I model these case studies on one fictional company so you can see the maths, not just hear the warning. This is the second state. It is the one that does the most damage the fastest, because the cost doesn’t trickle. It arrives in a single phone call from a client who is leaving.
The business we’re modelling
The same reference business runs through all four of these case studies, so the only thing changing is the behaviour.
The baseline
- Sector
- B2B logistics and supply chain
- Revenue
- £50M a year
- EBITDA
- 12%, so £6M
- AI investment
- £1.5M, spent over 12 months
- The promise on that spend
- 20% efficiency gain, around +£2M EBITDA by Year 2
A solid firm with a clear plan. Now add one thing the plan never accounted for: people working around it.
What the Shadow AI Tax looks like
The official rollout is slow, and the targets are not. So staff find their own way. They paste raw client contract data and pricing matrices into public AI tools to get their work done faster. No approval. No oversight. No IT sign-off. Just people under pressure reaching for whatever helps them hit the number.
This is not rare. Across the market, 49% of workers admit to doing exactly this, and 51% have wired unmonitored AI into their work systems without IT approval. Every one of those connections is a door left open.
In our model, the door opens onto something specific. The staff are uploading routing logic and margin data, the things that make the business worth buying from. That data flows into a public training loop. And one of the firm’s corporate clients notices its own routes and margins surfacing where they should never be.
The client terminates.
The maths
One lost account at this scale is not a rounding error. It is 8% of the company.
The outcome
- Lost revenue from the terminated client
- −£4M, 8% of ARR
- Regulatory action
- ICO investigation freezes the rollout
- Forensic and legal fees
- £150,000, unbudgeted
- New revenue
- £50M falls to £46M
- EBITDA hit
- Slashed by £630,000 (Capital Destruction)
And the part that doesn’t fit in a table: trust. The firm now carries a multi-year trust-recovery cycle. Every other client that hears about it asks the same question, and the sales team spends the next two years answering it instead of selling.
The result, on the P&L, is a single brutal number.
A free tool, reached for under pressure, took £4M of revenue and £630k of profit off the table, and left a reputation to rebuild. The most expensive AI in the business was the one nobody bought.
This is the Shadow AI Tax. It is the price of a culture where people had to go around the system to do their jobs, and one of them went around it with the data that mattered most. The tool was free. The leak cost £630k of profit and a reputation.
You’ve seen this happen to a far bigger name
In 2023, Samsung engineers pasted confidential source code and internal meeting notes into ChatGPT to speed up their own work. The intent was harmless. The exposure was not. Samsung responded by banning generative AI across the company. One of the most sophisticated technology firms on earth could not stop its own people from feeding secrets into a public tool, so it pulled the plug on all of it.
At full scale
Samsung · a company-wide ban after a leak
That is the Shadow AI Tax at full scale. The staff weren’t reckless. They were trying to be productive. The governance gap turned that into a corporate-wide shutdown.
Source: TechRadar, Samsung workers accidentally leaked company secrets via ChatGPT
The point
The most expensive AI in your business may be the AI you never bought. It runs in the gaps your rollout left open, with your client data inside it. The Friction Scan tells you whether those gaps are open right now.
This analysis is conducted out of curiosity, utilising hypothetical numbers and applying The Unbiased Strategic Clarity Methodology™. It is intended as a simplistic educational case study demonstrating the business impact of assumed organisational dynamics for leaders navigating the current technology era.
