diff --git a/README.md b/README.md
index 54ad50a..979f3bb 100644
--- a/README.md
+++ b/README.md
@@ -147,6 +147,11 @@ cloudcost policy run cloudcost.yml
 5. View your optimization findings:
 ```bash
 cloudcost findings list
+```
+
+   To see which resources to fix first, roll findings up per resource (sorted by combined impact; `--min-findings 2` shows only resources hit by more than one check):
+```bash
+cloudcost findings summary
 ```
 
 ## Mapping a real cloud source to the FOCUS schema
diff --git a/src/cloudcost/cli.py b/src/cloudcost/cli.py
index 037bb5e..f194abc 100644
--- a/src/cloudcost/cli.py
+++ b/src/cloudcost/cli.py
@@ -141,6 +141,60 @@ def list_findings(severity: str = typer.Option(None, help="Filter by severity"))
     except FileNotFoundError:
         console.print("[yellow]No findings generated yet. Run `cloudcost policy run` first.[/yellow]")
 
+@findings_app.command("summary")
+def summarize_findings(
+    input: str = typer.Option("data/findings.json", help="Path to findings JSON file"),
+    min_findings: int = typer.Option(1, help="Only show resources with at least this many findings"),
+):
+    """Roll up findings per resource, sorted by combined impact."""
+    from rich.table import Table
+
+    try:
+        with open(input, "r", encoding="utf-8") as f:
+            findings = json.load(f)
+    except FileNotFoundError:
+        console.print("[yellow]No findings generated yet. Run `cloudcost policy run` first.[/yellow]")
+        return
+
+    # A resource can trip several policies (e.g. stopped-but-billing AND
+    # untagged). Each finding stays independent in findings.json; this is
+    # just a rolled-up view for deciding which resource to fix first.
+    groups = {}
+    for f in findings:
+        resource_id = f.get("resource_id")
+        if not resource_id:
+            continue
+        group = groups.setdefault(resource_id, {"policies": [], "impact": 0.0})
+        group["policies"].append(f.get("policy_name", "unknown"))
+        group["impact"] += float(f.get("estimated_impact", 0.0) or 0.0)
+
+    rows = sorted(
+        ((rid, g) for rid, g in groups.items() if len(g["policies"]) >= min_findings),
+        key=lambda item: item[1]["impact"],
+        reverse=True,
+    )
+
+    if not rows:
+        console.print("[green]No findings to display![/green]")
+        return
+
+    table = Table(title=f"Findings by resource ({len(rows)} resources)")
+    table.add_column("Resource", overflow="fold")
+    table.add_column("Findings", justify="right")
+    table.add_column("Policies", overflow="fold")
+    table.add_column("Combined impact", justify="right")
+    for resource_id, g in rows:
+        table.add_row(
+            resource_id,
+            str(len(g["policies"])),
+            ", ".join(sorted(g["policies"])),
+            f"${g['impact']:,.2f}",
+        )
+    console.print(table)
+
+    total_impact = sum(g["impact"] for _, g in rows)
+    console.print(f"[blue]Total combined impact: ${total_impact:,.2f}[/blue]")
+
 @findings_app.command("notify")
 def notify_findings(
     webhook_url: str = typer.Option(..., help="Slack incoming webhook or generic webhook URL"),
diff --git a/tests/test_findings_summary.py b/tests/test_findings_summary.py
new file mode 100644
index 0000000..ca4e194
--- /dev/null
+++ b/tests/test_findings_summary.py
@@ -0,0 +1,58 @@
+import json
+
+from typer.testing import CliRunner
+
+from cloudcost.cli import app
+
+
+runner = CliRunner()
+
+
+def _write_findings(tmp_path):
+    findings_path = tmp_path / "findings.json"
+    findings_path.write_text(
+        json.dumps(
+            [
+                {"resource_id": "/sub/vm-a", "policy_name": "stopped_not_deallocated_vms", "severity": "high", "estimated_impact": 100.0},
+                {"resource_id": "/sub/vm-a", "policy_name": "untagged_resources", "severity": "low", "estimated_impact": 0.0},
+                {"resource_id": "/sub/disk-b", "policy_name": "unattached_disks", "severity": "medium", "estimated_impact": 250.5},
+                {"resource_id": None, "policy_name": "github_wasted_actions_minutes", "severity": "low", "estimated_impact": 5.0},
+            ]
+        )
+    )
+    return findings_path
+
+
+def test_findings_summary_groups_by_resource_sorted_by_impact(tmp_path) -> None:
+    findings_path = _write_findings(tmp_path)
+
+    result = runner.invoke(app, ["findings", "summary", "--input", str(findings_path)])
+
+    assert result.exit_code == 0, result.stdout
+    assert "2 resources" in result.stdout
+    # Rich folds long cells at the test runner's 80-col width, so check the
+    # combined impact per resource rather than the full policy list.
+    assert "$100.00" in result.stdout
+    assert "$250.50" in result.stdout
+    # Highest combined impact first.
+    assert result.stdout.index("disk-b") < result.stdout.index("vm-a")
+    assert "$350.50" in result.stdout
+
+
+def test_findings_summary_min_findings_filters_single_hit_resources(tmp_path) -> None:
+    findings_path = _write_findings(tmp_path)
+
+    result = runner.invoke(app, ["findings", "summary", "--input", str(findings_path), "--min-findings", "2"])
+
+    assert result.exit_code == 0, result.stdout
+    assert "1 resources" in result.stdout
+    assert "vm-a" in result.stdout
+    assert "disk-b" not in result.stdout
+    assert "$100.00" in result.stdout
+
+
+def test_findings_summary_missing_file(tmp_path) -> None:
+    result = runner.invoke(app, ["findings", "summary", "--input", str(tmp_path / "nope.json")])
+
+    assert result.exit_code == 0
+    assert "No findings generated yet" in result.stdout