AI ROI

AI ROI: Definition and 5 Measurement Dimensions

In short

AI ROI is the return on an AI investment, calculated as (Net Benefit / Total Cost) × 100, and measured across five dimensions: revenue gain, cost reduction, time savings, quality improvement, and risk reduction. The formula is standard; the dimensions are AI-specific.

AI ROI uses the same formula every CFO already knows: Net Benefit divided by Total Cost, expressed as a percentage. The maths is not new.

What is new is what goes into each side of the equation. AI investment has a different cost structure than software licensing, and AI value lands across more dimensions than traditional automation.

AI ROI is measured across five dimensions. Each one must be defined, baselined, and tracked separately, then aggregated into the net-benefit figure.

1. Revenue gain

Incremental revenue attributable to the AI investment — new customers, higher conversion, larger basket size, or expanded market reach.

2. Cost reduction

Direct cost taken out — manual processing replaced, vendor contracts retired, or headcount reallocated. The easiest dimension to defend in a finance review.

3. Time savings

Hours freed per role, monetised at fully-loaded cost. Time savings are the most common AI benefit — and routinely the most under-measured.

4. Quality improvement

Measurable lift in accuracy, consistency, first-contact resolution, or defect rate. Translates to lower rework cost and higher customer retention.

5. Risk reduction

Reduced exposure to compliance, security, or operational risk — typically expressed as probability-weighted avoided cost. Required for regulated industries.

The 5 Cost Components of an AI Investment

In short

Total cost in an AI ROI calculation has five components: (1) model and API costs, (2) infrastructure, (3) talent, (4) data preparation, and (5) governance. Most CFOs underestimate data preparation and governance — together they often exceed model costs.

Underestimating total cost is the fastest way to overstate AI ROI. Each cost component must be itemised and tracked through the full measurement horizon.

1. Model and API costs

Foundation-model API charges, fine-tuning costs, and any per-token or per-call usage fees. The most visible cost — and rarely the largest one.

Forecast usage at production volume, not pilot volume. The two often differ by 10-100x and break ROI calculations built on pilot data.

2. Infrastructure

Compute, storage, vector databases, observability, and integration layers. Shared infrastructure reduces per-use-case cost at scale — but only if governance is in place.

3. Talent

AI engineering, ML operations, product, and business-side ownership. Fully-loaded cost, including overhead and benefits — not just salary.

Talent is typically 40-60% of total cost over a 24-month horizon. Under-counting talent is the single biggest source of inflated ROI claims.

4. Data preparation

Data collection, labelling, cleaning, governance, and ongoing maintenance. The cost most CFOs underestimate — and the one that makes or breaks production performance.

5. Governance and compliance

Model documentation, EU AI Act conformity work, audit trails, incident response, and board-level oversight. Now a regulated cost line, not an optional overhead.

The 5 Benefit Components: Where AI Value Lands

In short

Net benefit in an AI ROI calculation aggregates five components: (1) revenue gain, (2) cost reduction, (3) time savings monetised at fully-loaded cost, (4) quality improvement translated to financial impact, and (5) risk reduction expressed as probability-weighted avoided cost.

The benefit side of AI ROI is where most measurement breaks down. Each component must be defined in financial terms before the pilot starts, not retrofitted afterwards.

1. Revenue gain

Attributed incremental revenue — new pipeline, higher conversion, larger basket, or expanded segments. Attribution model must be agreed with finance up front.

2. Cost reduction

Direct cost taken out of the run-rate. The Ljusgårda case is a textbook example: 2.5M SEK/yr in annual savings, an 83% cost reduction on the targeted process.

3. Time savings (monetised)

Hours freed per role × fully-loaded hourly cost. The Alice Labs public-sector case freed 6,400-8,000 caseworker hours per year — a material benefit when monetised.

Time savings only count when reallocated to higher-value work or removed from the cost base. Hours "freed" that stay in the cost base are not ROI.

4. Quality improvement

Measurable lift in accuracy, consistency, or customer outcomes, translated to financial impact via reduced rework, higher retention, or lower escalation cost.

5. Risk reduction

Probability-weighted reduction in exposure — compliance fines, security incidents, operational outages. Often the largest benefit in regulated sectors.

The 74% Value Gap: Why Measurement Matters

In short

BCG x MIT Sloan Management Review (2024) found only ~26% of GenAI investments deliver measurable business value. The 74% gap is largely a measurement gap — RAND (2024) identifies a missing measurable KPI as a top root cause of AI project failure.

BCG and MIT Sloan Management Review (2024) found that around 26% of GenAI investments deliver measurable business value. The other 74% are stuck.

That number is widely misread as "74% of AI projects fail to make money". The more accurate reading is: 74% of AI projects fail to measure whether they made money.

RAND's 2024 report (RR-A2680-1) identifies missing measurable KPIs as a top root cause of AI project failure. Without a baseline and a defined success metric, ROI is not unprofitable — it is unprovable.

McKinsey's 2024 State of AI puts adoption at 72% of organisations. The 46-point gap between adoption (72%) and measured value (26%) is a measurement debt, not a technology debt.

The implication for CFOs is direct. Before approving the next AI investment, demand three artefacts: a baselined KPI, a defined attribution model, and a measurement cadence. Without all three, the investment is unmeasurable by design.

Source: Alice Labs AI ROI Framework (2026)

Real Alice Labs ROI Cases: Concrete Proof

In short

Three Alice Labs cases demonstrate AI ROI across different value dimensions: Ljusgårda (2.5M SEK/yr savings, 83% cost reduction), a Nordic media client (+2,092% click increase, 8.77M impressions), and a public-sector deployment (6,400-8,000 hours/year freed). Each was measured against a baselined KPI from day one.

Industry averages are not a substitute for measured outcomes. Below are three Alice Labs cases where AI ROI was defined before the pilot and tracked through production.

Case 1 — Ljusgårda (cost reduction)

Annual savings of 2.5M SEK against the targeted process, an 83% cost reduction. The ROI dimension here is direct cost-out: manual work replaced, run-rate dropped.

Baseline was set in the Diagnosis phase. The financial case was agreed with finance before any model was selected, which is why the savings hold up under audit.

Case 2 — Nordic media (revenue gain)

AI-driven search and content optimisation generated a +2,092% click increase and 8.77M impressions. ROI here is revenue-driven — attributed pipeline, not cost savings.

The attribution model was agreed up front: incremental clicks, downstream conversion, and audience growth, each tracked separately and aggregated quarterly.

Case 3 — Public sector (time savings)

A public-sector deployment freed 6,400-8,000 caseworker hours per year. The hours were reallocated to higher-value casework — not left in the cost base.

Time savings were monetised at fully-loaded hourly cost. The capacity unlock also reduced backlog, which is a quality benefit captured separately.

What these cases have in common

Three different value dimensions — cost, revenue, time — and one shared discipline: the KPI was defined before the pilot, baselined, and measured through production.

That discipline is what closes the 74% value gap. It is also what the Alice Labs Implementation Index 2026 reports as the largest single driver of the 96% production rate and the 14-week median pilot-to-prod cycle.