Leia Intelligence

The AI Work Bank

A library of what AI actually does — and what it takes to get right.

We keep our client work confidential. The case studies below are public examples, selected because they’re representative of the problems we help organisations solve — each annotated with our own analysis of how we’d approach it.

Industry
Function
Showing 18 of 18 public examples
Data centresOperations & efficiency
Cutting data-centre cooling energy with machine-learning control
~40% less cooling energy — but only after the data foundation and safety envelope are in place.
Source · Google / DeepMind Read entry →
Data centresOperations & efficiency
Autonomous cooling agents for AI factories
RL agents pre-empt thermal spikes — ~25% less cooling energy when treated as an engineered rollout, not a drop-in.
Source · Phaidra (with NVIDIA, CoreWeave, Applied Digital) Read entry →
Energy & utilitiesOperations & efficiency
Balancing the grid with an AI energy platform
Orchestrating 2GW of distributed assets — the lesson is the unified data platform underneath, not the AI alone.
Source · Octopus Energy / Kraken Read entry →
Transport & infrastructureOperations & efficiency
Predictive maintenance for rail infrastructure
Cheap commodity sensors, expensive workflow — prove the false-positive rate on one asset class before scaling.
Source · New York City Subway (MTA) Read entry →
ConstructionOperations & efficiency
Automated construction progress tracking
Computer vision turns site walks into an objective source of truth — decision quality, not automation for its own sake.
Source · Buildots (clients incl. Intel, Sir Robert McAlpine, NCC) Read entry →
Financial servicesRisk, legal & compliance
Contract intelligence for commercial lending
360,000 lawyer-hours removed — automate the standardised majority, route the ambiguous tail to people.
Source · JPMorgan Chase (COiN) Read entry →
Financial servicesRisk, legal & compliance
Real-time payment fraud detection
Real-time fraud scoring at network scale — judged on the false-positive trade-off, not the catch rate alone.
Source · Mastercard (Decision Intelligence / Decision Intelligence Pro) Read entry →
Financial servicesKnowledge & enablement
A knowledge assistant for financial advisers
RAG over 100,000+ documents — the project is the governed knowledge base, not the model.
Source · Morgan Stanley (with OpenAI) Read entry →
InsuranceCustomer service & experience
Instant, narrow-scope insurance claims
Claims settled in seconds — automation works when scope is tight and the human hand-off is deliberate.
Source · Lemonade Read entry →
Financial servicesMarketing & content
AI-generated marketing copy
Measurable CTR uplift on generated copy — held to brand, compliance and tested uplift, not vendor headlines.
Source · JPMorgan Chase (with Persado) Read entry →
Consumer financeCustomer service & experience
AI customer service at scale, done right
AI support at scale, then a course-correction — scope to the simple tail and keep a person always reachable.
Source · Klarna (with OpenAI) Read entry →
Consumer goodsPeople & HR
AI in graduate recruitment
Faster, fairer early screening — but explainability isn't optional in regulated decisions about people.
Source · Unilever (with HireVue and Pymetrics) Read entry →
Professional servicesRisk, legal & compliance
Legal drafting and research assistance
Minimum viable trust — target high-effort, low-risk tasks, keep a lawyer in the loop, prove value before widening.
Source · Allen & Overy (now A&O Shearman), with Harvey Read entry →
HealthcareR&D & clinical
AI-assisted diagnosis from eye scans
Expert-level triage from eye scans — narrow, interpretable and uncertainty-aware, designed to defer to a human.
Source · DeepMind with Moorfields Eye Hospital Read entry →
PharmaceuticalsKnowledge & enablement
Scaling AI across the enterprise (an adoption model)
Not one tool but an adoption model — fix the data foundation, diagnose high-value use cases, upskill people.
Source · AstraZeneca Read entry →
RetailOperations & efficiency
Demand forecasting and inventory optimisation
Forecasting value comes from data integration and granularity — and 'adapts to shocks' beats any headline accuracy.
Source · Walmart Read entry →
RetailMarketing & content
Generative fashion imagery and digital twins
On-brand imagery in days, not weeks — a clean 'augment, don't replace' case with rights and disclosure guardrails.
Source · Zalando Read entry →
RetailCustomer service & experience
AR virtual try-on for online shopping
AR try-on that lifts conversion and cuts returns — strongest when scoped around inclusive accuracy from day one.
Source · Sephora (Virtual Artist, with ModiFace) Read entry →