Systems and records
Support repeated contact dataset: grouping contacts into one case
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Group contacts into cases before labeling, and label outcomes only per case after a maturity window. Link tickets through follow-ups, merges, parent cases and issue links first, then through a documented inference rule. Record the problem key, policy version, promises, handoffs and corrections, and mark the case resolved only when no further contact on that problem arrives in the window or downstream records confirm the fix.
System export
Support repeated contact
Key takeaways
- A solved ticket records finished work, not a solved problem; outcomes belong to the case.
- Follow-ups after closure, merges and channel switches split one problem into several tickets in most help desks.
- Explicit links are strong grouping evidence; inferred links need a rule version and a confidence grade.
- A case is resolved only after a maturity window with no repeat on its problem key, or downstream proof such as a posted credit.
- Grouping contacts across channels makes records more identifying, so privacy review should see whole cases.
How do you stop repeat contacts about one problem counting as separate successes?
Group contacts into cases before you label anything, and label outcomes only at the case level after a maturity window has passed. A ticket marked solved records that an agent finished a piece of work; it does not record that the customer's problem went away. When the same customer writes in again about the same problem, the earlier solve was an intermediate step, not a success.
In practice that means three things. Every contact carries a case ID assigned by explicit links where they exist and by documented inference where they do not. Every case carries one problem key, the policy version in force, the promises made and the corrections that followed. And every case outcome waits until a fixed window after the last contact has closed with no further contact on that problem key, or until downstream evidence such as a completed refund or a shipped fix confirms it.
A support repeated contact dataset built this way stops rewarding the pattern that inflates ticket metrics: the quick close that sends the customer back through another channel. The rest of this page gives the grouping evidence, a field-level worksheet, the reconciliation rules and an illustrative case.
Why a help desk records one problem as several solved tickets
Because the ticket, not the problem, is the unit every help desk counts, and several routine mechanics split one problem into several tickets. Knowing which mechanic created each split tells you which evidence to look for.
Status rules are the first source. In Zendesk, a solved ticket reopens if the requester replies, but once the system moves it to closed, a reply creates a follow-up ticket instead. The standard automation closes tickets four days after they are set to solved, and admins can change that window. A follow-up references the closed ticket and copies most of its details, but it is a new ticket with its own ID and its own solve.
Merges are the second. Zendesk closes the source ticket when it is merged and adds a closed_by_merge tag, and the merge comment carries only the most recent public comment from the source with a link back. Salesforce case merge lets an admin choose whether non-master cases are deleted to the recycle bin or retained with a merged status. In Dynamics 365 Customer Service, merged cases move their activities to the case they were merged into. Each system leaves a different footprint, and an export that drops closed or deleted records loses the evidence that two contacts were one problem.
Channel switches are the third. A customer who emails, then chats, then calls usually generates three records with no system link at all, which is why inferred grouping has to be part of the method rather than an afterthought.
Which links prove that two contacts belong to the same case?
Explicit system links are strong evidence; matching customers, products and timing is weaker evidence that needs a recorded rule and a confidence grade. Store the basis of every grouping so a reviewer can audit it later.
| Link evidence | Where it comes from | Strength | How to record it |
|---|---|---|---|
| Follow-up reference to a closed ticket | Help desk follow-up mechanism | Strong | grouping_basis = follow_up, with source ticket ID |
| Merge record or merge tag | Help desk or CRM case merge | Strong | grouping_basis = merge, keep the closed or retained source record |
| Parent and child case relationship | CRM case hierarchy | Strong for shared incidents, weak for the individual customer outcome | Group the incident; label each customer case separately |
| Duplicates link to an engineering issue | Issue tracker link types such as duplicates and is duplicated by | Strong for the defect, not for the customer | Attach the issue ID and its resolution date as outcome evidence |
| Agent note citing an earlier ticket | Internal comments | Medium | grouping_basis = agent_reference, with the cited ID |
| Same customer, same product, same problem category within a set window | Ticket fields and timestamps | Weak to medium | grouping_basis = inferred, with rule version and window length |
| Similar free text from the same customer | Text similarity over subject and body | Weak | Candidate only; a reviewer confirms before grouping |
The case grouping and outcome reconciliation worksheet
The worksheet below is the original asset: one row per case, with fields a customer operations lead can fill from a ticket export, a CRM and the billing or engineering systems that hold downstream evidence. Example values are illustrative.
| Field | What it records | Rule | Illustrative value |
|---|---|---|---|
| case_id and member_contacts | Case key and every contact ID in it, with channel | A contact belongs to exactly one case | C-104: email T1, chat T2, phone T3 |
| grouping_basis and confidence | Why each contact joined the case | Inferred links need a rule version | T2 follow_up; T3 inferred, rule v2, 14-day window |
| problem_key | Customer, product or account, and problem category | A new problem on the same account opens a new case | Account A, invoicing, duplicate charge |
| policy_version | Support or refund policy in force at first contact | Later policy changes do not relabel earlier promises | Refund policy revision in force on day 0 |
| promises | Commitments made to the customer, with owner and due date | Each promise is later marked kept, late or broken | Credit within 5 business days, billing team |
| handoffs and corrections | Transfers between teams and any reversal of an earlier answer | A correction marks the earlier answer as wrong, not resolved | Tier 1 to billing; first answer corrected |
| solves and reopens | Each solve timestamp and each reopen or follow-up | Counted per case, not per ticket | 3 solves, 0 reopens, 1 follow-up |
| maturity_window_end | Date after which no further contact means resolved | Window starts at the last solve | Last solve plus 30 days |
| case_outcome | Resolved, failed, unknown or open | Assigned only after the window or downstream proof | Resolved, on the third contact |
| outcome_evidence | Record that confirms the outcome outside the help desk | Required for resolved when a promise was made | Credit memo posted on day 10 |
What proves the problem stayed resolved after the ticket closed?
Silence for a defined window, plus downstream proof where the fix happens outside the help desk. Neither alone is enough. Silence can mean the customer gave up, and a closed engineering issue can mean the fix shipped for some customers but not this one.
Help desk metrics only see part of the picture. Zendesk's one-touch reporting counts tickets solved with one agent reply, and a ticket that reopens and gets a second reply stops counting as one-touch. A follow-up created after the original closed is a separate ticket, so the original can keep its one-touch credit even though the customer came back. That gap is exactly why the case, not the ticket, has to carry the outcome.
Downstream evidence depends on the problem type. Billing problems close with a posted credit or refund in the billing system. Defects close with a fix released in a version the customer actually runs. Account changes close with the setting visible in the product's audit log. Record the evidence ID and its date in the worksheet, and leave the outcome unknown when the evidence system is not available.
Ticket status versus case outcome: the reconciliation rules
Apply the rules in order and record which rule fired. The ticket-level column shows what a status-based export would report; the case-level column is the label that belongs in a dataset.
| Pattern in the records | Ticket-level view | Case-level label | Reason |
|---|---|---|---|
| Three contacts, last one solved, no contact in the window | Three successes | One case resolved on the third contact | Earlier solves were followed by recontact |
| Solved, then a follow-up two days after closure | Two successes | One case; first solve not a resolution | Follow-up shares the problem key |
| Two duplicates merged into one ticket | One success, one closed | One case | Merge is explicit evidence of one problem |
| Workaround given, bug fix ships later | One success at the workaround | Resolved at the fix date, if the customer runs that version | Workaround is an interim answer |
| Credit promised, never issued, customer stops writing | One success | Failed: promise broken | Downstream evidence contradicts the solve |
| No contact after the solve, no downstream system available | One success | Unknown | Silence alone is not proof |
| New contact on the same account about a different problem | Reopen or new ticket | New case | Different problem key |
Worked scenario (illustrative): one billing error, three channels
The details here are illustrative and describe no real company. A customer emails on day 0 about being charged twice. The agent explains that a credit will be applied and sets the ticket to solved on day 1. The standard four-day close moves it to closed on day 5.
On day 6 the customer replies to the old email thread. Because the ticket is closed, the reply creates a follow-up ticket. A second agent confirms the credit is pending and solves it the same day. On day 9 the customer calls. The phone ticket has no system link to the first two, but it matches the account, the invoicing category and the 14-day inference window, so it joins the case as an inferred member. The phone agent hands off to billing, billing posts the credit on day 10, and the ticket is solved.
A ticket-level export reports three solved tickets, two of them one-touch. The worksheet reports one case, resolved on day 10 by the third contact after a handoff, with one promise kept late and a correction to the first answer's timing. With a 30-day maturity window the label becomes final on day 40 if the customer does not write again about this charge.
The counterexample matters as much. If the same customer writes on day 25 about a different invoice, the problem key differs and a new case opens; the first case's outcome stands. If instead the day 25 contact says the credit was reversed, it joins the first case and the outcome changes to failed.
Which assumption changes the outcome, and how to show the uncertainty
The maturity window and the inference rule move the labels most, so publish both and show how the labels change when they move. A short window turns slow repeat contacts into false resolutions; a long window attaches unrelated later contacts to old cases.
Show uncertainty directly in the data rather than in a footnote:
Checklist · 5 items
- Record the window used for each case and, for a sample, the label under a shorter and a longer window, so a buyer can see how sensitive the outcome is.
- Grade every grouping link as explicit or inferred, and let downstream users filter to explicit-only cases.
- Keep unknown as a first-class outcome instead of defaulting silence to resolved.
- Version the problem-key definition, because a coarse category such as billing merges distinct problems and a fine one splits real repeats.
- Mark cases whose outcome depends on a system you could not export, such as a billing platform or issue tracker, so the gap is visible.
0 of 5 ticked. Tick each item as you go.
Linking contacts raises the privacy stakes
Grouping makes each case more identifying than any single ticket, because it joins an email address, a phone recording, chat transcripts and account details into one record. A privacy reviewer should see the grouped case, not only the individual tickets, before anything leaves the company.
Use one consistent pseudonym per customer across channels so the grouping survives de-identification, remove direct identifiers and payment details from free text, and decide separately whether call audio or screenshots belong in scope. Keep the distinctions clear: access to the help desk, privacy permission over the people in the records, ownership of the records and a contractual right to license them are four different questions, and de-identification answers none of the last three by itself.
When reconciled case histories become a licensable dataset
Case-level histories with explicit grouping, promises, corrections and mature outcomes are scarcer than raw ticket exports, because most exports flatten exactly the links this worksheet preserves. They are also the records an AI support agent needs to learn when a reply actually solved the problem.
Support teams that want to offer such histories can work with SourceX, which carries selected case records through rights review, contracting, preparation and delivery. Checking fit takes a short assessment with no upload, and grouped cases leave the help desk only after the company has approved the scope and signed.
Frequently asked questions
What should a case grouping and outcome reconciliation worksheet include?
One row per case with every member contact and its channel, the basis and confidence of each grouping link, a problem key, the policy version, promises with owners and due dates, handoffs and corrections, solves and reopens, a maturity window and a case outcome with downstream evidence. The rules column matters as much as the fields, because it stops ticket statuses from becoming labels.
Is a reopened ticket the same as a repeat contact?
No. A reopen happens when a customer replies to a solved ticket and the same ticket goes back to open. A repeat contact can also arrive as a follow-up after closure, a new email, a chat or a call, each with its own ticket ID, so reopen counts alone undercount repeats.
How long should the maturity window be?
Long enough to cover how customers usually come back about that kind of problem, which differs between a password reset and a billing dispute. Choose it per problem category, record it in each case, and show for a sample how labels change under a shorter and a longer window.
Can merged tickets be recovered from an export?
Often only partly. Depending on the system and settings, merged source records may be closed with a tag, retained with a merged status or deleted, and the target may carry only the latest comment from the source. Export closed and merged records deliberately, and record which merge behavior the system used.
Who should review a grouped support dataset?
A customer operations lead should confirm the problem keys, the reconciliation rules and the downstream evidence, and a privacy reviewer should review grouped cases, because linking contacts across channels makes records more identifying than single tickets.
Sources
- In Zendesk, requesters cannot reopen closed tickets but can create a follow-up request that references the original closed ticket. Source
- A solved Zendesk ticket reopens when the requester replies; a closed ticket stays closed and a reply creates a follow-up; the standard automation closes tickets four days after they are solved. Source
- When Zendesk tickets are merged, the source ticket is closed with a closed_by_merge tag and the merge comment includes the most recent public comment with a link to the closed ticket. Source
- Zendesk one-touch tickets are those solved with one agent reply; a reopened ticket that gets a second agent reply no longer counts as a one-touch resolution. Source
- Salesforce case merge combines duplicate cases into a master case; admins choose whether non-master cases are deleted to the recycle bin or retained with a merged status. Source
- In Dynamics 365 Customer Service, merged cases' activities are moved to the case they were merged into. Source
- Dynamics 365 parent and child case settings let admins close all child cases when the parent closes or prevent closing the parent until children close; by default they close independently. Source
- Jira work items can be linked with the duplicates and is duplicated by link pair, and admins can configure link types. Source
See if your company qualifies
A short company assessment. No data uploads are needed.