The same person entered three times, once with a maiden name, once with a typo, once from a referral letter, collapses into one record with the history intact. You get a candidate list of likely matches, the evidence for each, and a merge that a human approves rather than a script that quietly overwrites.
Start from the Eligibility Insight Dashboard. It is a live Taskade Genesis app that reads a list of records and shows you what it found and why, which is the review surface a merge queue is built on rather than a deduplication tool. Explore it, then clone it into your own workspace in about ten seconds and point it at your own exports.
The build gives you an import step that accepts several sources at once, normalisation of the obvious things such as name casing, date formats, and phone number shapes, a match score with the reasoning shown, a side-by-side comparison for review, and a merge that keeps the fullest value from each side and records what came from where. Nothing merges silently. Every decision is reversible because the source rows are retained.
The cleanup runs in whichever layout suits the step:
- Table view as the working grid of candidate pairs and merged results
- List view for a straight review queue of pairs awaiting a decision
- Board view to move records through matched, reviewed, and merged
- Plus the rest of the 7 project views
Automations keep it from happening again. On paid plans, Taskade automations check every new intake submission against the existing list and flag a likely duplicate before it becomes a second record. Across 100+ bidirectional integrations, exports pull in from the systems you already run while alerts push out to whoever owns data quality.
The AI agents carry 34 built-in tools including file analysis and persistent memory. The agent learns your corrections, so if you decide twice that two similar names at the same address are different people, it stops proposing that pattern.
Workspace DNA is what keeps a cleanup from becoming an annual event. Memory holds every source row and every decision your team has made about a pair. Intelligence, drawing on 15+ frontier models from OpenAI, Anthropic, Google, and open-weight providers, scores candidates and shows its reasoning. Execution checks each new submission against the list at the moment it arrives, so the duplicate is prevented rather than discovered later.
Running it in your practice
- Export everything at once. Booking tool, spreadsheets, and any list kept separately. Duplicates hide in the gaps between sources, so partial imports miss them.
- Clone the app and set your match signals. Which fields carry weight, and which combinations you consider decisive.
- Review the first fifty pairs yourself. Your corrections teach the agent your practice's edge cases faster than any configuration screen.
- Set the conflict rule. Most recent wins, most complete wins, or a nominated source wins. Deciding once removes hundreds of small judgements.
- Leave the check running on new intake. Prevention costs nothing once it is on, and it is the only thing that stops the pile rebuilding.
Every downstream reminder, recall, and status page gets more accurate once one person is one record, which is why this is worth doing before the rest.
Clone it, invite the team, and the rules are yours. What counts as a match, what wins in a conflict, and what gets retained are decisions your practice makes.
Next: read document intake if your sources include scans, browse the Community Gallery for cleanup apps others publish, or open create to start fresh. This pairs with turning scanned sheets into records, which is where duplicates usually enter, and with household grouping, which handles the family members who look like duplicates but are not.
Frequently Asked Questions
How does it decide two records are the same person?
By the signals you weight: name similarity, date of birth, contact details, and address. Each candidate pair carries a score and the reasoning behind it, so you can see exactly why it was proposed before approving anything.
Does anything merge automatically?
Only if you decide it should, and most practices keep every merge behind a human approval. Even then the source rows are retained, so a wrong call can be undone.
What happens to conflicting values?
You set the rule: most recent wins, most complete wins, or a chosen source wins. The comparison view shows both values side by side so the exception is easy to handle manually.
Can it handle family members with the same address?
That is the classic false positive, and the agent is tuned to treat shared address plus different date of birth as different people. For deliberate grouping, see household grouping.
Can we run it on an ongoing basis?
Yes, on a paid plan. A new submission is checked against the list at the moment it arrives, so the duplicate is caught at intake instead of discovered a year later.
What about records that came from a scan?
They are treated the same once extracted. Because the reader keeps a link to the source document, a merged record still points at the page a value came from.
Is any of this altering clinical information?
No. It reconciles administrative identity fields: names, contact details, and identifiers your practice uses. Anything clinical stays in whatever system your practice keeps it in.
