Adata Berna All articles
Analytics Strategy

The Handover That Never Happened: Why Analytics Teams Break When One Person Leaves

Adata Berna
The Handover That Never Happened: Why Analytics Teams Break When One Person Leaves

Photo: President (1981-1989 : Reagan). White House Photographic Office. 1981-1989, Public domain, via Wikimedia Commons

A resignation letter rarely mentions the queries.

When a senior analyst leaves a US mid-market company, the exit interview covers building access, equipment return, and a final paycheck. It does not cover the fact that she alone knew the orders table double-counts refunds after the fifteenth of the month, or that the revenue figure the CFO quotes to the board is not the number in the dashboard but the number in her personal reconciliation file, hand-adjusted for three enterprise contracts that renew on odd cycles.

Those details leave with her. What remains is a reporting stack that looks intact and is not.

The Knowledge That Never Gets Written Down

Every analytics function accumulates a layer of undocumented operating knowledge, and it almost always takes three forms.

The first is provenance: which table is actually trustworthy. Not which table is documented as trustworthy — which one experienced people open when they need an answer they are willing to defend. In most mid-market environments these are different objects, and the gap between them is knowledge held in heads rather than in a catalog.

The second is logic: why a particular filter exists. The exclusion of a test account, the join that silently drops three hundred historical records, the date field that must be shifted by one day because the source system timestamps in a different zone. None of it is written anywhere, and all of it is load-bearing.

The third is adjustment: the manual reconciliations that live in a spreadsheet on someone's desktop. A quarterly accrual here, a reclassification there, a note that the marketing attribution model is overridden for two business units because the vendor never fixed its tracking.

None of this is malicious. It is the residue of solving problems under deadline. It becomes visible only when the person holding it walks out.

The Bill Arrives in Weeks, Not Months

Week one is quiet. Week two, a dashboard returns a figure that looks roughly four percent low, and nobody is certain enough to say whether four percent is wrong. By week four, three analysts are rebuilding logic they did not write, and the monthly close has slipped two days.

The costs compound in a predictable order. First comes re-derivation: hours spent reconstructing definitions from output rather than from source. Then duplication, as two teams independently rebuild the same report and arrive at different answers. Then decision latency, as leadership waits for numbers they used to receive on the second business day. Then trust erosion, which is the expensive one — once an executive has been burned by a number that changed after the fact, every subsequent report is read with suspicion.

Consulting benchmarks routinely place the cost of replacing a skilled analyst at a substantial fraction of annual salary once recruiting, onboarding, and lost productivity are counted. The more damaging figure is rarely the replacement. It is the rework performed by everyone else in the meantime, and the decisions that were deferred while it happened.

Why Documentation Projects Fail

Most firms have tried to fix this, usually after an audit or a departure. Most of those efforts failed, for reasons that are structural rather than attitudinal.

Documentation gets written episodically, at moments of crisis, and is never maintained afterward. It is written for auditors rather than for colleagues, which means it describes controls instead of assumptions. It is written at the wrong altitude — either a forty-page architecture document nobody reads or a comment block that says the obvious. And above all, it has no owner. A document without a name attached to it is a document that will be stale within two quarters.

Four Practices That Actually Hold

Continuity in analytics is not achieved through heroics or through a documentation sprint. It is achieved through a small number of habits that survive contact with a busy quarter.

A named owner for every core metric. Not a team, not a department — a person. The owner is accountable for the definition, signs off on changes, and answers the question of why the number moved. Twenty to thirty metrics is usually the right scope for a mid-market firm; beyond that, ownership becomes nominal.

SQL in version control, with peer review. Queries that produce executive reporting should live in a repository, not in a browser tab. A lightweight review requirement — one colleague reads the logic before it ships — catches both errors and the quiet accumulation of undocumented exceptions. It also means the reasoning behind a change is captured in the commit history rather than in a hallway conversation.

A runbook for each recurring report. One page, plain language: source systems, filters applied, known quirks, refresh schedule, and the name of the person to call. The quirks section matters most. It is where the tribal knowledge goes to become institutional knowledge.

A reconciliation ledger. Any manual adjustment to a reported figure gets logged with a date, an amount, a reason, and an approver. This single practice converts the most fragile knowledge in the organization — the adjustments living in one person's spreadsheet — into an auditable record.

Run the Handover Before You Need It

Handover checklists are usually deployed in the final two weeks of someone's tenure, which is the worst possible time to use them. Run the exercise quarterly instead, on a rotating basis, against a team member who is not going anywhere.

A workable checklist covers access and credentials, a walkthrough of owned reports and their quirks, the location of any local files that feed official numbers, an inventory of scheduled jobs with their owners, a list of open questions the departing — or rotating — person never resolved, and a named successor for each item. The last one is the point of the exercise.

A Ten-Minute Test

There is a simpler diagnostic. Ask whether an analyst who joined last month could reproduce last quarter's executive reporting pack from source systems without asking anyone a question.

If the answer is no, the bus factor is one, and the meter is running on a cost you have not yet booked. The work of fixing it is unglamorous and cannot be delegated to a tool. But it is the difference between an analytics function that survives a Friday resignation and one that spends the following quarter explaining itself.

All Articles

Related Articles

Closer to the Floor: Putting Manufacturing Analytics in the Hands of Shift Leaders

Closer to the Floor: Putting Manufacturing Analytics in the Hands of Shift Leaders

Fast Enough: The Case Against Buying Real-Time Analytics You Will Never Use

Fast Enough: The Case Against Buying Real-Time Analytics You Will Never Use

Fewer Numbers, Better Decisions: The Executive's Case for Radical Metric Discipline

Fewer Numbers, Better Decisions: The Executive's Case for Radical Metric Discipline