Can your AI strategy drive real ROI?

The measured size of it

2-4 years: Typical AI use case ROI time
29%: Potential ROI boost via debt reduction

An executive team approves a multimillion-pound investment in generative AI to transform their customer service, only to discover a year later that they have merely replaced expensive human errors with faster, automated ones. They have high adoption rates and plenty of activity, but the balance sheet remains unchanged.

The gap between activity and transformation

Measuring AI success by the number of tools deployed is a mistake that creates a false sense of progress. Many organisations confuse the act of using AI with the result of business transformation, leading to a disconnect where leadership feels ahead while the operating model remains static.

In the insurance sector, this trend is particularly acute. According to a report by KPMG published by Reinsurance News, nearly half of executives believe they are in the top quartile for AI transformation, yet almost none have fully redesigned their core models, such as underwriting or sales, around the technology. The gap between activity and transformation is where most ROI is lost.

When AI is used only for routine automation or content generation, it delivers marginal efficiency. Real ROI requires a structural shift where the business is redesigned to run on AI, rather than simply using AI to perform old tasks.

Why AI ROI takes longer than traditional software

AI does not follow the standard payback period of seven to 12 months typical for technology investments. Because AI requires data cleaning, workforce reskilling, and workflow redesign, the returns are often delayed and non-linear.

Deloitte reported that most organisations find satisfactory ROI on a typical AI use case takes between two and four years. This delay happens because AI rarely delivers value in isolation; it is usually entangled with broader digital transformations or organisational restructuring, making it difficult to isolate the specific financial contribution of the AI itself.

There are two distinct timelines to manage:

Generative AI: Often provides quicker wins through productivity gains in content and coding, with some firms seeing returns within a year.
Agentic AI: These autonomous systems managing multi-step processes have higher complexity and longer timelines, with returns often taking three to five years.
Data Foundation: The time required to move from fragmented silos to a unified data layer often dictates the start date of actual value realisation.
Workforce Fluency: The gap between providing a tool and having a workforce proficient enough to use it for high-value output.
Governance Lag: The period spent establishing ethical guardrails and security protocols before a tool can be scaled beyond a pilot.

The components of a total cost of ownership

Many business cases fail because they treat AI as a software subscription rather than an operational overhaul. If you only account for the licence fee, your ROI calculation will be artificially inflated.

A pragmatic TCO must include the following costs:

Data Engineering: The cost of cleaning, structuring, and labelling fragmented legacy data.
Integration: The technical labour required to connect AI to CRMs, ERPs, and telephony systems.
Human Capital: The cost of training employees and managing the cultural resistance to automation.
Governance: The overhead of establishing security guardrails and ethical oversight.
Maintenance: The ongoing cost of monitoring for model drift and retraining as data patterns shift.
Technical Debt: The hidden cost of updating legacy infrastructure to support modern AI API calls.

IBM research indicates that addressing technical debt from legacy systems can improve AI ROI by up to 29% by reducing the friction of integration.

Distinguishing hard ROI from soft ROI

Financial stakeholders require hard numbers, but focusing exclusively on direct cost savings ignores the compounding strategic value of AI. A comprehensive strategy tracks both tangible financial gains and intangible operational improvements.

Hard ROI

These are direct impacts on the profit and loss statement.

Cost Reduction: Lowering the cost per interaction in a contact centre or reducing the hours spent on manual invoice processing.
Revenue Growth: Increasing conversion rates through AI-driven lead scoring or increasing average order value via personalisation.
Loss Avoidance: Reducing fraud losses through real-time anomaly detection.
Capacity Expansion: Handling a higher volume of transactions without increasing headcount.
Cycle Time Reduction: Shortening the time from lead generation to closed sale.

Soft ROI

These are indicators of long-term health that eventually convert to financial gain.

Decision Quality: Using predictive analytics to make more accurate forecasts, reducing waste and stockouts.
Employee Experience: Reducing burnout by automating drudge work, which lowers recruitment and turnover costs.
Customer Loyalty: Improving Net Promoter Scores (NPS) through faster, more accurate service, which increases customer lifetime value.
Brand Positioning: Establishing the company as a forward-thinking leader, attracting top-tier talent.
Risk Mitigation: Early detection of compliance failures before they result in regulatory fines.

How to calculate and validate the return

To move beyond vanity metrics like the number of prompts sent, you must establish a rigorous baseline before deployment. Without a before snapshot, you cannot prove the after improvement.

The standard formula for AI ROI is: AI ROI (%) = [(Total Quantified Value − Total AI Cost) ÷ Total AI Cost] × 100

To ensure this number is honest, use these validation methods:

Control Groups: Compare a team using AI against a team using the old process to isolate the AI impact.
Incrementality Testing: Run experiments to see if a conversion would have happened organically or if it was specifically driven by the AI.
Multi-touch Attribution: Instead of giving all credit to the last click, use models that recognise how AI influenced the customer at multiple stages of the journey.
A/B Testing: Deploy two different model configurations to determine which specific AI logic drives the higher financial return.
Holdout Groups: Maintain a small segment of the business that does not use the AI to measure the true lift in performance over time.

Moving from a pilot mindset to an AI-first model

The pilot trap occurs when a company runs dozens of small experiments that never scale. These pilots may prove a concept, but they do not deliver enterprise-wide ROI.

An AI-first company differs from a traditional company in its fundamental architecture.

Source: Worktual Decision Speed: Traditional is Days/Weeks (via reports); AI-First is Seconds/Minutes (via live AI) Scaling Model: Traditional is Add headcount to grow; AI-First is Scale via automation Data Usage: Traditional is Periodic analysis; AI-First is Continuous real-time processing Workflow: Traditional is Human-led, AI-assisted; AI-First is AI-led, Human-governed

Data from Worktual suggests that AI-first companies, particularly in contact centre environments, can see ROI exceeding 300% because they redesign the workflow so AI is the primary operator, with humans stepping in only for high-judgment moments.

Sources

Source: Insurance sector faces major gap between AI confidence and meaningful business transformation: KPMG, Reinsurance News