Status: Public Preview (Import and Direct Lake) — behaviour may change.

What it does

APPROXIMATEDISTINCTCOUNT(<column>) returns an estimated count of unique values. It's a faster, lighter alternative to DISTINCTCOUNT when the column has very many values and an exact figure isn't needed.

Key facts

• Support in Import and Direct Lake storage modes is in preview. The function itself has worked against some DirectQuery sources for longer.
• Import and Direct Lake (without fallback) use an engine-side HyperLogLog algorithm, with an error rate of roughly 1.6%.
• In DirectQuery, and in Direct Lake that has fallen back to DirectQuery, it pushes a corresponding aggregation to the source. Listed sources are Azure SQL, Azure Synapse dedicated SQL pool, BigQuery, Databricks and Snowflake.
• It takes a column reference only, not an expression, but the column can be any data type.
• It works in measures, calculated columns and calculated tables. It doesn't apply to visual calculations.
• No rows to count returns BLANK.

When to use / skip

Use it for unique-visitor or unique-order style measures over tens of millions of distinct values, where a 1.6% wobble is invisible to the audience. Skip it for anything that gets reconciled to finance, audited, or compared with another system's exact figure; use DISTINCTCOUNT there.

Configuration decisions

• Which measures can honestly tolerate an estimate, and whether the report labels them as approximate.
• Whether to keep an exact twin measure for reconciliation.
• Whether the model is staying in Import, or heading to Direct Lake or a DirectQuery source where the behaviour differs.

Gotchas

• On low-cardinality columns it can be slower than DISTINCTCOUNT and use more memory, especially with many group-by groups or high concurrency. It isn't a free upgrade.
• The estimate isn't additive in the way people hope, so subtotals and totals may not reconcile visibly.
• Direct Lake fallback silently changes the algorithm to the source's own, so the error profile can differ by day.

Consultant notes

• The first question from any finance user will be "why doesn't this match the ledger". Settle that in design, not UAT.
• Benchmark on the real column. The win only shows up at genuinely high cardinality.
• Preview in Import means don't hang a certified, headline KPI on it yet.

---

Revisit when Import and Direct Lake support reaches general availability.