The pattern repeats in every diversified group
I have walked into the same conversation in seven different countries this year. A diversified holding — apparel and food and energy and real estate, six or eight subsidiaries — wants to “do AI.” Each subsidiary CEO has already signed something. The Chief of Staff at the holding has a spreadsheet, but it is incomplete. The General Counsel is uneasy and cannot articulate why. The CFO sees the line items but not the picture.
Here is what is actually happening: the holding is running an undocumented AI estate. Nobody designed it. It accumulated.
The four invoices nobody adds up
Most holdings track the obvious vendor invoice. They miss four others:
| Cost | Where it hides | Typical magnitude |
|---|---|---|
| Capability overlap | Two BUs paying for similar features | ~38% of vendor spend |
| Compliance multiplication | Each vendor audited separately | $15K–$60K per BU per year |
| Onboarding cycles | Internal time across BU teams | ~120 person-hours per vendor |
| Lock-in & switching | Data trapped in vendor format | Hard to value until it bites |
Add these four to the visible invoice and the consolidated AI line is rarely 5–10% over plan, as the spreadsheet suggests. It is typically 30–60% over.
What governance failure looks like in practice
I sat with the General Counsel of a regional retail group last quarter. She knew three things: the apparel BU had signed with one vendor, the home-goods BU with another, and the cosmetics BU with a third. She did not know:
- Which jurisdictions hosted each vendor’s data centres
- Whether any of the three had signed Data Processing Agreements compatible with her group’s privacy posture
- Which of the three would notify her of a breach within 24 hours versus 72
- Whether any vendor was using customer prompts to train shared models
This is not a theoretical risk. It is the position most diversified holdings are in right now. The General Counsel is the last person to learn that AI exists in her organisation, and the first person who will be sued when something goes wrong.
The argument for one platform with isolated vaults
The honest version of the consolidation argument is not “save money” — though you do. It is governance becomes possible.
- One platform contract at the holding level, with master DPA and incident-response terms.
- Isolated vaults per subsidiary — each BU's data, prompts, and fine-tunes stay in their own encrypted partition. No cross-contamination.
- Per-BU brand voice & customisation — Pandora does not sound like Billabong; the helpdesk for the energy subsidiary does not sound like the one for the F&B chain.
- Holding-level governance console — General Counsel sees usage, prompt logs, incidents, and compliance status across every BU from one screen.
- Group volume pricing — one invoice, one renewal cycle.
This is not technically novel. What makes it hard is organisational: getting BU CEOs to surrender procurement autonomy in exchange for a better deal. The argument that lands is not “the holding will save 35%.” It is: “your subsidiary will get better tools, faster, with the holding’s governance carrying the risk.”
The window is short
Most holdings I work with are 18–24 months from this becoming a forced consolidation. The trigger is usually one of three: a regulator asks a question nobody can answer, a breach at one BU exposes the holding’s lack of central oversight, or a CFO finally adds up the four hidden invoices.
You can run that consolidation under pressure, after the trigger event. Or you can design it now, while there is still time to do it well.
There is no third option.
Our Perspectives
The procurement data is starting to look ugly. In Gartner's 2025 enterprise survey, 41% of multi-business-unit companies reported using six or more AI vendors across the group, with no central inventory of who-uses-what. Average overlap between vendor capabilities: 38%. That means more than a third of what each subsidiary pays for is duplicated somewhere else in the holding. Add the cost of seven separate compliance audits, seven onboarding cycles, and seven security reviews, and the consolidated number is rarely 5-10% over plan — it's typically 30-60% over. The data argues for a single procurement function with a master agreement and BU-level workspaces. That's not glamorous, but it is the difference between AI-as-investment and AI-as-leakage.
Let me name what's actually happening. Each BU CEO got pitched by a slick vendor, signed a fast deal to look modern in front of the board, and now nobody at the holding level can answer four basic questions: how much are we spending on AI in total, where is our customer data sitting, what are we exposed to legally if any of these vendors gets breached, and what happens to operations when one of them gets acquired or pivots? In practice, the holding is running an unmanaged, undocumented AI estate — and the General Counsel is the last to know. This isn't a strategy problem. It's a basic governance failure dressed up as innovation. Until someone at the holding level owns the AI estate the way they own the insurance program, the bill will keep growing and the risk will keep being invisible.
Here's the upside hiding inside this mess: a holding that consolidates AI across BUs gets compounding advantages no standalone company can buy. Imagine the apparel BU's customer-segmentation model getting reused by the F&B BU. Imagine the legal compliance vault built for one subsidiary serving every other one. Imagine cross-BU benchmarking — which helpdesk converts best, which sales playbook ships fastest — feeding back into every BU's roadmap. Holdings that wake up to this are sitting on an AI flywheel competitors can't replicate. The shift from seven vendors to one platform isn't just cost cleanup; it's how a portfolio of separate businesses becomes something more than the sum of its parts. Move first.
I have spent enough time inside diversified holdings — family groups, conglomerates, regional players — to know that the seven-vendor problem is rarely about technology. It is about authority. No CEO of a subsidiary wants the holding office picking their tools. No General Counsel wants to be told that a subsidiary's vendor put customer data in a non-compliant region. And no holding CFO has the patience to read seven different AI vendor contracts to figure out what was actually bought. The fix is not a procurement memo. It is a deliberate operating model: one curated platform with isolated vaults per BU (so the autonomy concern dies), one governance console at the holding (so legal sleeps at night), and one cost line in the consolidated budget. That is what we built into the Group Edition of the Ibizai AI Suite — not because it is technically clever, but because every holding I have advised has eventually had to invent it under pressure. Better to design it than to inherit it.
Sources & References
- Enterprise AI Adoption Survey 2025 — Gartner (2025-09-01)
41% of multi-BU enterprises use 6+ AI vendors with no central inventory
View source - State of AI in the Enterprise — McKinsey & Company (2025-11-01)
Average AI tool overlap across business units in diversified groups: 38%
View source - The Hidden Cost of Shadow AI — MIT Sloan Management Review (2025-08-01)
Shadow AI procurement averages 30-60% over consolidated plan
View source
AI education should be as intelligent as the technology it teaches. Our program adapts to your role, industry, and experience level to deliver exactly what you need — nothing more, nothing less.
Comments (0)
No comments yet. Be the first to share your thoughts!