← Back to Stories

Stories · TMEGS

Plume cuts analytics costs 58% by moving off Databricks to an AWS lakehouse

Mactores and AWS Professional Services moved Plume’s analytics from Databricks to a data lakehouse on AWS, built on Amazon EMR on EKS, Delta Lake on Amazon S3, AWS Lake Formation and Amazon SageMaker Unified Studio.
  • Data Platform
  • TMEGS
  • Amazon EMR on EKS
  • Delta Lake
  • Lake Formation
  • SageMaker Unified Studio
plume-home-960 (1)

Workspaces moved

13

production

7

non-production

What Mactores delivered for Plume

  • The figure is an average taken across every workspace Plume moved
  • No workspace had to fall back to Databricks after its cutover.
  • Fourteen weeks for production, four for everything else.
  • There was no Mactores-run transition period after go live.

Before

$1,513K

projected six-month run-rate on Databricks

After

$635K

six-month run-rate on the AWS lakehouse

Saved

$878K

projected over six months

Six-month run-rate, drawn to scale
Before

$1,513K on Databricks
Now $635K Just under 45% of the track; the empty part is what Plume no longer spends. $635K on AWS

Each track spans $1,513K. Hover the After bar for the value.

The customer

About Plume

Plume provides a SaaS platform to broadband ISPs, and its data platform runs the analytics behind those operator deployments.
  • Headquarters Palo Alto, US West
  • Industry Telecom · ISP tech
  • Reach Hundreds of operator deployments, worldwide
  • Before the move 20 Databricks workspaces
Plume at a glance. Source: Plume

The challenge · 01

The problem

Plume’s Databricks footprint had grown to 20 workspaces, and the bill had grown with it. Two things could not be traded away to bring it down.
  • Correctness Every number had to survive the move A figure that changed in transit would surface in front of the ISPs Plume serves.
  • Operational readiness Someone had to run it on day one A new platform would only pay off if Plume’s own team could operate it the moment the workloads switched.

Three conditions on the move

Databricks cost optimization was the goal. These three conditions decided how it could be reached.
  • Migrate all 20 workspaces end to end, production and non-production alike.
  • Validate parity before cutover, to take the risk out of each switch.
  • Hand Plume AWS-native runbooks it could use from the first day after cutover.

 

How Mactores delivered · 02

An agent took the volume work; engineers made the calls

Mactores ran the move on its agent-native delivery model: the agent and the engineers took turns on every workspace until it was running on AWS.

Databricks 20 workspaces Plume’s whole estate at the start
  1. Agent

    Step 1, agent: Discovery

    Maps what each workspace holds before anything moves

  2. Engineers

    Step 2, engineers: Architecture

    Designs the target lakehouse on AWS

  3. Agent

    Step 3, agent: Conversion

    Rewrites workloads for Amazon EMR on EKS

  4. Agent

    Step 4, agent: Parity validation

    Compares results on AWS with Databricks

    Parity validation A workspace moved only after its outputs on AWS matched its outputs on Databricks.
  5. Parity gate No cutover without a match
  6. Engineers

    Step 5, engineers: Cutover

    Switches each workspace once parity passes

  7. Engineers

    Step 6, engineers: Hand-off

    Turns the platform over to Plume’s team

    Plume runs it from day one
Mactores’ agent-native delivery
Lands on Plume’s AWS lakehouse
  • Amazon EMR on EKS
  • Delta Lake on Amazon S3
  • AWS Lake Formation
  • Amazon SageMaker Unified Studio
How Plume’s 20 Databricks workspaces reached the AWS lakehouse. Numbers give the order of the steps, and colors show who owned each one. Hover the step marked Details for more.

The work

What runs Plume’s analytics now

This is Mactores’ Data Platform Modernization work in practice: Databricks out, AWS services in, each with one job.

Amazon EMR on EKS

Runs the converted analytics workloads on Kubernetes

Delta Lake on Amazon S3

Stores the tables in an open format

AWS Lake Formation

Controls who can read which data

Amazon SageMaker Unified Studio

Gives data teams one place to work

Delivery · 03

Production first, then the rest of the estate

AWS Professional Services and Mactores ran the program jointly, from 15 December 2025 to 17 April 2026.

  1. Kickoff

    Eight forward-deployed engineers and the migration agent begin the move

  2. Phase 1: production

    Thirteen workspaces that carry live analytics move first

  3. Phase 2: non-production

    Seven development and test workspaces follow on the proven path

  4. Go-live

    The last workspace goes live on the AWS lakehouse

Time and scope, drawn to scale
Weeks
14 in phase 1 and 4 in phase 2, of 18
Workspaces
13 in phase 1 and 7 in phase 2, of 20

In production · 04

The outcome

Plume’s engineers have operated Amazon EMR themselves since the day after cutover, working from runbooks built for AWS.
All 20 workspaces, live on AWS

20 of 20 · 0 rolled back

What we ship · Data Platform Modernization

Moving your own data platform to AWS?

Mactores plans and runs data platform moves on AWS, from the target architecture to the day your team takes over.

Know more about Mactores' data platform expertise

Curious about the migration agent behind this move? See Mactores’ AI agent expertise.