Data reconciliation that knows which source to trust for which field
When your POS, Google Business Profile, Yext, HR, and CMS disagree about hours or services or manager bios, automatic reconciliation decides which one to trust — by rules you configure.
The problem
Your POS says the Denver store is open until 10 PM. Google Business Profile says 9 PM. The website says "See store for hours." Yext says 10 PM Monday through Saturday, 9 PM Sunday. Which one is right? Whoever logged in last.
Your ops analyst spends Wednesdays going line by line through a Google Sheet trying to figure out which source to trust for which field at which location. Multiply that by manager bios that drift between HR and the location page, services that exist in the POS but not the CMS, license status changes that legal forgot to tell marketing about, and acquired-brand portfolios where four source systems disagree across six brands.
Generic enterprise data-quality platforms (Informatica DQ, IBM InfoSphere) solve this for data engineers at $50,000 to $500,000 a year with three-to-twelve-month implementations. CDPs (Tealium, Segment) resolve customer identity, not location records. SQL deduplication scripts produce CSVs that nobody reviews on time. The operations analyst owns the spreadsheet, and the spreadsheet owns the operator.
What success looks like
Every disagreement between source systems resolves against rules you configure. POS owns hours. Google Business Profile owns categories. HR owns manager bios. Your compliance officer owns license status. When two equally trusted sources disagree on the same field, freshness breaks the tie. Anything the rules cannot resolve routes to a single human review queue with full context — and the reviewer's prior decisions on similar conflicts.
Every decision your team makes feeds back as training signal. After two to three months, repeat conflicts auto-resolve and the human queue shrinks to genuine edge cases. License status and HIPAA fields always route through compliance regardless of freshness. PE roll-up reconciliation handles cross-brand portfolios without you mapping every acquisition by hand.
The audit trail builds itself — every decision logs the source inputs, the chosen value, the applied rule, and the reviewer (if any) for six-to-seven-year regulator-defense retention.
How most operators solve this today
Two tiers of tools exist, plus DIY. The middle tier — anything purpose-built for multi-location marketing — is missing:
Enterprise data quality (Informatica DQ, IBM InfoSphere, Talend Data Fabric, Trifacta/Alteryx)
$5,000 to $500,000/year
Built for data engineering teams at enterprise scale. Three to twelve months of implementation. Not built for multi-location operations data (addresses, hours, services, manager bios, license status).
Customer data platforms (Tealium, Segment, mParticle)
$120/month to $50,000+/year
Resolve customer identity across touchpoints. Not built for location-record reconciliation. Different data shapes, different rules, different audit-trail requirements.
SQL deduplication scripts + manual CSV review
Engineering time + ops analyst time
Data team writes queries. CSV lands in someone's inbox. Nobody reviews it on time. No rules engine, no learning, no audit trail. Falls apart past 50 locations.
In-house person + master spreadsheet
$60,000 to $90,000/year, 10-20 hours/week
Operations analyst owning the master spreadsheet. Workable under 50 locations. Becomes the rate-limiter past that.
Build it in-house
Senior engineer ($130-220k) + ongoing maintenance
You can build a rules engine. The learning loop on top of it is the hard part. Years of decisions to train a model you could have inherited from a vendor.
What changes when this is an agent skill
Every cross-source disagreement resolves against rules you configure. You say which system wins for which field per vertical. When two equally trusted sources disagree, freshness breaks the tie — with staleness thresholds you can tune. Anything still unresolved routes to a single human review queue with full context plus the reviewer's prior decisions on similar conflicts. License status and HIPAA fields always route through compliance regardless of the rules.
The system learns from your team. Every decision your reviewer makes feeds back as training signal. After two to three months, the auto-resolution rate climbs and the human queue shrinks to genuine edge cases. Industry-specific rules apply automatically where regulators require them. PE roll-up portfolios that share back-office systems across acquired brands get per-brand configuration.
Every decision logs the source inputs (system, field value, timestamp), the chosen value, the rule applied, the reviewer if any, and the downstream outcome — retained for six to seven years for regulator defense.
The total cost replaces $50,000 to $500,000 a year of enterprise data-quality software plus the operations analyst's time on the master spreadsheet.
Agents that include this skill
Skills live inside agent rentals. To get this skill in production, hire any of the agents below — context-tuning at onboarding is included in the first month.
Master Record Canonicalization Agent
Ingests every operator source system, resolves per-location fact conflicts, and emits the canonical master record downstream agents consume.
FAQ
- What does the reconciliation actually do?
- When two or more source systems disagree about the same field — hours, services, address, manager bio — it picks the right value against rules you configure. Where the rules cannot decide, it routes the conflict to a single human review queue with full context.
- How is this different from Informatica or IBM InfoSphere?
- Those are enterprise data-quality platforms aimed at data engineering teams, with three-to-twelve-month implementations. This is purpose-built for multi-location operations data (addresses, hours, services, manager bios, license status) with rules configuration your operations director can read.
- Is this the same as entity resolution?
- Functionally yes. Data engineers call it entity resolution. Operators call it "which system is right about my Denver store?" Same operation, different vocabulary.
- How does the learning loop work?
- Every decision your team makes in the review queue feeds back as training signal. After two to three months, the auto-resolution rate climbs significantly and the human queue shrinks to genuine edge cases the rules cannot reach.
- What if multiple sources are equally trustworthy for the same field?
- Configure them as co-equal. Freshness breaks the tie. If freshness is also equal, the conflict routes to a human reviewer with full prior-decision context.
- What about regulator-sensitive fields like license status or HIPAA-flagged data?
- Industry-specific rules apply. License status always routes to the compliance officer regardless of freshness. HIPAA fields require explicit human confirmation per record.
- How does this work with our data ingestion?
- The ingestion layer pulls data from each source. This reconciles the conflicts between them before anything goes live. Both feed your master record, which then propagates changes to your downstream marketing systems.
- What does the audit trail look like?
- Every decision logs the source inputs (system, field value, timestamp), the chosen value, the rule applied, the reviewer if any, and the downstream outcome — retained for six to seven years.