Home Use Cases Work Insights About Contact
Back to Our Work

How an Enterprise Data Management Company Built Record Matching with Human Approval Before Two Records Become One

Operations2
Open the walkthrough

Seven-Step AI Workflow with Human Oversight

Each step shows the workspace, what the AI completed, and what still needs human review.

Steps 3, 4 and 7 need a person. Step 3 is waiting for approval right now. Click it to open the full walkthrough.

Estimates are directional and based on stated assumptions. All names, organizations, and identifying details have been anonymized in accordance with our confidentiality agreements.

The Transformation

How an Enterprise Data Management Company Built Record Matching with Human Approval Before Two Records Become One

Before
A person used to export each system's overnight changes into a spreadsheet and line the formats up by hand.
A steward used to search the store by company name and merge duplicates on judgment alone.
A person used to chase five systems by email to answer one question about an account.
An analyst used to pull last week's extracts and paste the figures into each report by hand.
After
every overnight change, in one shape, with the broken feeds held
every record scored against the store, the grey band routed to a person
22 pairs, briefed, waiting on the steward's call
confirmed changes landed everywhere, the gated one waiting on its owner
one question, answered out of the record, with the fields quoted
Executive Summary

Everything the business holds on one company, matched into one entity and kept true across the systems that disagree about it. The agents pull every overnight change, score every match, land the confirmed ones, answer questions out of the mastered record and send the reports nobody gathers by hand. A person decides what merges, signs the changes the approval rules name, and owns the thresholds. Seven stations, run as a loop: the rule a steward's overrides earn on Friday is the rule the scorer scores with on Monday.

What Was Broken

A person used to export each system's overnight changes into a spreadsheet and line the formats up by hand
A steward used to search the store by company name and merge duplicates on judgment alone
A person used to chase five systems by email to answer one question about an account
The real cost
An analyst used to pull last week's extracts and paste the figures into each report by hand.

What We Built

Seven stations and 15 subagents. Each subagent carries its own tasks and its own refusal.

A4
Feed Puller
Pull every changed record off the six source systems since the last sync
A3
Field Check
Check every record for the fields a match cannot run without
A4
Candidate Search
Search the entity store for every record that could be the same company
A3
Confidence Scorer
Score each candidate pair on the fields that agree and the fields that fight
A1
Pair Briefing
Draft a one-screen brief for each pair: the fields that agree, the fields that fight
A3
Decision Record
Record each decision, who made it, and the brief that was on screen
A3
System Writer
Push each confirmed change back to every system that carries that entity
A4
Workflow Runner
Trigger the workflow each change kind is configured to run
A2
Approval Desk
Match each change against the approval rules the teams configured
A4
Entity Search
Search the entity store and the records behind each entity
A1
Answer Writer
Write the answer in plain lines, each carrying the record it came from
A4
Report Builder
Assemble each scheduled report from the entity store at send time
A3
Anomaly Check
Check every figure against that report's own run of weeks
A4
Decision Audit
Check every steward decision against what the scorer recommended
A2
Change Proposal
Propose the rule or threshold change that would have scored this week's overrides right

How It Runs

1

Record Intake

Nobody is asked anything here

Agents pull every changed record off the six source systems, put each one into the common shape with its source kept, and hold any feed or record that does not check out. Nothing waits on a person here. A quarantined record is never guessed into shape and never loaded broken.

2

Match Scoring

Nobody is asked anything here

Agents search the entity store for every record that could be the same company, score each candidate pair on the fields that agree and the fields that fight, and split the day into merge, review and new. Nothing waits on a person here. A pair in the grey band goes to the steward's queue, never to an auto-merge.

3

Merge Review

A person answers here

Agents draft a one-screen brief for each pair in the grey band, map where every conflicting value came from, and recommend which value survives on each field. A person makes the call. This is where a person works. Everything else exists so that 22 pairs, not 3,196 records, reach this screen.

4

Change Landing

A person answers here

Agents push every confirmed change back to the systems that carry that entity, run the workflows configured for each change kind, and hold any change that hits an approval rule until its owner signs. A person signs here. A change a team put an approval rule on does not land without the person that rule names.

5

Answer Writing

Nobody is asked anything here

Agents answer a question out of the entity store and the records behind it, quote the field and the system every claim came from, and say not found where no system answers. Nothing waits on a person here. An answer the record cannot support is refused, not softened.

6

Report Building

Nobody is asked anything here

Agents assemble each scheduled report from the live entity store at send time, flag any figure that moved further than its own history, and hold a report whose source feed never refreshed. Nothing waits on a person here. A flagged number ships flagged, and a stale one does not ship at all.

7

Match Tuning

A person answers here

Agents check every steward decision against what the scorer recommended, group the overrides that repeat into the pattern behind them, and propose the rule or threshold change that would have scored them right. A person accepts or rejects each one. A person decides here. A threshold decides what merges without a person, so moving it quietly would change who decides.

Where a Person Decides

Step 3, Merge Review. Decide the 22 pairs, the briefs are ready. Rule on the five-against-five pair, or park it for more evidence. a merge is a steward's signature.
Step 4, Change Landing. Sign or refuse the Woodgrove account change, the case is one screen. Look at the two stopped writes only if the support desk asks. an approval rule is a person, not a step.
Step 7, Match Tuning. Accept or reject the two proposed changes. Say whether the two wrongly-merged pairs are a price worth paying. the thresholds are yours to move.

Operating Model

This changes how work flows through the team.

Role
Responsibility
Merge Review owner
Decide the 22 pairs, the briefs are ready. Rule on the five-against-five pair, or park it for more evidence. a merge is a steward's signature.
Change Landing owner
Sign or refuse the Woodgrove account change, the case is one screen. Look at the two stopped writes only if the support desk asks. an approval rule is a person, not a step.
Match Tuning owner
Accept or reject the two proposed changes. Say whether the two wrongly-merged pairs are a price worth paying. the thresholds are yours to move.
Record Intake, when it goes wrong
Look here only if a source team says their change never landed.
Match Scoring, when it goes wrong
Look here only if a match rate moves in a way nobody explained.

What Transfers, What Must Be True

What transfers
A person is in the loop wherever two records become one company, a change lands under an approval rule, or the match rules themselves move. Three stations refuse to proceed without a person: Merge Review, Change Landing and Match Tuning. Everywhere else the agents pull, score, land, answer and report, and reach a person only where the call is theirs to make.
Merge Review stops for a person, and a merge is a steward's signature.
Change Landing stops for a person, and an approval rule is a person, not a step.
Match Tuning stops for a person, and the thresholds are yours to move.
Every subagent says what it will not do. 15 of them do.
What must be true in your environment
The agent can read the systems your records already live in. This one reads 23.
Somebody owns Merge Review and has time for it.
Somebody owns Change Landing and has time for it.
Somebody owns Match Tuning and has time for it.

Failure Modes

What breaks this pattern:

✗ The grey band merges itself

The scorer sees a week of clean auto-merges and starts merging the borderline pairs too. Two different companies become one record, and every downstream system copies the mistake before a person sees it.

✗ Approvals skipped to clear the queue

The queue behind an approval step grows, so the runner pushes changes through without the sign-off. The one change that approval rule existed to catch lands with all the routine ones.

✗ Answers past the record

The writer fills the gap between what the record holds and what the question asked. The reader gets a confident answer no system can back, and builds their own work on it.

✗ Outliers smoothed before the report ships

The check drops or averages the strange number so the report reads clean. The one figure that pointed at a real problem never reaches the person the report was built for.

Directional Outcomes

What the agent counts, and the station that counts it.

These counts are the tallies from one monitored run of the agents. They are not monthly or annual totals.

changed records pulled off six systems
Counted at Record Intake
3,240
stale feeds loaded over today's records
Counted at Record Intake
0
quarantined records with the missing field named
Counted at Record Intake
31
records scored against the entity store
Counted at Match Scoring
3,196
grey band pairs auto-merged
Counted at Match Scoring
0
pairs routed to the steward's queue
Counted at Match Scoring
22
questions answered out of the record
Counted at Answer Writing
9
answers written past the record's history
Counted at Answer Writing
0
answers refused at the record's edge
Counted at Answer Writing
1
scheduled reports sent on their day
Counted at Report Building
5
stale figures shipped as fresh
Counted at Report Building
0
figures flagged for a second look
Counted at Report Building
1
Our measurement policy: We do not publish precise ROI without baseline methodology. Every figure above carries its basis.

What Runs Where

Every step names the subagent that does the work, the record it writes, the thing that raises a question for a person, and what it is allowed to touch. This is drawn from the source, not from a diagram somebody kept in sync by hand.

1Record Intake
subagentfeed-intake
writesrecords/incoming/<record-id>.json
raiseshold-stale-feed
may touchrecords/incoming/**, records/quarantine/**, the source systems read-only
2Match Scoring
subagentmatch-scoring
writesmatches/scored/<record-id>.json
raisesgrey-band-match
may touchmatches/**, records/incoming/** read-only
3Merge Review
GATE
subagentmerge-briefing
writesmatches/review/<pair-id>.md
raisesawait-steward-decision
may touchmatches/review/**, everything else read-only
4Change Landing
GATE
subagentchange-rollout
writeschanges/outbound/<change-id>.json
raisesapproval-rule-hit
may touchchanges/**, the target systems through their connectors
5Answer Writing
subagentask-the-record
writesanswers/<question-id>.json
raisesanswer-not-in-the-record
may touchanswers/**, entities/** read-only
6Report Building
subagentreport-assembly
writesreports/<report-id>.md
raisesstale-source-hold
may touchreports/**, entities/** read-only
7Match Tuning
GATE
subagentmatch-tuning
writesrules/proposed/<rule-id>.json
raisesaccept-rule-change
may touchrules/proposed/**, the decision history read-only

Stack

Every system this agent reads or writes.

System
Read at
Stations
billing
Record Intake
1 of 7
the CRM
Record Intake
1 of 7
the ERP
Record Intake
1 of 7
the answer log
Answer Writing
1 of 7
the approval rules
Change Landing
1 of 7
the contracts store
Record Intake
2 of 7
the decision record
Merge Review
2 of 7
the dependency map
Change Landing
1 of 7
the distribution lists
Report Building
1 of 7
the entity store
Match Scoring
4 of 7
the held-feed log from intake
Report Building
1 of 7
the incoming records
Match Scoring
1 of 7
the last thirty days of scored pairs
Match Tuning
1 of 7
the marketing platform
Record Intake
1 of 7
the proposal log
Match Tuning
1 of 7
the report schedule
Report Building
1 of 7
the six source systems through their connectors
Change Landing
1 of 7
the six systems' field history
Merge Review
1 of 7
the six systems' record history behind it
Answer Writing
1 of 7
the standing match rules and thresholds
Match Scoring
2 of 7
the steward's queue
Merge Review
1 of 7
the support desk
Record Intake
1 of 7
the workflow definitions
Change Landing
1 of 7
Next Step

Want to see if this pattern fits your internal tools?

No build commitment·Real samples, not a demo·Estimate in writing