Question and scope

This protocol measures how often one cross-team request appears as separate records in more than one tracker. The result is a duplication rate, a count of affected requests, and a short list of the channels where duplicates begin. A team can use the rate to decide whether intake needs a single front door or whether its existing tools need clearer ownership. The protocol does not estimate the cost of the duplication, because time and rework are not visible in most trackers.

Scope is one coordination team, one defined window, and a named set of systems. Include requests that crossed a team boundary and were open at any point during the window, whether they lived in a ticketing tool, a spreadsheet, a project board, or a shared inbox. Exclude internal tasks that never left the team and requests that were cancelled before any work started. State the systems and the observation window so another analyst can rebuild the same population.

What counts as a duplicate

A duplicate exists when two or more records describe the same deliverable for the same receiving team and both records were treated as actionable. The wording does not have to match, and the titles may use different names for the same output. What matters is that a reasonable coordinator could assign either record and expect the same work to be done. A record that merely mentions the request without asking for action is a reference, not a duplicate.

Distinguish a duplicate from a related request. Two records can share a stakeholder or a system while asking for different outputs, and merging them would hide real scope. A request to correct a vendor address and a request to resend a vendor packet may be related, but they are separate commitments. The coding team should write a one-sentence rule for this boundary before reading records, then apply it consistently.

Unit of analysis and exclusions

The unit is one distinct request, represented by a cluster of one or more records. Each cluster has one deliverable, one receiving team, and one needed-by date, even when the records disagree about the date. Record every source record in the cluster, including the one selected as primary. A cluster with one record is a non-duplicate and stays in the denominator.

Exclude records that were created by a system sync of the same item, because those are copies rather than independent entries. Exclude test tickets, template records, and entries that were closed as spam before review. Report the exclusion count by category so the denominator is transparent. Keep a dated log of every exclusion, because a later reviewer will need to see what was removed and why.

Building the candidate set

Start with an export from each tracker that covers the observation window. Include closed records as well as open ones, since a duplicate may have been closed quickly after someone noticed the overlap. Add requests that appear in recurring meeting notes but never reached a tracker, because those informal entries are often the second copy. The goal is a complete list of candidate records, not a clean list of confirmed duplicates.

Give every record a stable identifier before any matching begins. Record its source system, creation date, requester, receiving team, and current status. Freeze the export and save the date it was taken, since live systems change while an analyst is coding. Matching against a moving dataset makes the result impossible to reproduce and invites disputes about which record was present.

Normalizing match fields

Normalize the fields that will drive matching without erasing the original values. Lowercase the requester names, expand common abbreviations, and map each team alias to one canonical team name. Convert dates to a single format and flag records with no date. Keep the raw values beside the normalized ones so a reviewer can check the transformation. Store the normalization rules in the working file so the next analyst can repeat them exactly.

Do not normalize away meaningful differences. Two similar titles may describe different deliverables, and two different titles may describe the same one. Use the normalized fields to shortlist candidate pairs, then read the full record before deciding. The matching model should reduce reading time, not replace judgment about the work. When a pair is ambiguous, write a note and send it to the second reviewer instead of guessing.

Coding match confidence

Code each candidate pair as confirmed, probable, possible, or not a match. Confirmed means the records share a deliverable, a receiving team, and an actionable request, and at least one person has said they are the same. Probable means the evidence strongly suggests one request but no one has confirmed it. Possible means the pair shares several fields but the deliverable may differ, while not a match means the difference is clear.

Report confirmed and probable separately, and keep possible pairs out of the headline rate. A generous definition can make duplication look common, while a strict one can make it disappear. Pair every rate with the rule that produced it and the count of pairs in each confidence band. That presentation lets a reader judge whether the headline reflects the team's actual intake problem.

Sampling and verification

Use a census when the number of records is small enough to review in full, which is common for one coordination team over a quarter. Above roughly two hundred records, draw a random sample and publish the sampling method. Stratify by source system so a high-volume tracker does not dominate the result. Freeze the sample before matching so the analyst cannot choose easier pairs.

Have a second reviewer recode at least ten percent of the sample, including every confirmed duplicate. Resolve disagreements by returning to the source records and the written rule, not by choosing the answer that feels more plausible. Record the resolution and update the rule if the disagreement reveals an ambiguity. If agreement stays low, rewrite the definitions and recode the full sample before reporting.

Decision rules before coding

Decide before coding whether a record that was closed as a duplicate still counts in the numerator. Counting it captures the intake problem, while excluding it measures only unresolved duplication. Both are useful, but the report must say which one it uses. A common approach is to report both and treat the closed-record count as evidence of how quickly the team noticed.

Decide how to treat records that were created after the original request was already delivered. These may be duplicate requests for a new cycle rather than copies of the old work. Code them as a separate category and report the count outside the main rate. The distinction prevents a recurring monthly request from being mistaken for a coordination failure. Write the rule in the coding guide before the team reviews any candidate pairs.

Analysis and reporting

Report the number of distinct requests, the number of duplicate clusters, and the share of requests with at least one duplicate record. Show raw counts beside every percentage, especially when the population is small. Add the median number of records per cluster and the share of clusters that span more than one system. Those figures show whether duplication is a rare edge case or a routine intake pattern.

Break the result down by source channel, receiving team, and request age. Suppress groups with fewer than five units so one unusual case does not look like a trend. Pair the rate with the time each duplicate stayed open, because a duplicate that is caught in an hour is a different problem from one that runs for a month. Use the duplicate request check to turn the findings into an intake routine.

Limits and sources of error

The sources cited here describe internal control, project management, and records practice in general terms; they do not provide a benchmark for duplication in a specific team. A record that looks unique may be a duplicate whose matching detail was never written down. A record that looks duplicated may be a new request for the same output in a later period. Treat the rate as a description of the records, not as proof about how the team behaves.

Informal requests are undercounted when they never reach any tracker. Check a sample of messages and meeting notes to estimate the size of that gap, and report it as a limitation. If a large share of requests is invisible to the systems, fix intake before setting a target. A precise rate built on an incomplete candidate set is less useful than a range with a clear caveat.

Review cadence

Review the result at a fixed point and end with one change, one owner, and one check date. A practical first change is to require a search for an existing request before a new record is created. Repeat the measurement after two full cycles before setting a target, because a single month may reflect a reorganisation or a tool migration. Keep the definitions and the sample method with the result so the next reviewer can compare like periods.

Retire the measure if the intake path is stable and the rate stops changing. If the team keeps it, publish the confirmed and probable shares together, because a strict headline can hide a larger probable problem. Use the weekly blocked-work review to confirm that merged requests actually stay merged. The purpose is a queue that answers a status question once, not a perfect count.

Sources