Just finished a proof-of-concept with Consensus. The marketing says "intelligence platform," but after three weeks of trying to build actual insights, I'm calling it: it's a well-packaged data warehouse with a nice UI. It ingests, stores, and queries. That's it.
The "intelligence" part implies some layer of analysis, prediction, or automated insight generation. Consensus doesn't do that. You pipe in your data (logs, metrics, whatever), you write SQL or use their visual builder to query it. That's a warehouse. Their differentiator is the pre-built connectors and the opinionated schema they force your data into.
Where this falls apart:
* **No anomaly detection out-of-the-box.** I had to manually craft baseline queries and set thresholds. That's just monitoring 101.
* **Correlation is manual.** Seeing a spike in errors and a drop in conversions? You're the one joining those tables, not the platform.
* The "recommendations" tab was just query suggestions based on table size. Not intelligent.
If you need a centralized place to query your telemetry data and you don't want to manage a ClickHouse cluster, it's fine. But don't expect it to tell you *why* your p99 latency doubled. You'll be digging through dashboards you built yourself.
Example: To even get a simple "unusual error rate" alert, I had to write:
```sql
SELECT
service_name,
COUNT(*) as error_count,
COUNT(*) / LAG(COUNT(*), 1) OVER (ORDER BY time_bucket) as rate_change
FROM consensus.standardized_events
WHERE level = 'error'
GROUP BY service_name, time_bucket
HAVING rate_change > 2.0;
```
That's a warehouse query. An "intelligence platform" would surface that rate_change automatically and ask if I wanted to alert on it.
-shift
shift left or go home
You've hit on a crucial distinction that's become increasingly blurred in vendor marketing. Your experience with Consensus mirrors my team's evaluation last year. The "intelligence" label is being applied to any system that can store and retrieve time-series data with a chart on top.
Your point about anomaly detection is key. A true intelligence platform would embed statistical baselining (think Holt-Winters, STDDEV clustering) into its query execution layer, surfacing deviations without manual query crafting. I reviewed a POC where we pushed a year's worth of application metrics into Consensus; the platform identified no anomalies until we explicitly built the models ourselves in SQL. That's a query engine, not an analytical one.
The forced, opinionated schema you mention is actually the core of their business model - it lets them optimize storage and guarantee query performance, but it inherently limits the types of intelligence you can derive. If your data doesn't fit their model of "events," you can't easily perform network analysis or trace exemplar identification. You're right to call it a managed warehouse. For our use case, we stayed with a vanilla data lake and used separate, specialized layers for actual machine intelligence, which proved more cost-effective long-term.
No free lunch in cloud.