If you're delivering embedded analytics through Sisense, your dashboard is only as good as the data behind it. But how often do you think about what's happening upstream in the pipelines and ingestion architectures that determine whether that data is fresh, reliable, and cost-efficient before it ever reaches a visualization?
QBeeQ is known in the Sisense ecosystem as a plugin and integration provider, but that's only part of what we do. This post is a window into the broader, strategic side of our work: architectural decision-making that happens well upstream of the BI layer but directly impacts the analytics experience your customers see.
The client here uses Sisense to deliver embedded dashboards. We built their Snowflake-based data platform from the ground up: integrations, ELT, data modeling, and governance with Sisense sitting at the top of the stack. The next question became: is our data replication approach from MS SQL Server into Snowflake the right one long-term, or should we move to Snowflake Openflow?
A Platform Already in Motion
Our relationship with our client didn't start with this POC. We previously helped them design and build a Snowflake-based data platform from the ground up, including data integrations, ELT processes, data models, and governance. That work gave them a solid, modern foundation for their analytics and reporting, including the embedded dashboards they deliver to their customers via Sisense.
This POC was the next strategic step in that journey. With the platform stable and maturing, the question became: are we using the best possible approach for replicating data from their MS SQL Server into Snowflake? And more specifically, should we continue developing our custom ingestion approach, or move toward a more ready-to-go solution using Snowflake Openflow?
Our client’s business depends on near-real-time data availability. Their operational data lives in MS SQL Server and feeds directly into Snowflake, where it powers analytics and reporting for their customers across retail, telecom, banking, and energy.
These aren't just technical concerns. They directly affect the quality and reliability of the analytics experience delivered to its end customers. The key business drivers behind this evaluation were:
Faster availability of incremental data, especially for large transactional tables
Lower operational risk and improved robustness
Reduced maintenance effort over time
Cost optimization at scale
Putting Two Approaches to the Test
Rather than making a decision based on vendor documentation or assumptions, we designed a structured POC to test both approaches in parallel, on the same three tables and similar data volumes, across four dimensions:
performance
cost
stability
maintainability
Option 1: Custom ADF + SQL CDC (Our Existing Solution)
The key strength of this approach is control. We define the schema, manage data types, handle transformations during ingestion, and have full visibility into every step of the pipeline. This is the approach we built as part of the initial platform work:
CDC (Change Data Capture) - a native SQL Server feature that tracks only changed rows (inserts, updates, and deletes), eliminating the need to reload entire tables on every run
Azure Data Factory (ADF) - Microsoft's cloud orchestration service, used here to move delta data from SQL Server to Azure Blob Storage
Snowflake - which then loads and transforms the data using Snowpipe and scheduled tasks
It's more engineering-intensive to set up, but the result is a robust, predictable, and highly tunable architecture.
Option 2: Snowflake Openflow
Openflow is Snowflake's newer, more plug-and-play approach to data ingestion. It uses Change Tracking (CT) on the SQL Server side, a lighter-weight mechanism than CDC, and ingests data more directly into Snowflake with less custom orchestration required.
On paper, it's appealing: faster to configure, fewer moving parts, and tighter integration within the Snowflake ecosystem. But as with any "managed" solution, there are trade-offs. And that's exactly what we needed to quantify.
What We Found
Performance
Both approaches handled initial loads equally well. The gap appeared with delta loads — Openflow's Change Tracking completed incremental updates in 1–2 minutes versus 5–12 minutes for ADF + CDC. For near-real-time use cases, that's a genuine advantage worth noting.
Cost
This is where the picture shifts dramatically, and where decision makers should pay close attention:
Openflow consumed approximately 12.5 Snowflake credits per day, translating to roughly $50/day or ~$1,500/month for this workload alone.
ADF + CDC came in at an estimated $8–15/month
That's not a marginal difference. That's a 100x cost gap on a three-table POC.
Now consider what that looks like at scale. Most production environments don't replicate three tables; they replicate dozens, sometimes hundreds. If costs scale proportionally, an Openflow-based architecture could run into tens of thousands of dollars per month for a full production workload, compared to what remains a very modest cost with the custom ADF approach.
For a data platform that's meant to be a long-term foundation, the compounding effect of that cost difference is enormous. Over a single year, the gap between these two approaches could easily reach $200,000 or more — money that could instead fund additional data products, analytics capabilities, or engineering capacity. For decision makers evaluating build vs. buy, or assessing the true TCO of a "managed" solution, this kind of analysis is exactly what's needed before committing to a direction.
Stability & Maturity
The Openflow SQL Server connector was still in preview status at the time of the POC. Runtime and canvas updates caused instability during our testing, and the solution requires ongoing infrastructure coordination.
Azure Data Factory, by contrast, is a mature and battle-tested technology with a proven track record in production environments.
Architecture & Maintainability
Openflow introduced meaningful architectural complexity:
requires Snowflake Business Critical edition (for PrivateLink connectivity)
adds authentication and data sharing overhead
offers limited control over target data types, and
handles soft deletes only, so hard deletes require additional handling
Additional Snowflake-side transformations (PARSE_JSON, CLEAN_JSON_DUPLICATES) were also necessary.
Our custom ADF solution, while more involved to build:
gives full control over schema and transformations
handles deletes cleanly
can be extended through a reusable automation procedure that simplifies onboarding new tables over time
Open Flow
ADF + CDC
Architecture Complexity
⚠️ Requires Business Critical + PrivateLink
✅ Simpler, fewer dependencies
Stability
⚠️ Preview — instability during POC
✅ Mature, production-proven
Cost/month
⚠️ ~$1,500
✅ ~$8–15
Initial load
✅ Comparable
✅ Comparable
Delta load speed
✅ ~1–2 min
⚠️ ~5–12 min
Onboarding new tables
✅ Fast and easy
⚠️ More setup, but automatable
Schema & type control
⚠️ Limited
✅ Full control
Delete Handling
⚠️ Soft deletes only
✅ Full delete support
Our Recommendation
The POC gave our client something genuinely valuable: a data-driven basis for a strategic architectural decision, rather than a choice made on assumption or vendor marketing. Our recommendation leaned toward continuing with the custom ADF + CDC approach as the long-term foundation. Not because Openflow lacks merit, but because the cost differential is substantial, the stability risks are real at this stage of the product's maturity, and the architectural overhead introduces complexity that isn't justified by the performance gain alone.
That said, the delta load performance advantage of Openflow is meaningful. As the connector matures and if near-real-time requirements intensify, it remains a viable path to revisit. The POC ensures that if and when that moment comes, the decision will be informed, not reactive.
What This Kind of Work Looks Like in Practice
In the Sisense ecosystem, you may know us through our plugins and integrations. But this post reflects something equally important: our ability to act as a full-stack data platform partner. What your users see in Sisense is a function of decisions made far upstream. When those decisions lack rigor, the effects show up as latency, data gaps, and rising costs.
Running a structured POC like this is often underestimated, but is one of the most valuable things we do for our clients. It's not just about finding the faster or cheaper option. It's about understanding the full picture: performance under realistic conditions, true cost at scale, architectural implications, and long-term maintainability.
Every organization's situation is different. The right ingestion architecture depends on your data volumes, latency requirements, existing tooling, team capabilities, and cost constraints.
If you're thinking about your broader data strategy, not just what Sisense can do, but whether everything underneath it is built right, that's a conversation we're well-positioned to have.
Thinking about your data platform strategy? Let's talk .
QBeeQ is data strategy consulting firm made up of former Sisense employees and customers. We are a Sisense Gold Implementation Partner and Snowflake Select Partner
