TL;DR 60-second read

Keep reporting alive. Then pick what proves itself.

Dataland is reported to shut at the end of September. That leaves about 11 business days, too few to pick a tool and migrate. So continuity comes first: confirm the date, export and record the essential reports, time the key tasks, and name an owner and fallback.
Define the work before the tool. Watch Aideen do three real tasks (a search, a heavy report, an export) so "search" and "millions of rows" mean something measurable.
Try what the company already has in October. Databricks dashboards and Metabase on one governed data slice. Sigma comes in as a challenger only if they fail a real requirement.
Decide on evidence. Accuracy, freshness, permissions and complete exports are pass/fail. Only then score usability, speed, flexibility, admin and cost, and choose the simplest option that passes.

Advisory note Revised 15 Sep 2026

Replace theworkflow.

Keep the team's essential reporting running past month-end first. Then choose the simplest supported option that proves it can do their real tasks. Databricks is a reasonable starting point and Sigma a credible challenger, but the evidence doesn't yet justify prescribing either.

Aideen Teo

11 business days Dataland is reported to shut at the end of September. A ten-day pilot starting 16 September ends on the 29th, with no room left to migrate. So continuity and tool selection run as separate tracks.

Two tracks

Keep reporting running. Then choose.

Track 1 This week, before 30 Sep

Continuity

Confirm what actually stops, the exact date, and whether a read-only extension or archive window is possible.
Record the essentials: reports, filters, calculation rules and who needs access.
Export the raw datasets and report definitions. A static export only covers work where stale data is acceptable.
Capture a Dataland baseline (time, clicks, rows) on the key workflows while it still exists. Without it, "close enough to Dataland" can't be measured.
Name a technical owner and an interim path, such as downloads from a governed Databricks SQL warehouse or curated Metabase questions.

Track 2 October

Selection

Watch three real tasks end to end, so the requirements come from the work.
Test existing tools first: Databricks dashboards and Metabase as configured today.
Bring in Sigma as a targeted challenger if existing tools fail a meaningful requirement.
Gate on accuracy, freshness, permissions and exports before scoring convenience.
Switch only on reconciled evidence. Trial freely, buy nothing until then.

The unanswered question

What does "search" mean here?

Finding one order is a different job from finding a dashboard or asking a business question. If the day is mostly about pulling up individual records fast, a polished analytics dashboard could still fail as a replacement.

Find a recordA specific order, customer or transaction, in seconds

Find a reportThe right dataset or preset dashboard

Ask a questionAn aggregate cut: totals, trends, comparisons

"Millions of rows" is three numbers

Stored

How big the underlying dataset is

Shown

How many rows the screen needs to page through

Exported

How many rows actually leave in the download

Ask Aideen to demonstrate

Her most frequent search and filtering task.
The slowest or most consequential report.
An export, and what she does with the file afterwards. Reconciliation, combining other data, fixing categories or building a customer deliverable each point to a different solution.

Current state

Six tools. Diagnose before discarding.

As described in the discussion. Two of the existing tools may be failing because of how they're set up, not because of the product.

DatalandPrimary reporting and exploration

Shutting down

Fast search and filtering, preset dashboards, quick retrieval across millions of rows, easy download. Confirm the exact shutdown date and any extension.

MetabaseReporting and querying

Diagnose

Its graphical query builder handles filters, joins, grouping and custom columns without SQL. "Needs SQL" often means no curated models, restrictive permissions or unprepared data. Find out which.

TableauReporting and dashboards

Diagnose

Feels slow and tied to BizOps. Is that query performance, permissions, or a change process with too few creators? A new vendor won't fix an ownership bottleneck.

DatabricksCompany data and AI platform

Preferred, if checks pass

Already strategic, but that doesn't prove the team's data is there, fresh enough, modelled correctly and accessible to them. Check all four.

ClaudeAI assistant

Ad hoc route

Connected to Databricks, a credible route for one-off analysis by people comfortable checking its work. It doesn't replace a governed reporting layer.

MonolithUnderlying source data

Source

Every reporting tool depends on it, so governance and data modelling matter whichever front end wins.

The architecture principle

Separate the foundation from the experience.

The useful question is which front end best serves users on top of a governed foundation. Databricks is the preferred foundation for storage, compute, access control and shared definitions, subject to the readiness checks above.

User experience
Existing tools first (Databricks AI/BI dashboards, Metabase), Sigma as a challenger: search, filters, exploration, downloads.
Semantic & governance
Shared business definitions, trusted metrics, access policies, curated tables and views, ownership and lineage.
Data & compute
Databricks lakehouse and SQL warehouse, tuned for interactive queries with several users at once.
Source systems
Monolith and other operational sources, with pipelines into governed analytical models.

What people touchWhat everything stands on

Candidates, in order

Quickest viable route first.

First

Databricks AI/BI dashboards

Parameterised, filterable dashboards on a governed SQL warehouse. The team's stated needs (search, filter, page through rows, export) are a dashboard and warehouse-tuning problem, not a natural-language one.

Genie comes later, for genuine extra questions. Genie Agents take up to 50 tables, and Databricks recommends five or fewer. User usage is free until 31 January 2027, then metered.

Alongside

Metabase, on a curated table

Already paid for and familiar. Point its query builder at one curated Databricks table and try the demonstrated tasks.

The test: how many of the key workflows work in Metabase today? If most do, that's the October bridge and possibly the long-term answer.

Challenger

Sigma on Databricks

A spreadsheet-style interface over the warehouse. Bring it in if existing tools fail a meaningful requirement, or if a free parallel trial doesn't slow the continuity work.

Watch: access depends on how auth is set up (a shared service credential vs each user's own OAuth). Creator seats concentrated in BizOps just move the bottleneck. Vendr's median contract is $66,540 a year (range $17,500 to $147,142, 152 purchases), before warehouse compute and implementation.

Same governed data, same tasks, same ordinary user accounts

What not to do

Let tool selection eat the time needed to keep reporting running on 1 October.

Choose from dashboard screenshots or a vendor demo.

Assume a new vendor fixes a slow change process or a shortage of report builders.

Point natural-language querying at loosely defined metrics. Fluent answers can still disagree.

October, ten business days

A pilot with the deadline removed.

Depends on curated data, access, people and trial licences being in place. If any of those slip, the dates slip with them.

Days 1–2

Turn the demonstrated tasks into 5 to 8 test workflows

Cover record lookup, a high-row-count query, a multi-filter report, an ad hoc question and an export-heavy task. Compare against the Track 1 baseline.

Days 2–4

Prepare one governed slice in Databricks

One named owner, probably BizOps or the platform team, with a delivery date. Same data, permissions and definitions for every candidate. Tune the warehouse for interactive use.

Days 4–7

Build only the minimum in each candidate

The equivalent workflow in dashboards, Metabase and, if triggered, Sigma. No over-built dashboards.

Days 7–9

Test with ordinary users and accounts

Same tasks, short orientation, no coaching, several users at once. Record time, clicks, failures and whether they self-serve.

Day 10

Decision review

Gates first, then scores. Bring in the data platform team, security and procurement before any purchase.

Pilot scorecard

Pass the gates. Then score.

A tool that returns wrong numbers or shows data to the wrong people fails, however fast it is. These four are pass/fail.

Pass / fail

Accuracy

Record IDs, counts and totals reconcile against an agreed source of truth, not just the old dashboard.

Pass / fail

Freshness

Data updates as quickly as the work needs, compared with what Dataland delivered.

Pass / fail

Permissions

Access and row-level restrictions hold when tested with ordinary user accounts.

Pass / fail

Complete exports

The filtered result at the needed row count, untruncated, in a usable format. Check workspace download settings, and get security sign-off before customer data lands on laptops.

Then weight what's left

Self-service usability

30%

Can a normal user answer the question without SQL or BizOps help?

Speed under load

25%

Time to first result and filter latency on large datasets, with several users at once.

Exploration flexibility

20%

Change filters, groupings, calculations and views without rebuilding the report.

Administration

15%

Effort to build, maintain, secure and support, and who does it.

Total cost

10%

Licences, warehouse compute, implementation, and Genie usage once billing starts.

Decision rule

Only options that pass all four gates get scored. Among those, prefer the simplest supported stack that lets a typical user finish the tasks alone, at a speed that doesn't feel like a loss.

Sigma wins only if it clearly cuts time-to-answer or reliance on specialists by enough to justify its licence, implementation and support.

Who owns what

Aideen accepts or rejects the business workflow.
A named technical owner implements and supports it.
Data platform, security and procurement join before anything is bought.
A metric owner settles disputed numbers, with a date for the semantic layer.

Risk register

What could go wrong, and when.

Cost, timeline, data, security and people risks behind the recommendation, sorted by severity. Dates and prices checked against vendor sources on 15 September 2026.

High blocks continuity, correctness or security Medium raises cost or delays the decision

HighTimeline

Shutdown date unconfirmed

Dataland's end-of-September shutdown is reported, not confirmed, and no extension or archive window has been agreed.

When it bitesEnd of Sep 2026About 11 business days from 15 Sep.

MitigationConfirm the date and ask for a read-only extension this week. Export and baseline the essential reports before the cut-off.

OwnerAideen + technical owner

HighCost

Genie's free period ends

Genie One and Genie Agents usage by users is free until 31 January 2027, then metered at the full rate. Genie Code is already billed beyond a free allowance.

When it bites31 Jan 2027Costs start with no action needed.

MitigationSet Genie budgets and monitor usage from day one. Model February 2027 costs before the team depends on it.

OwnerData platform team

HighData

Databricks isn't ready for this team

Nobody has yet confirmed that the team's data is in Databricks, fresh enough, correctly modelled and accessible to them.

When it bitesBefore pilot day 2Blocks every candidate.

MitigationRun the four readiness checks (present, fresh, modelled, accessible) on the governed slice before building anything.

OwnerData platform / BizOps

HighSecurity

Exports capped or blocked

Workspace download settings may cap result size, and millions of rows of customer data on laptops needs security approval.

When it bitesFirst export testExport is the team's core workflow.

MitigationTest full-volume filtered exports early in each candidate. Get security and DLP sign-off before rollout.

OwnerSecurity + technical owner

MediumCost

Warehouse compute grows

Every front end, including Metabase and Sigma, runs its queries on the Databricks SQL warehouse. More users and heavier filters mean more compute.

When it bitesScale-up after pilotHidden inside the Databricks bill.

MitigationMeasure compute cost during the pilot with several users at once, and include it in each option's total cost.

OwnerData platform team

MediumCost

Sigma licence cost

Vendr's buyer data puts the median Sigma contract at $66,540 a year (range $17,500 to $147,142), before implementation and warehouse compute.

When it bitesIf Sigma is triggeredFigures move; recheck at quote.

MitigationGet a quote for the expected Creator and Viewer mix before any trial ends. Compare on total cost, not licence alone.

OwnerProcurement

MediumProduct

Names and limits keep changing

Genie Spaces were renamed Genie Agents in July 2026. Agents take up to 50 tables, and Databricks recommends five or fewer. Internal notes and training may already be out of date.

When it bitesAt decision dateVendor docs change monthly.

MitigationRe-check vendor documentation on the decision date. Design agents around a few curated tables, not the whole catalogue.

OwnerTechnical owner

MediumSecurity

Sigma access model

What each user sees depends on whether Sigma connects with a shared service credential or each user's own OAuth sign-in.

When it bitesSigma setupRow-level rules can silently loosen.

MitigationTest with ordinary user accounts. Prefer per-user authentication wherever row-level restrictions matter.

OwnerTechnical owner

MediumData

Metrics without an owner

Without agreed definitions, dashboards and natural-language answers can return different numbers for the same question.

When it bitesBefore pilot day 4Undermines trust in any tool.

MitigationName a metric owner with a delivery date for the semantic layer before the pilot builds start.

OwnerBizOps

R10

MediumPeople

The bottleneck moves

If report building and change requests stay concentrated in BizOps, a new tool reproduces the Tableau problem. Sigma Creator seats can do the same.

When it bitesLicence planningNot fixed by switching vendor.

MitigationAgree who can build and change reports, and how requests flow, before buying seats.

OwnerAideen + BizOps

R11

MediumTimeline

Pilot dependencies slip

The October pilot needs curated data, access, people and trial licences in place on time.

When it bitesOctober 2026Dates slip with any dependency.

MitigationKeep the continuity track independent of the pilot. Move pilot dates rather than squeezing the continuity work.

OwnerTechnical owner

Owners are suggested roles, to be confirmed with the teams involved.

Recommendations

Six moves, in order.

Protect continuity now

Confirm the shutdown, record the essential reports and rules, export and baseline them, and agree an extension or supported fallback with a technical owner.

Watch three real tasks

A search, a heavy report and an export, end to end, including what happens to the downloaded file.

Test existing tools first

Databricks dashboards and Metabase against those tasks. Diagnose Tableau's slowness before blaming the product.

Bring in Sigma on evidence

A targeted challenger on the same data and tasks, with warehouse compute and maintenance counted in its cost.

Give the semantic layer an owner and a date

Trusted metrics, joins and access need defining before the pilot's day 4, and before any conversational BI scales.

Switch only on reconciled evidence

All four gates passed, several users tested, workflow accepted by Aideen, support accepted by the technical owner.

Discussion summary and subsequent recommendations

Discussion summary

The team relies on Dataland for reporting on the Sendoso platform, and Dataland is due to be deprecated at the end of September.
Metabase and Tableau are also in use. Metabase was described as needing SQL for meaningful self-service; Tableau as slow and restricted because changes depend on BizOps.
The company already uses Databricks and Claude, but Databricks reporting isn't yet as usable as Dataland for this team.
The team values rapid search, filtering, preset dashboards, speed across millions of rows and easy downloads.

Subsequent recommendations

Prioritise continuity before the shutdown, as a separate track from tool selection.
Base requirements on demonstrated tasks, and test existing tools on a governed Databricks slice before adding a new vendor.
Evaluate Sigma as a challenger, and decide through pass/fail gates followed by a weighted score.

A summary of points raised, not a transcript. The recommendations were developed after the discussion and revised following review on 15 September 2026.

Bottom line

Secure the reporting.Then prove the tool.

Keep Aideen's essential reporting uninterrupted, then pick the simplest supported option that shows it can do her team's actual tasks: accurate, fresh, properly permissioned, with complete exports.