Finance in 2026: AI goes practical as cost and cyber risks intensify
Finance leaders are shifting from AI hype to practical deployment. Budgets are tight, scrutiny on cyber risk is rising, and boards expect clear returns on every tool, process, and control.
The theme is simple: automate the repetitive, protect the balance sheet, and keep judgement human. The firms that win will build systems people trust, not machines they fear.
AI in finance: from fear to implementation
Hugh Scantlebury, CEO at Aqilla, sees teams moving from anxiety to execution. The question isn't "should we use AI?" - it's "how do we use it responsibly, predictably, and with oversight?"
- Automate high-volume data tasks to speed up reporting and reduce manual entry.
- Establish guardrails: audit trails, approvals, and clear ownership over AI outputs.
- Measure reliability and bias; don't scale a process you don't fully understand.
- Keep the model simple where possible. People-led automation beats machine-led accounting.
Customers aren't asking for fully machine-run finance. They want clarity, confidence, and tools that remove busywork so analysts can focus on interpretation and decisions.
Prompt literacy becomes essential
Charis Thomas, Chief Product Officer at Aqilla, expects "prompt literacy" to become a basic job skill. Like learning to search properly years ago, finance teams now need to ask better questions and validate outputs.
- Teach teams to structure prompts, provide context, and ask follow-ups.
- Treat AI + search like a loop: verify sources, keep traceability, document assumptions.
LLMs are great at giving the next logical answer. Discovery and verification still matter - especially when the numbers move decisions.
If you're building this capability, see practical prompt training resources and examples here: Prompt course library.
Tax: agentic AI moves into controlled workflows
Russell Gammon, Chief Innovation Officer at Tax Systems, expects agent-based AI to show up in real tax workflows - carefully, with controls. Guidance from regulators like HMRC is likely, which should increase confidence.
- Start low-risk: data prep, number checks, reconciliations, and summarisation.
- Use agents to support first- and second-level reviews: flag anomalies, highlight risks, draft summaries.
- Build trust gradually, then expand scope as accuracy and controls prove out.
By late 2026, expect scale: more efficient reviews and faster cycles across tax functions.
Cost pressure forces simplification
Bruce Martin, CEO of Tax Systems, notes UK businesses face persistent cost pressure going into 2026. Inflation may be easing, but prices and employment costs are still rising, which drags on hiring and investment.
- Reset cluttered processes. Remove steps added over the years that don't add value.
- Equip teams with modern tools - stop asking people to do more on outdated systems.
- Chase new capability, not just speed. The biggest wins come from insights you couldn't produce before.
In a tight year, streamlined operations and effective tools separate the cautious from the confident.
Market infrastructure: uptime, latency, security - or you're out
Terry Storrar, Managing Director at Leaseweb UK, says finance and trading workloads have zero tolerance for disruption. Data shifts in milliseconds, and clients expect performance and security by default.
- Non-negotiables: uptime SLAs, low latency, strong security controls, and rapid scale.
- Test AI/ML only if it proves a real productivity edge - no trend-following.
- Choose providers that can show real metrics, not promises.
Cyber risk is financial risk: budget for RoM
2025 made the link painfully clear. A supply-chain breach at a real-estate finance vendor forced major banks to scramble, while UK Finance flagged significant losses to AI-driven investment scams in the first half of the year. For context on sector trends, see UK Finance Fraud Facts.
The average cost of a financial services breach is now estimated at around $5.56 million per incident - before litigation, fines, and reputational damage. IBM's latest report is a useful benchmark: IBM Cost of a Data Breach.
- Treat cybersecurity spend as protection for earnings, capital, and customer trust.
- Adopt a return on mitigation (RoM) mindset: what risk, in dollars, did we actually remove?
- Use external research communities and offensive testing to find issues before attackers do.
- Test AI assets and third-party attack surfaces as thoroughly as core products.
AI will scale activity on both sides. Institutions that don't test their AI and vendor exposure as rigorously as they ship features will keep ending up in headlines for the wrong reasons.
What finance leaders should do now
- Pick two AI use cases to scale this quarter: one for data automation, one for review support. Add guardrails and audit trails.
- Standardise prompt practices and training across FP&A, reporting, and tax.
- Pilot agentic AI in tax on low-risk tasks; expand only after measured accuracy lifts.
- Run a process reset: remove non-essential steps and retire manual workarounds.
- Review infrastructure partners for latency, uptime, and security. Demand proof.
- Shift cyber budgeting to RoM. Prioritise offensive testing of AI and third parties.
Looking for practical tools to cut busywork in your team? Start here: AI tools for finance.
Your membership also unlocks:
