The first ninety days off Tableau

A BI migration plan that starts with a file list is already going wrong. The estate is not the work and the import is not the work. The work is agreeing what the numbers mean, and every plan that skips it discovers it in week seven with no budget left for the conversation.

Here is a sequence that puts the hard part early, keeps the old estate running, and gets something real in front of people in a fortnight.

gantt
    dateFormat  YYYY-MM-DD
    axisFormat  %b %d
    todayMarker off
    section Prove it
    Deploy, import real workbooks        :done, p1, 2026-01-05, 14d
    section Agree it
    Settle the top metrics  (slips here) :active, p2, 2026-01-19, 28d
    section Compare it
    Run both estates, reconcile          :p3, 2026-02-16, 28d
    section Move it
    Readers first, then authors          :p4, 2026-03-16, 21d

Thirteen weeks. The second bar is the one that slips, and it slips because it is the only bar that is not a technical task.

A worked example

One tranche, thirteen weeks, as it actually ran

A 1,200-person insurer moving its claims and underwriting reporting. Not the whole estate — one department, 512 dashboards owned, 133 migrated. What each phase produced and where it slipped.

Phase	Planned	Actual	What happened
Deploy and import	2 wks	2 wks	Three real workbooks imported. The fidelity report came back 41/52 sheets on the worst one.
Agree the definitions	3 wks	6 wks	13 definitions across 5 metrics. Two needed the CFO to decide, and one of those moved a team’s bonus basis.
Reconcile	4 wks	4 wks	10 dashboards cell-for-cell. Nine resolved in a day; one found two people seeing data they were not entitled to.
Move readers	3 wks	2 wks	486 readers, no training, no support tickets about how to use it.
Total	12 wks	14 wks	The overrun was entirely in one phase, and it was the right phase to overrun in.

Illustrative composite. The figures are internally consistent and every step is one the product supports today.

Two weeks over, and all of it in the definitions phase — which is the phase every plan underestimates and the only one where the overrun bought something. The 379 dashboards they did not migrate are why the reconciliation phase hit its estimate.

Weeks 1–2 — Prove it on your own worst workbook

Deploy into your cloud with your identity provider, then import real workbooks. Not the clean demo one — the oldest, the one with the most calculated fields, and the one the board sees.

You get a fidelity report before committing to anything: sheet by sheet, calculation by calculation, filter by filter. The ratio of converted to needs-review across three real workbooks predicts the estate better than any survey.

The import page after a Tableau workbook: four of four sheets converted with zero partial and zero manual, four of four calculations translated, one table of 600 rows and one dashboard created — **The number to argue about is the second one in each pair.** Sheets converted is the headline; sheets *partial* is the work. A real estate workbook will not come back 4/4, and the report saying so is the product working.

Who is in the room	For what	How long
A platform engineer	Deploy the container, the bucket and the database. Wire the identity provider.	2–3 days
The analyst who owns the worst workbook	Choose what to import, read the fidelity report, say whether the shortfalls matter.	2 days
Nobody else	Deliberately. This phase produces evidence, not consensus.	—

The deliverable of phase one is a number, not a demo

By the end of week two you should be able to say: “across three representative workbooks, N% of sheets converted cleanly and M calculations need a person.” That sentence is what the rest of the plan is built on, and it is the one thing a vendor cannot give you.

Weeks 3–6 — Agree the definitions

This is the phase that slips, the phase that matters, and the only one where the vendor cannot help much. The imported field names give you a first draft of the business model — considerably better than a blank page — and then three teams discover their definitions of net revenue disagree.

The model editor listing dimensions and metrics with their business names, types, number formats, descriptions and synonyms — **The artefact this phase produces.** Not dashboards — twenty metrics with a description, a format and a named owner. Everything afterwards is cheap; this is the expensive part and it was always outstanding.

The migration does not create the disagreement. It ends a decade of being able to avoid it, at a moment when you have budget and attention — which is the best time you will ever get.

Three rules that keep this phase from running away:

Settle the top ten, not all of them. A small number of metrics carry almost every decision. The long tail can drift another year without anyone noticing.
The people who own the numbers, not the people who own the tool. If this is run by the BI team alone, the definitions will be technically correct and commercially wrong, and finance will reject them in month four.
Keep a decision log. “Net revenue excludes intercompany, decided 14 March, owner: Group FC.” In eighteen months somebody will reopen it, and without the log they will reopen all of it.

Budget four weeks and expect six

Every migration we have seen underestimates this phase and none has regretted spending longer on it. If it finishes early, that usually means one team decided on behalf of the others and the argument has been deferred rather than settled.

Weeks 6–10 — Run both, and reconcile

The migrated dashboards run alongside the originals. Nothing is switched off. The only activity that matters is comparing numbers and working the review list.

Activity	What good looks like	Watch for
Reconcile against exported crosstabs	Pick the ten dashboards people actually open. Export the original as a crosstab, compare cell for cell.	Differences are usually a filter, not a calculation
Work the review list	Every sheet the report flagged gets a person and a verdict: fixed, accepted, or dropped.	“We will come back to it” means dropped
Set access rules on the data	Row and column policies on the datasources, not per dashboard. Test every surface with a restricted account.	At least one surface you had not thought of
Retire, do not migrate	Most estates have three to five times more dashboards than anyone opens. Use the usage logs.	Migrating everything is the most expensive mistake available

That last row deserves emphasis. The single biggest cost overrun in BI migration is treating the file list as the scope. Start from what people opened last quarter.

The access policies tab listing a row policy limiting west managers to the West region and a column policy hiding profit and profit ratio from sales reps — **Set these in week eight, not week thirteen.** Rules on the datasource are inherited by every dashboard built afterwards — which only helps if they exist before the dashboards do.

Weeks 10–13 — Move the readers first

Readers are the largest group and the easiest to move, because they only filter and read. They also account for most of your licence bill, so moving them first is where the commercial case actually lands.

THE COMMON SEQUENCEMove the authors first

The BI team goes first because they are keenest. Readers stay on the old estate, both run in full, and the licence saving arrives in month nine if at all.

THE BETTER SEQUENCEMove the readers first

Readers need no training — filter, drill, export, ask. The saving starts immediately, and authors follow as their own workbooks are migrated rather than all at once.

The old estate stays until its last-opened dashboard is retired. Plan to pay for both for a quarter. Any plan that does not is a plan to rush the reconciliation.

2 weeksto evidence, on your own workbooks

4–6 weeksto agree the definitions — the real work

1 quarterrunning both estates, budgeted honestly

Readers firstwhere the saving actually is

xychart-beta horizontal
    title "Weeks: planned against actual, by phase"
    x-axis ["Deploy", "Definitions", "Reconcile", "Readers"]
    y-axis "Weeks over or under plan" -1 --> 4
    bar [0, 3, 0, -1]

Every phase hit its estimate except one, and that one ran three weeks over. It is the only phase that is not a technical task, and the overrun is where the value was.

Where it stalls, and the tell

Stall	The tell	What to do
The definitions phase	Meetings that end with “let us take that offline” three times running.	Escalate to whoever owns the P&L. This is a business decision being treated as a technical one.
The review list	It has not shrunk in two weeks.	Assign each item an owner and a date, or accept it explicitly. An unowned list is a dropped list.
Reconciliation	“Close enough” appears in a status update.	Stop. Every unexplained difference is a trust problem that surfaces later, in public.
Nobody uses the new thing	Migrated dashboards, flat usage.	You migrated dashboards nobody opened. Go back to the usage logs.

What we would not promise

Thirteen weeks for a large estate. This is the sequence for a first meaningful tranche — a department, a few dozen dashboards. A thousand-workbook estate is several of these, run in parallel, over a year.
That the fidelity report will be all green. On an old workbook it will not be, and a vendor promising otherwise has not read your workbook.
That we can settle your definitions. We can surface the disagreements in week three instead of year five. Somebody in your organisation still has to decide.

What converts, artefact by artefact → · How to tell whether it worked →

← All posts Request a demo