← Build, buy and runFree decision tool · Build, buy and run

Lakehouse, Mesh, or Fabric?

Seven inputs covering timeline, budget, organisational autonomy, and how your data is used. A weighted algorithm returns a clear recommendation before you commit months to the wrong architecture.

Listen to this briefing

Pick your Lakehouse, Mesh, or Fabric

0:00

How the scoring works

Each of the seven inputs pushes one or more of the three paths up or down by a fixed weight. Set all seven, read the recommendation and the risk flags, then move the inputs you are least sure of to see whether the call holds.

How the recommendation is scored

What each input weighs in the lakehouse, mesh and fabric scores

InputLakehouseMeshFabric
Timeline urgency↓ ×1.5↑ ×1↑ ×2.5
Budget available↑ ×2.2↓ ×0.8↓ ×1.8
Organisational autonomy↓ ×1.3↑ ×2.5↑ ×0.5
Analytical workload↑ ×2.2↑ ×0.5↓ ×1
Live query latency sensitivity↑ ×1not usednot used
Distinct data silosnot used↑ ×2.2↑ ×0.8
Legacy migration tolerance↑ ×1.8↓ ×1↓ ×2.2
A higher answer raises this option's scoreA lower answer raises this option's score

Each path's weighted total is scaled to a score out of 100 and the highest score wins. When the top two scores are within 8 points, the tool returns the named hybrid path for that pair instead.

Weights as used by this tool's scoring model. Each answer is placed on a 0 to 1 scale before weighting, so a neutral 3 sits at the midpoint.

Your situation

0 of 7 inputs set. Drag or tap each slider, including any you would leave at 3.

Not set

How quickly do you need this data foundation ready for your first AI use case?

Not set

What's your implementation budget for this data foundation?

Not set

How independently do your business units already manage their own data?

Not set

How much of your AI roadmap depends on heavy cross-system analytics, batch reporting, or model training?

Not set

How much does your AI use case depend on fast, real-time answers at query time?

Not set

How many separate systems, departments, or business units hold data you need connected?

Not set

How willing are you to migrate data out of existing legacy systems into a new store?

Your recommendation

Answer to start. Set any input and the three scores appear here. The recommendation firms up once all seven are set.

The three paths

Centralised Lakehouse

When it applies

You have budget and runway for a 12–18 month build, your AI roadmap leans heavily on cross-system analytics or model training, and you'd rather control everything from one place than negotiate ownership across business units.

Typical profile
  • Timeline: 12–18 months to production
  • Cost: Highest upfront investment of the three paths
  • Risk: Can become a new bottleneck if one team controls all access
  • Upside: Fastest for big, cross-system analytical questions once built

Data Mesh

When it applies

Business units already operate with real autonomy, you have many distinct data silos rather than a few big ones, and you're prepared to invest in the cultural shift of treating data as a product each team owns.

Typical profile
  • Timeline: 9–15 months to production
  • Cost: Moderate infrastructure spend, real organisational investment
  • Risk: Harder to govern consistently across many independent teams
  • Upside: Scales well; the people closest to the data keep it accurate

Data Fabric

When it applies

You need something working fast, budget is limited, and disturbing legacy systems is not an option. A virtual metadata catalog connects existing systems in place without moving the underlying data.

Typical profile
  • Timeline: 3–6 months to a working data layer
  • Cost: Lowest upfront spend of the three paths
  • Risk: Can be slower than a Lakehouse when AI needs to search data live
  • Upside: Fastest to deploy; barely disturbs old systems
Terence Kok
Before You Go

This one exists because I kept seeing the same expensive mistake: a company commits to a 18-month Lakehouse build when a lighter-weight Data Fabric would have gotten them to their first AI use case in a fraction of the time, or the reverse, rushing into a virtual catalog that can't handle the heavy analytics they needed. Seven sliders won't replace a proper architecture review, but they will stop you from committing months of engineering time on a hunch.

Terence Kok

Last updated 29 Sept 2026

More in Build, buy and run

Next in this group · Decision toolFine-Tune vs. Retrieval-Only ClassifierIs this dataset safe to train into a model, or should it stay behind retrieval?Classify your data →
← All 8 in Build, buy and run