one field to another from a table you paste in (exact, normalized, or contains).Transactions, trades, loans, receipts

Guard data — change which rows land.

TypeWhat it doesRuns on
FilterKeeps matching records out of your destination entirely.Every data type
SplitReplaces one record with several lines that add up to the original, to the cent.Transactions, trades
Transfer MatcherDetects money moving between your own accounts and tags (or suppresses) both legs.Transactions
Duplicate ScreenFlags look-alike duplicates — same account, amount and description inside a window.Transactions

Watch — notify, never mutate.

TypeWhat it doesRuns on
AlertEmails, Slacks or webhooks you when a synced record matches your conditions.Every data type
Feed MonitorTells you when a feed stops syncing, or keeps syncing but stops producing records.Feeds (no data type)
BudgetAlerts when spend, income, count or a balance crosses a threshold for a period.Workspace totals
Recurring RegistryTracks subscriptions and bills; flags price increases, new ones, and missed charges.Transactions
Fee LedgerClassifies bank fees and interest, and alerts when a month's fees creep upward.Workspace totals
Anomaly DetectorFlags big first-time merchants, amount spikes against history, and transaction bursts.Transactions
DigestSends a scheduled daily, weekly or monthly summary of everything above.Workspace totals

AI & learning — fill what is still empty.

TypeWhat it doesRuns on
MemoryLearns how you categorize records in your destination and reapplies it next sync.Transactions, trades
AI CategorizerFills empty categories from your taxonomy, gated by a confidence threshold.Transactions, trades
Merchant CleanupTurns raw bank descriptors into clean, canonical merchant names.Transactions
Review QueueNot a type — the inbox where below-threshold AI suggestions wait for your decision.Transactions

Some types have no data type

Feed Monitor, Budget, Fee Ledger and Digest watch feeds and workspace totals rather than a record shape, so they have no Data type selector — they contribute from every feed you attach them to. For the record-level types, data types an enrichment cannot work on are shown disabled with the reason (a Duplicate Screen needs a Duplicate Flag field, so it is transactions-only).

The order things run in#

Enrichments do not simply run top to bottom of the grid. Each type belongs to a stage, stages run in a fixed order, and your drag-and-drop order decides the sequence within a stage. That is what makes the results predictable no matter how the grid is arranged:

  1. Guard

    Filter, Duplicate Screen, Transfer Matcher and Split decide which records continue. A suppressed record never reaches a later stage, never teaches Memory, and never fires an alert.

  2. Memory

    Memory fills fields that are still empty, from what you have corrected before.

  3. Lookup

    Lookup Tables apply their mapping — early, so your rules can still override them.

  4. Rules

    Rulesets apply. Later rules win over earlier ones, so put the most specific rules first.

  5. AI fill

    The AI Categorizer and Merchant Cleanup fill only what is still empty, and only above your confidence threshold. Everything else goes to the Review Queue.

  6. Observe

    Alerts, budgets, recurring and fee tallies, and anomaly scoring read the final records. They notify and count; they never change a value.

Between the rules and the AI fill there is one more stage, transform, which rewrites derived values (templates, regex extraction, upper/lower/title case, absolute/negate/round). Rules evaluate before it, and everything in the observe stage sees the transformed result. The in-app rule editor writes plain set-a-value actions, so today these derived actions are authored over the REST API.

Watch types that look at feed health or a whole period (Feed Monitor, Budget, Fee Ledger, Digest, and the Recurring and Anomaly alerts) are also evaluated on a scheduled hourly pass, so they can fire between syncs.

Preview before you sync#

Every feed's Mapping tab has a Live Preview that runs your real records through the whole pipeline without writing anything:

  • The Data tab shows the rows as your destination will receive them, with enrichment-changed cells highlighted.
  • Rows a Guard enrichment would exclude are struck through and labelled with the enrichment that suppressed them, and Split children appear indented under their parent.
  • Alerts show as "would fire (dry run)" instead of sending.
  • The Enrichments tab shows each enrichment with the number of records it matched.

After a sync, the job's execution summary in job history records the same thing for what actually ran: which enrichments fired, how many records each touched, and whether alerts were delivered.

What each plan includes#

Enrichment types come in tiers. Every plan can create the basics; higher plans add the rest.

PlanEnrichment types
Free, StarterRuleset, Alert, Memory, Filter, Feed Monitor
StandardThe above, plus Lookup Table, Split, Transfer Matcher, Duplicate Screen, Budget, Recurring Registry, Fee Ledger, Anomaly Detector, Digest
Professional and aboveEverything, including the AI Categorizer and Merchant Cleanup

How the tier is enforced

Types your plan does not carry appear locked in the New Enrichment dialog with the plan they need. If you downgrade, existing enrichments of a type you can no longer create keep running — except the AI types, which stop spending model time immediately (their documents and settings are kept, so an upgrade resumes them). A workspace can hold up to 100 enrichments in total.

AI types and credits

The AI Categorizer and Merchant Cleanup ride the AI entitlement on Professional and above, and do NOT draw down your AI assistant credits — those are for Ask Penny. There is no model to choose: one managed model serves every workspace, and your accuracy dials are the confidence threshold and the Review Queue.

Next steps#

Build your first enrichment — a ruleset that categorizes transactions as they sync.

Create your first enrichment

  • Creating enrichments: the create → configure → preview → enable walkthrough, using a Ruleset.
  • Memory: let BankSync learn your categorizations instead of writing rules for them.
  • Alerts: email, Slack and webhook notifications on matching records.
  • Review Queue: accept, edit or reject the AI suggestions that fell below your threshold.
  • Configuring field mappings: decide which columns enriched values land in.
  • Changing your plan: what each tier includes, and how upgrades apply.

Use this page with your AI assistant

Every BankSync doc is available as plain Markdown for agents and LLMs.