AI Benchmarking

AI Benchmarking

BCG Expand's AI benchmark compares what financial institutions spend on AI, and how far they have taken it, against a like-for-like peer group. It gives leaders a yardstick for AI budgets, vendor terms, adoption and governance.

Explore our solutions

The Opportunity

  • Set AI budgets against a size-adjusted peer yardstick, not internal history
  • See the full AI cost curve across seats, APIs, provisioned capacity and self-hosted compute
  • Recover paid-for capacity, from never-activated seats to idle provisioned throughput
  • Negotiate vendor terms and concentration with peer reference points

How we can help

Spend and intensity

  • AI spend as % of technology budget and revenue

  • Spend per FTE and per active user

  • Licensing, API and infrastructure mix

  • Monthly run-rate against peers

    • Capacity and models

  • Seat utilisation and never-activated seats

  • PTU and GPU utilisation

  • Closed-weight, open-weight and in-house share

  • Effective cost per 1M tokens by hosting mode

    • Vendors and adoption

  • Top-vendor and top-3 concentration

  • Discount and commitment drawdown

  • Spend by function, rank and line of business

  • Agentic share of consumption

    • Leaders are scaling AI without a peer yardstick

    10 maturity dimensions benchmarked like for like, from strategy to governance
    5 spend lenses: intensity, capacity, models, vendors and adoption
    60+ of the world’s top financial institutions work with Expand
    80+ Expand benchmarks live today, built over 25 years

    How it works

    1. Scope

    Agree the perimeter, data period and currency. We share the data request, glossary and value-chain definitions.

    2. Extract

    Provide system exports from gateways, telemetry, admin consoles, ERP and HR systems, using pseudonymised identifiers, plus a short survey.

    3. Normalise

    Expand analysts map every submission to a common taxonomy and value chain. You verify the result before it is final.

    4. Compare

    Receive your report against a size-adjusted expected spend and peer cohort, followed by a results discussion and peer roundtable.

    Submissions are held confidentially by BCG Expand and used solely for this benchmark. Results are shared only in aggregated, anonymised form.

    The Opportunity

    • Compare AI use cases with peers on one glossary, so a pilot means the same thing everywhere
    • Separate measured AI benefit from modelled claims before it reaches the budget
    • Check that controls keep pace as more AI runs without a human in the loop
    • Test whether talent and operating model will hold as AI scales
    • See whether AI access is turning into daily use; where adoption is concentrated by functions, seniority, and location

    How we can help

    Use cases and return

  • Live use cases by value-chain area and AI type

  • Growth from 2025 to 2026

  • Top 10 use cases by annualised benefit

  • How benefit is evidenced

    • Autonomy and controls

  • Unattended share of AI spend

  • Human review by business area, today and in 12 months

  • Review depth for AI-generated code

  • Controls in practice against policy

    • Talent and operating model

  • Operating model and AI leadership

  • Access, active and daily-use funnel

  • Training hours by population

  • Expected shift in organisation shape

    • Autonomy and oversight

  • Unattended and agentic share of spend

  • Human-in-the-loop posture by use case

  • Review depth for AI-generated code

    • Governance controls

  • Test coverage, staged rollout and red-teaming

  • Kill-switch, audit log and model failover

  • Policy against practice

    • Talent and operating model

  • AI headcount and developer mix

  • Training hours by population

  • Organization model and AI leadership

    • Leaders are scaling AI without a peer yardstick

    10 maturity dimensions benchmarked like for like, from strategy to governance BCG AI Radar 2026
    5 spend lenses: intensity, capacity, models, vendors and adoption
    80+ Expand benchmarks live today, built over 25 years
    60+ of the world's top financial institutions work with Expand

    How it works

    1. Scope

    Agree the perimeter, data period and currency. We share the data request, glossary and value-chain definitions.

    2. Extract

    Provide system exports from gateways, telemetry, admin consoles, ERP and HR systems, using pseudonymized identifiers, plus a short survey for what no system records.

    3. Normalize

    Expand analysts map every submission to a common taxonomy and value chain. You verify the result before it is final.

    4. Compare

    Receive your report against a size-adjusted expected spend and peer cohort, followed by a results discussion and peer roundtable.

    Submissions are held confidentially by BCG Expand and used solely for this benchmark. Results are shared only in aggregated, anonymised form.

    Contact Us

    Product - Contact Us
    Reset

    How can we assist you today?

    Learn more about our core areas of expertise by selecting your topic of interest.

    Back to Top