AI to ROI
Nine in ten firms use AI. About one in twenty is paid for it.
Adoption is near-universal; value is concentrated in a small group that fixed its data, redesigned the work, and tracked value in rand against a baseline. Every part of that sits within the finance office’s reach.
The AI value gap.
The evidence is consistent across every major study. The firms that get paid fix the data first, redesign the work end to end, put the effort into people and process, and track value monthly against a baseline.
Use AI in at least one function
McKinsey, 2025
88%
Report any EBIT impact from AI
McKinsey, 2026
37%
Get 5% or more of EBIT from AI
McKinsey, 2026
6%
Achieve AI value at scale
BCG, 2025
5%
AI projects that delivered the expected ROI
IBM, 2025
25%
Firms that abandoned most AI initiatives
S&P Global, 2025
42%
All figures reported. McKinsey, The state of AI (5 Nov 2025; 25 Aug 2026). BCG, The Widening AI Value Gap (Sep 2025) and Where’s the Value in AI? (Oct 2024). IBM Institute for Business Value, CEO Study (6 May 2025). S&P Global Market Intelligence via CIO Dive (14 Mar 2025). Gartner (26 Feb 2025).
Three hard lessons, turned into design principles.
Misunderstanding
Value is measured first
Every use case starts with a rand target and a baseline, and is traced through the Capital-to-EV Linkage Matrix™ before it is built.
Overspending
Start small, prove, then scale
One or two modules on real data, gated by agreed success measures, then the next. Each module costs less than the last because the data, security and model layers are shared.
Hype fatigue
Evidence grades on everything
Counted, proven, derived or modelled on every figure, and blanks shown as blanks. Language models explain numbers; they never produce them.
Every capital, one traceable chain: from AI intervention to enterprise value.
The Capital-to-EV Linkage Matrix™ follows the six capitals a board reports on under King IV through six steps. Read left to right, it is an engineering diagram. Read right to left, it is the board’s question: which intervention moves this multiple?
Step 1
AI intervention
The HC-to-EV™ bridge on the staff-cost line; WorkforceIQ™ skills genome, redeployment, span and layer engineering
Step 2
Spend & throughput
Staff cost per unit of output down, revenue per FTE up
Step 3
Operating metric
Productivity per FTE; regretted attrition
Step 4
Board KPI
Cost-to-income; organisational health
Step 5
Income statement
Staff costs, the largest controllable line
Step 6
EV & multiple
PPP to ROE to a P/B re-rate, the most direct wedge
C4, Human capital. Start from the intervention: the HC-to-EV™ bridge on the staff-cost line; WorkforceIQ™ skills genome, redeployment, span and layer engineering. Follow it through spend and throughput to the operating metric, the board KPI, the income statement and the multiple.
The Capital-to-EV Linkage Matrix™ is AlphaOmega proprietary methodology. Six-capital taxonomy per the IIRC Integrated Reporting Framework as adopted in King IV. Directional linkages for method; not investment advice. We engineer the controllable drivers; the market prices the rest.
Most AI projects fail on data, not algorithms. So the first ninety days are the data.
Know what you have, who owns it and how good it is, then connect it. Every module, agent and command centre after that is faster and cheaper, because it stands on one governed data model.
- AI projects without AI-ready data expected to be abandoned
- 60%AI projects without AI-ready data expected to be abandonedGartner, Feb 2025
- Firms that struggle to centralise their data
- 64%Firms that struggle to centralise their dataCisco, Oct 2025
- Of AI challenges that are people and process
- 70%Of AI challenges that are people and processBCG, Oct 2024
1. Inventory
Know what you have
Every data set the decisions need, where it lives and who uses it.
2. Ownership
Name an owner for each
A named owner per data set, with a definition the business agrees.
3. Quality
Measure it
A quality baseline with severity-graded findings, fixed at source rather than in spreadsheets.
4. Connect
One version of the truth
Pipelines from the top two or three systems into one governed model, reconciled to the ledger.
5. Govern
Guardrails and classification
Four-tier classification, access by role, POPIA checks, an audit log on every query.
6. Activate
Intelligence on top
Dashboards, agents, models and the command centre, all on the same foundation.
From the first assessment to an intelligence-led operating model by 2030.
Five stages, each with a measurable gate. The path is the same for a mid-market company and a sovereign-style fund; the depth and the number of modules change.
Q4 2026
Assess
Readiness and the evidence base
- An AI readiness assessment and maturity level
- A data inventory with owners and a quality baseline
- An acceptable-use policy; consumer AI brought inside a governed environment
- Two priority use cases with rand targets
The gate
Maturity level set, data owners named, policy approved
H1 2027
Foundations
Structure the data, connect the sources
- The core data model live
- Pipelines from the top systems, reconciled to the ledger
- The first module in production on your own data
- Usage and cost reporting for every AI call
The gate
One version of the truth for the board pack
H2 2027
Prove
Measured value on real data
- Two modules proving their success measures
- An AI command centre for the executive team
- Value tracked monthly in the management pack
- Evaluation suites before every release
The gate
Rand value against baseline, reported monthly
2028–29
Scale
A portfolio across functions
- Agents and modules across finance, risk, operations and people
- Your own domain model, fine-tuned in your tenancy
- Workflows redesigned, not just tooled
- Board oversight of AI risk as a standing item
The gate
AI value in the budget cycle
2030
Lead
An intelligence-led operating model
- AI part of most key decisions, with the evidence chain attached
- Data managed as an asset with quality KPIs
- Six capitals tracked to enterprise value
- Continuous test-and-learn
The gate
AI-led maturity; value traced to the multiple
Agree the baseline before the first build.
Every rand you spend on AI is then tracked against it, monthly, from the first module onward.