Keeping one business event consistent across systems that disagree
When the same fact lives in four systems, the question is not how to copy it but what to do when they disagree — which they will, because updates fail, arrive late, or are changed by hand at the other end. Modeled as an orchestrated process, the event is applied to each system, the result is compared, and only a genuine conflict reaches a person.
A reference architecture: how RoboAgentix would model this process on UiPath. It is not a description of a delivered project, and no client, result or metric is implied.
What starts it
A business event is published — a customer record changes, a contract is signed, an employee moves.
Why it is orchestrated rather than scripted
- Systems accept updates at different speeds, and the process has to wait for the slow one rather than declare success early.
- A partial failure across four systems needs a defined path back, not a retry loop that gives up quietly.
- Divergence is normal and recurring, so the reconciliation has to be part of the process, not a monthly clean-up.
How it runs on UiPath Maestro
The process is defined in BPMN and Maestro runs it across agents, robots and people, holding its state for as long as the work takes — including while it waits on an approver.
Step by step
- 01
Event publishedSystem
A change is committed in the system that owns the fact, and published as one event.
- 02
Apply to each systemRobot
The change is applied to each downstream system through its connector, in the order dependencies require.
- 03
Compare the resultsAgent
What each system now holds is read back and compared against what was sent, field by field.
- 04
Do they agree?Maestro
The orchestrator routes on whether every system reflects the event, and on whether the difference is one the rules can settle.
- 05
Resolve the conflictPerson
A genuine conflict reaches a data steward with both values, their sources and their timestamps.
- 06
Apply the resolutionRobot
The agreed value is written back to whichever systems were wrong, and re-read to confirm.
- 07
Differences reconciledSystem
Every system reflects the event, and the divergence and its resolution are written to the audit record.
What each agent is allowed to decide
An agent is only useful in a governed process if its authority is written down. Each one below has a defined input, a decision it may make on its own, and the condition that obliges it to hand over.
Reconciliation Agent
Explains why the systems differ, rather than only reporting that they do.
Reads
- What was sent and what each system now holds
- Which system is authoritative for each field
- Update timestamps and recent change history
Decides on its own
Must escalate when
Remediation Agent
Proposes what should happen to each divergent field.
Reads
- The reconciliation finding
- The precedence rules for that record type
- The constraints of the receiving system
Decides on its own
Must escalate when
What the robots do
Deterministic steps stay with robots — the same action every time, including in systems that never exposed a usable API.
Distribution robot
Applies the event to each downstream system and captures what each one returned.
Runs in
Remediation robot
Writes the agreed values back and re-reads to confirm they held.
Runs in
Do they agree?
Every system reflects the event, or the only differences are ones the precedence rules resolve without judgment.
- Passes
- Closed as reconciled.
- Does not pass
- Held for a data steward, with the process open until the conflict is settled.
Modeled as a decision table, so the threshold can be changed by the people who own the policy without redeploying an automation.
The human step
Who
What they see
What they decide
People are in this process because judgment or authority belongs to them — not as a fallback for a step that failed.
What this design deliberately does not do
An automation is only trustworthy if its limits are written down as plainly as its capabilities.
- It does not become a master record. Nothing here holds its own copy of the fact for the other systems to be reconciled against — the system that owns the field stays the source, and the process only makes the others agree with it.
- It does not roll back a system that already accepted the change. There is no distributed transaction across the connected systems of record. A wrong value is corrected by a later write that is itself recorded, not by unwinding the first one.
- It does not edit the precedence rules on its own. When a steward decides a rule was wrong, the decision table is changed by a person and applies to the next event, not retroactively to instances already closed.
- It does not sweep for drift nobody published an event for. A hand edit at one end or a bulk load that never raised an event is outside this process; it is only seen if a later event touches the same record.
Failure has a defined path
Systems go down, confidence drops, approvers go on leave. Each of these is a modeled branch rather than a stuck instance.
- A downstream system is unreachable, or refuses the update, while the others have already taken it.
- The instance stays open and holds the change queued for that system rather than reporting success. The systems that did accept it are left as they are, the instance is marked partly applied, and it cannot close until that system either accepts the change or the difference is put in front of a steward.
- The Reconciliation Agent cannot tell whether a difference is lag, a rejected update or a real conflict, and its confidence in the classification falls below the threshold.
- It fails closed and calls the field a conflict. The Remediation Agent is not asked for a correction on that field, no write is attempted, and the field goes to the data steward marked as unclassified with both values and their timestamps.
- The data steward never acts on the conflict.
- The process keeps waiting rather than timing out. The field stays marked divergent, the task is reassigned along the escalation path defined for that record type, and later events touching the same field join the open instance instead of starting a second one that would overwrite the disputed value.
- A system accepts the write, but the read-back returns something other than what was sent — a truncation, a rounding rule, a mapping the receiving end applies.
- It is classified as a transformation by that system, not as a conflict, and the same write is not retried because it would land the same way. If a rule for that field is registered the difference is recorded as expected; if not, it is held for a steward as a mapping question rather than a data question.
Data, rollout, and what to watch
Data it touches
- The process moves fields that already exist in the connected systems and creates no new store of its own, beyond the instance state and the before-and-after values kept for the reconciliation record.
- Fields the organization marks restricted are compared as a match or no-match result rather than by value, so a divergence is still detected without the value entering an agent's context or its reasoning trace.
- Per-system credentials stay in Orchestrator, scoped to the records and fields each connector touches, and are never among the inputs an agent reads. The agents see values and timestamps; the robots hold the access.
- Instance state and the retained before-and-after values sit in the tenant and region the organization chooses, and carry the retention period of the record they describe rather than one set by the process. When the source record is deleted or anonymized, the instance history for it follows.
How it goes live
- 01
Shadow
Events are consumed and nothing is written. Each system is read rather than updated, the comparison runs, and the reconciliation findings accumulate against real traffic. The gate to the next phase is that differences the agent calls lag do clear on their own, that the precedence rules cover the fields which keep appearing, and that the fields with an ambiguous authoritative source have been named and settled.
- 02
Supervised
The distribution robot applies the event for real, but every remediation write is proposed rather than executed and a data steward releases it. Stewards see the correction the agent proposed beside what they would have chosen. The gate is that the two agree over a defined review period; where they do not, the precedence rule is rewritten rather than the agent re-instructed.
- 03
Unattended with exceptions
Differences the precedence rules settle are written back without a person. Real conflicts, ambiguous authority, and anything that would overwrite a more recent human edit still go to Action Center. Record types cross this gate one at a time, each on its own evidence from the supervised phase.
Signals worth watching
- The share of events that reach full agreement on the first pass, and whether it shifts after a release in any connected system.
- Which system is most often the one still behind at the comparison step, and whether that is the same system month to month.
- Fields that diverge again on a record already reconciled, which usually means something at the other end is writing the old value back.
- How often a steward decides against the precedence rule the agent applied — a rule that is regularly overruled is a rule that needs changing, not an agent that needs tuning.
Named without target values. What good looks like depends on your volumes and your risk appetite, and is set with you rather than asserted here.
The UiPath components this design uses
UiPath Maestro
Runs the BPMN process and holds its state for as long as the work takes, including while it waits on a person or a system that is down.
UiPath Agent Builder
Where each agent is defined — its instructions, its tools, its inputs and outputs, and the escalation that sends work to a person.
UiPath Integration Service
Connectors to the systems of record, so the process talks to them through a maintained integration rather than a bespoke script.
UiPath Orchestrator
Schedules the robots, holds queues and retries, and issues the least-privilege credentials each automation runs under.
UiPath Action Center
Where a human step actually lands. The approver gets a task with the agent's reasoning and sources attached, and the process resumes on their decision.
Systems it touches
- The system that owns the fact
- Downstream systems of record
- Integration and event layer
- Reporting and data warehouse
What the run leaves behind
- The event as published, and what was sent to each system
- What each system returned, and where they diverged
- The reconciliation finding and the rule applied
- Any human resolution, its reason and its author
Other representative workflows
Talk to us about a process like this.
The most useful first conversation is about a real process — where it stalls, who approves what, and which systems it touches.