How AllianceBernstein Cut Snowflake Spend 25% and Built a Safety Net for Cortex AI Agents

By Perry Tapiero
June 25, 2026 | 5 min read

AllianceBernstein didn’t need another Snowflake dashboard. They had already tried that.

Actually, they had tried more than that. The team had evaluated two other Snowflake cost optimization tools before Yuki. The promises sounded good. The results didn’t hold up. Month after month, their internal assessments showed no real impact. As Albert Obeng put it, the tools felt “more of a gimmick” than a real solution.

That matters because AllianceBernstein was not trying to optimize a small analytics playground.

They were running a modern Snowflake data platform supporting analytics, transformations, Sigma, Looker, dbt, Fivetran, and a growing portfolio of internal Cortex AI agents pushing signals to investment teams.

The stakes were getting bigger.

More workloads. More AI use cases. More dynamic query patterns. More pressure to keep Snowflake costs predictable without slowing down the business.

With Yuki, AllianceBernstein cut Snowflake credits by 25.8% on average over six months, routed more than two-thirds of 4.3M queries to smaller warehouses automatically, and reduced peak queue time from over 17 minutes to under 2.5 minutes.

This is how they did it, and why the bigger win was not just savings.

It was confidence.

About AllianceBernstein

AllianceBernstein is a global asset management firm providing investment management and research services to institutional, high-net-worth, and retail investors worldwide.

Within that environment, the team manages complex, data-intensive investment strategies across credit, distressed, and private opportunities. Their Snowflake-based data platform powers analytics, transformations, and internal AI agents built on Snowflake Cortex that work together to push signals to investment teams.

That makes Snowflake more than a warehouse.

It is part of the operating layer for investment intelligence.

And when that layer scales, cost control cannot be an afterthought.

Data Stack

  • Snowflake
  • dbt, Sigma, Looker
  • Cortex AI Agents
  • Fivetran

The Challenge: Snowflake Gives You Power, but Power Needs Guardrails

Snowflake gives data teams a lot of room to move. That is the whole point.

You can scale compute. Add workloads. Build transformations. Serve dashboards. Support AI agents. Push more use cases into production without constantly fighting infrastructure.

But that flexibility comes with a very real tradeoff:

The easier it is to spin up compute, the easier it is to burn credits before anyone notices.

For AllianceBernstein, three problems were showing up at the same time.

1. Optimization tools that didn’t actually optimize

The team had already tried two other Snowflake cost optimization tools.

The results did not match the pitch.

They were getting visibility, but not control. Reports, but not enough execution. Promises, but not sustained monthly impact.

That is the trap a lot of enterprise data teams fall into.

A dashboard can tell you where the fire started.

It cannot put the fire out.

And if your data team has to manually interpret, prioritize, and execute every recommendation, you do not have automation. You have another backlog.

AllianceBernstein needed a system that could take action inside the workload path, not just analyze spend after the fact.

2. No visibility until the damage was done

Without a layer actively making routing decisions, a runaway dbt model or misbehaving query could burn credits for days before anyone noticed.

That is not just a cost problem.

It is a control problem.

By the time a team sees the spike in a report, the money is already gone. Someone has to investigate. Someone has to find the workload. Someone has to decide whether it was legitimate or wasteful. Someone has to fix it.

That cycle is slow.

And Snowflake credits do not wait for a weekly review.

3. A growing AI footprint with no safety net

AllianceBernstein was also building Cortex-based agents across multiple investment domains, connecting them via REST endpoints into internal applications.

That changes the cost profile.

AI workloads are not always neat, scheduled, and predictable. Agents can trigger different query paths depending on the question, workflow, context, and downstream application.

Some queries are lightweight.

Some need more compute.

Some workloads grow quietly until they become expensive.

Without confidence in the cost layer, scaling AI on Snowflake starts to feel risky.

And that is the worst possible tradeoff: the business wants more AI, but the data team has to worry about what the next workload will do to the bill.

The Real Problem: Visibility Is Not a Safety Net

Most Snowflake cost optimization conversations start with visibility.

That makes sense. You need to know where spend is coming from.

But visibility is not the finish line.

AllianceBernstein had already learned this the hard way: knowing what happened after the fact is not the same as controlling what happens in production.

The real requirement was different.

They needed a layer that could:

  1. understand workload behavior
  2. route queries to the right warehouse
  3. reduce overprovisioning automatically
  4. alert the team before runaway workloads became expensive
  5. support AI growth without forcing manual tuning

That is the difference between a dashboard and a safety net.

A dashboard says, “Here is what happened.”

A safety net says, “We caught it before it became a problem.”

The Solution: A Real Layer in the Path of Every Query

After evaluating alternatives, AllianceBernstein chose Yuki because the architecture made sense.

Yuki sits in the path of every query and automatically routes each one to the right-sized warehouse based on workload behavior.

That was the key difference.

This was not another reporting layer sitting on the side.

It was not another recommendation engine asking the team to go do the work.

It was an execution layer.

Albert and the team understood how Yuki worked before they turned it on, and that transparency mattered. After two tools that failed to deliver meaningful impact, the team needed to trust the mechanism, not just the message.

Yuki gave them that.

“We tried two other tools before Yuki. They were more gimmick than substance. What resonated with Yuki was the architecture: a real layer in the path of every query, routing it to the right warehouse. Now we have the peace of mind to keep building.”
Albert Obeng, Head of Technology @ AllianceBernstein

That line is the story.

Peace of mind does not come from another chart.

It comes from knowing the system is actively controlling the work.

Start Small, Prove It, Then Scale

AllianceBernstein did not flip a switch across everything on day one.

That is important.

Enterprise teams do not need more “trust us” deployments. They need proof.

So the team rolled Yuki out on lower-risk workloads first, validated the platform, then expanded it across their heavier hitters: Sigma, transformation warehouses, and beyond.

That rollout pattern is exactly how Snowflake optimization should work in complex environments.

You do not start with a massive change management project.

You start where the risk is controlled.

Then you prove the savings.

Then you expand.

That made adoption practical, measurable, and easier to trust.

Built-In Early Warning for FinOps

The cost savings mattered.

But one of the most important outcomes was something AllianceBernstein did not have before: live signal when a workload was running out of hand.

That is a different kind of value.

About a week before the case study call, one workload started running out of hand. The team spotted it through Yuki and fixed it immediately. Without Yuki, Albert described it as “a door left open with money flying out the window.”

That is the kind of problem traditional cost reporting misses.

Not because the reporting is bad.

Because the timing is wrong.

If the alert comes after the credits are burned, it is not a warning. It is an invoice explanation.

Yuki helped AllianceBernstein catch the issue while there was still time to act.

For a dedicated FinOps person, that is a huge unlock. The team also worked closely with Yuki to customize the views they needed, giving FinOps a clearer operating layer instead of forcing them to reverse-engineer spend after the fact.

Results: 25.8% Average Monthly Credit Savings Over Six Months

From the first month Yuki was active, AllianceBernstein saw consistent credit savings.

Across six months of production usage, from October 2025 to March 2026, monthly savings ranged from 22% to 30%, averaging 25.8%.

That consistency matters.

One good month can be noise.

Six months of sustained savings is a pattern.

And for a Snowflake environment supporting analytics, transformations, BI, and AI agent workloads, sustained savings are more valuable than a one-time cleanup.

Because the goal is not to “find savings once.”

The goal is to keep the platform efficient as the business keeps adding work.

Routing: More Than Two-Thirds of 4.3M Queries Sent to Smaller Warehouses

Out of 4.3M queries routed through Yuki, more than two-thirds were automatically sent to a smaller warehouse.

Zero manual tuning required.

That number tells you why the savings were consistent.

Most teams overprovision because they do not know which queries actually need larger compute and which ones can run safely on smaller warehouses.

So they size for the scary workloads.

Yuki changes that pattern by making the decision per query.

If a query does not need a larger warehouse, it should not pay for one.

That sounds obvious.

But without automation, it is almost impossible to execute at scale.

Performance: Peak Queue Time Dropped From 17 Minutes to Under 2.5 Minutes

The story is not just lower credits.

Performance improved too.

Peak queue time fell from over 17 minutes to under 2.5 minutes, while average queue time dropped from 40 seconds to 20 seconds.

That is the part many teams still get wrong.

They assume cost reduction means performance risk.

So they overprovision.

But overprovisioning is not the same as performance engineering. It is just the most expensive way to avoid hard decisions.

The better move is routing.

Separate the work. Send each query to the compute it actually needs. Keep heavier workloads from crowding out lighter ones. Reduce waste without creating contention.

That is how AllianceBernstein got both: lower spend and better queue behavior.

Cost Control Became Confidence to Keep Building on Snowflake

This is the part that makes the case study bigger than a savings story.

AllianceBernstein is building AI agents on Snowflake Cortex. Those agents push signals into investment workflows, which means the data platform is becoming more dynamic, more strategic, and more exposed to unpredictable workload patterns.

Before Yuki, scaling that footprint came with risk.

After Yuki, the team had a safety net.

With cost and routing under control, AllianceBernstein now has more confidence to keep expanding AI agent use cases on Cortex without constantly watching the meter run. The savings created headroom, but the bigger value was knowing the platform could absorb growth with better control.

That is where cost optimization becomes business enablement.

Not “we saved money.”

More like: “we can keep building.”

As Yuki CEO Ido Arieli Noga put it:

“AllianceBernstein didn’t want another dashboard or another promise. They wanted a layer they could actually understand and trust in production. That’s what Yuki delivers: per-query routing decisions that show up as real credit reduction every month, with the confidence to keep building on Snowflake.”
Ido Arieli Noga, CEO and Co-Founder @ Yuki

What This Unlocked for AllianceBernstein

Most cost stories stop at the savings number.

That is not the full story here.

The 25.8% average savings matter. The 4.3M routed queries matter. The queue-time reduction matters.

But the bigger unlock is operational confidence.

For AllianceBernstein, Yuki helped unlock:

  • More confidence to scale Cortex AI agents without creating a runaway Snowflake bill
  • Less reliance on manual inspection and after-the-fact cost reviews
  • Better FinOps signal when workloads start behaving unexpectedly
  • More efficient warehouse usage across high-volume workloads like Sigma and transformations
  • A clearer path to expand Snowflake usage responsibly

That is the real advantage.

When teams trust the cost layer, they move faster.

They stop treating every new workload like a budget risk.

They stop waiting for the bill to find out something went wrong.

They stop forcing engineers to become full-time cost police.

And they start building with guardrails already in place.

Takeaways

1. Dashboards are useful, but they do not execute

Visibility matters, but it is not enough. AllianceBernstein had already tried tools that did not produce real impact. What changed was adding a layer that could act inside the query path.

2. Per-query routing beats static warehouse assumptions

Not every query needs the same compute. Yuki routed more than two-thirds of 4.3M queries to smaller warehouses automatically, which helped reduce waste without manual tuning.

3. AI agents need cost guardrails before they scale

Cortex AI agents introduce dynamic workloads. Without a safety net, teams may hesitate to expand AI use cases. With routing and live workload signal, scaling becomes less risky.

4. FinOps needs early warning, not invoice forensics

Catching a runaway workload while it is happening is very different from explaining it after credits are gone. Yuki gave AllianceBernstein signal early enough to act.

5. Savings are good. Confidence compounds.

A 25.8% average reduction is meaningful. But the bigger win is knowing the platform can keep growing without turning every new workload into a cost-control project.

Want to See What You Could Save?

If your Snowflake environment is growing across dbt, BI, transformations, and AI agents, dashboards alone will not keep costs under control.

You need a layer that acts.

Yuki helps teams route each query to the right warehouse in real time, reduce overprovisioning, catch runaway workloads earlier, and keep Snowflake performance predictable without manual tuning.

Download the full AllianceBernstein case study

By Perry Tapiero
Perry Tapiero leads marketing at Yuki, driving demand generation and brand growth for B2B and B2C SaaS companies in FinTech, AdTech, and Cybersecurity. With 15+ years of experience, he specializes in go-to-market strategies, ICP refinement, and managing multi-million-dollar campaigns using HubSpot and Salesforce. Previously at other companies, he led ABM, PBM, and product marketing initiatives that drove ARR growth and helped achieve Gartner Magic Quadrant recognition. Perry was a regular contributor for marketers and now shares his insights on LinkedIn.

Table of Contents

Free cost analysis

Take 5 minutes to learn how much money you can save on your Snowflake account.

By clicking Submit you’re confirming that you agree with our Terms and Conditions.

Follow us on LinkedIn

Related posts

Free cost analysis

Take 5 minutes to learn how much money you can save on your Snowflake account.

By clicking Submit you’re confirming that you agree with our Terms and Conditions.

Skip to content