↗
When things go right
Migrations that delivered measurable results.
8×more streaming members after the move
Netflix: seven years to an all-cloud platform
After a database corruption in August 2008 halted DVD shipments for three days, Netflix rebuilt its systems as cloud-native services on AWS instead of lifting and shifting them. It shut its last data-centre workloads in January 2016. By then it had eight times the streaming members of 2008, viewing had grown by three orders of magnitude, availability was approaching 99.99%, and cloud cost per streaming start was a fraction of the data-centre cost.
Lesson: Re-architect rather than copy, and migrate in stages you can roll back.
100+ PBof data moved, 4 data centres closed
Spotify: 2,000 services to Google Cloud
Starting in 2015, around 100 teams in four regions moved about 2,000 services, 20,000 daily data-pipeline runs and more than 100 petabytes of data to Google Cloud. All user traffic ran on GCP by May 2017 and the four on-premise data centres were retired by 2018, freeing engineers to work on product instead of infrastructure.
Lesson: Give every team a clear migration path, tooling and a hard deadline.
3 mo → minto build a development environment
Capital One: first US bank all-in on the cloud
Over an eight-year programme begun in 2012, Capital One closed all eight of its on-premise data centres and moved to AWS, with about 2,000 applications running in the cloud. Building a development environment went from three months to minutes, disaster-recovery testing improved by 70%, and transaction errors and incident resolution fell by 50%.
Lesson: In a regulated industry, invest in controls and skills before you move the data.
$74.6Moperating costs saved in two years
Dropbox: migrating off the public cloud
Dropbox moved about 90% of its users’ data from AWS to its own storage infrastructure, keeping AWS for the rest. Its 2018 IPO filing credited the project with $74.6 million in operating-cost savings over the following two years. It shows that the right target platform depends on scale and workload, not fashion.
Lesson: Model the total cost of each target platform before you choose one.
Read the full article: When things go right
↘
When things go left
Implementations that cost far more than planned.
£48.65Min regulatory fines, plus £32.7M redress
TSB: a bank migration that locked out customers
In April 2018 TSB migrated its customer data to a new banking platform. All of its branches and a significant share of its 5.2 million customers were hit by problems with online, mobile, telephone and branch banking, and normal service took until December 2018 to restore. UK regulators fined TSB £48.65 million in 2022 for poor planning, governance and risk management, on top of £32.7 million already paid to customers.
Lesson: Rehearse the cutover and test for real production volumes before go-live.
≈ €500Mreportedly written off after 7 years
Lidl: an SAP inventory project cancelled
Lidl began planning its SAP-based eLWIS inventory system in 2011 and stopped it in July 2018, returning to its legacy system. Lidl values stock at purchase price while standard SAP retail uses sales price, and the heavy customisation this required drove up cost and hurt performance. Spend was reported at around €500 million (an estimate not confirmed by Lidl or SAP).
Lesson: Settle core process differences before configuring the software.
$64Mof net sales it could not ship
Revlon: an SAP go-live that stopped shipments
After Revlon switched its North Carolina manufacturing plant to a new SAP ERP system in 2018, disruption left it unable to ship about $64 million of net sales to US retail customers. Its share price fell 6.4% when it disclosed the problems, and investors filed a class-action lawsuit in 2019.
Lesson: Plan go-live around production and supply-chain risk, with a fallback ready.
£25M → £100Mbudget versus expected cost
Birmingham City Council: an Oracle ERP overrun
Birmingham’s Oracle Fusion finance and HR programme, started in 2018, was first budgeted at £25.3 million and revised to £38.7 million in 2021. By May 2023 it was expected to cost £80–100 million, while projected savings fell from £26.9 million to about £10.9 million. Reviews cited delays, cost overruns, weak controls and a product poorly suited to local government.
Lesson: Check that the product fits your sector before you sign, and keep governance independent.
Read the full article: When things go left
