Lots of organisations are using AI; far fewer can say what it’s worth. The Bank of England and FCA’s 2024 survey found 75% of UK financial services firms already use AI. UK government research on AI adoption points to skills and understanding as common barriers. The question is no longer “should we try AI?” but “which uses are worth keeping?”
Step 1: Measure the baseline first
Before switching anything on, record for two weeks: how long the task takes, how often it happens, how many people do it, and the error or rework rate. Without a baseline, any ROI number is a guess.
Step 2: Count all the costs
- Subscriptions and usage fees (many tools charge per seat, per task or per resolution).
- Set-up time: configuration, testing, writing prompts and training people.
- Ongoing review time. If a person checks every output, that time is a cost.
Step 3: Be strict about what counts as a saving
Time saved is only a financial return if it is redeployed (to revenue-generating work, backlog or quality) or avoids a cost (overtime, a new hire, a contractor). Twenty minutes saved per person scattered across a week often disappears.
Step 4: Watch quality, not just speed
Track errors, complaints, rework and customer satisfaction alongside time. A faster process that creates more rework hasn’t paid back.
Step 5: Decide at a fixed date
Set a review date (say 60 or 90 days) and a threshold in advance: keep, change or stop. This is how you avoid pilots that never end.
Sources
- Bank of England and FCA, "Artificial intelligence in UK financial services – 2024" Regulator survey
- GOV.UK, AI adoption research Government research
- Gartner press release, 25 June 2025: over 40% of agentic AI projects predicted to be cancelled by end of 2027 Analyst prediction