Merging duplicate customers or reconciling two systems means more than spotting a discrepancy: someone has to define the matching logic, tune it against real data, review the results, and decide what the final output looks like. That's why this work turns into a project, and why so many stall. Recipes carry it through to completion.

Each Recipe is a guided workflow that sequences platform actions to configure a use case end to end. AgentQ reads your data and proposes each step, grounded in what your data actually looks like, not generic defaults. Accept it, adjust it, or configure it yourself. Every step produces an inspectable, editable platform asset, so ongoing validation runs without rework.

What used to take a data engineer days now takes a data steward under three minutes, and it ends with the problem resolved, not documented. The check keeps running against new data, catching the next duplicate or break on arrival and writing a golden record downstream.

Entity Resolution Recipe

The same real-world entity, spelled five ways. Every company has this problem: one customer appears under different names, addresses, or account numbers in every system that touches them. And simple name matching can just as easily merge two different people and miss a true duplicate.

The Entity Resolution Recipe gives teams a guided way to easily find and resolve those matches.

Choose the fields that matter, decide how heavily to weigh each, and set your match threshold. AgentQ can recommend a configuration such as fuzzy matches on names and exact matches on dates of birth, or you can adjust every step yourself.

Before running the full scan, a dry run tests the setup against a sample of your data, so you can tune the configuration before go-live.

From there, matched records are grouped for review. You select the golden record for each entity, materialize a clean, deduplicated dataset for downstream use, and keep a record of which record survived and why. The check stays active, so the next duplicates are caught on arrival.

Where it shines:

• Matching beyond a name. Compare multiple fields like name, address, and account ID to find true duplicates without merging look-alikes.
• Adapting to messy data. Tune thresholds to your dataset, whether you’re working with a noisy partner feed or a clean internal table.
• Keeping the full picture. Materialize the golden set while retaining excluded records for reference, so nothing disappears along the way.

Configuration takes less than three minutes, and scans can complete in as little as five minutes depending on dataset size. One data leader described it as “a TurboTax solution for entity resolution.” That’s a pretty good way to think about it: the expertise is built into the workflow, without taking control away from the user.

Data Reconciliation Recipe

Independent systems can be well-integrated and still disagree. A warehouse and a lake, or a system of record and the report built from it, are all supposed to line up, yet they rarely match perfectly. When they sit on different technologies, telling the “harmless differences” from the real breaks gets much harder.

The Data Reconciliation Recipe helps teams quickly determine which differences are expected and which ones actually need attention, regardless of the technology stack the source and target are in.

Select a source and target table, identify how records should be paired, choose the fields to compare, and set tolerances for differences that are acceptable. AgentQ can recommend the setup and suggest how to resolve mismatches, or you can configure it manually.

Once the scan runs, it classifies every mismatch as added, removed, or changed. You can review differences side by side, down to the individual record, then materialize the reconciled result with a clear record of where the winning values came from. The comparison stays active as a check, so the next break is caught immediately.

Where it shines:

• High-volume reconciliation. Apply the same logic across thousands of comparisons instead of maintaining one-off scripts.
• Financial data. Treat a small market-value variance differently from a share-quantity discrepancy by setting tolerances field by field.
• System migrations. Separate harmless formatting or data-type changes from actual breaks between the old and new systems.

The goal isn’t simply to tell you that two datasets are different. It’s to help you answer the question that actually matters: do my systems reconcile? And give you the evidence to back up the answer even when you’re reconciling tens of thousands of files a day.

Go deeper

• Release notes: September 1
• Release notes: August 18

Have questions? Reach out to your FDE or contact us at support@qualytics.ai

‍

‍