Your AI ROI next quarter starts with terminating a contract

Every board of directors asks management about the AI strategy. Where is the gain? What is the return on investment? When do we start to see it on the top and bottom line?

Management has initiated some projects: an AI assistant in customer service, a series of automations in bookkeeping, or perhaps a proof of concept in HR. It sounds like something, but when the math is done, the bottom-line effect is marginal. The board has been asking for six months, and the answer is still the same. The efforts are running, but the effect is nowhere to be seen. Management has not done anything wrong; they have just invested in the wrong place.

Internal AI transformation is the hardest place to get a return: legacy systems, workflows that have taken years to fine-tune, employees who need to be retrained. Add to this IT security, legal, and data governance. A CFO wanting to implement AI in the finance function is fighting against an ERP system from the last millennium. The time horizon is years, the risk is high, and the return is difficult to quantify.

The problem is not just Danish. Uber has publicly admitted that the company spent its entire AI coding tool budget for 2026 by April. COO Andrew MacDonald said in late May that the correlation between rising token consumption and useful features for customers "is not there yet". Walmart has capped employees' AI usage. And Danish Chamber of Commerce's director of digitalisation, Frederikke Saabye, recently told TV 2 that quite a few Danish companies have rolled out enterprise licences broadly without a business case, and that very few can isolate the effect in the business figures.

Successful companies measure business impact, but the Danish AI debate rarely goes any further than that. It tells companies that they need to get a return, but not what they should measure, or where the return is to be found. Meanwhile, advisory firms, IT consultants, and change management consultants sell "AI strategy", internal implementations, and transformation in 12-month courses. They all have a commercial interest in keeping the company on the internal track. That is where their margins are.

The fastest and most frictionless path to AI return lies in the external services you already buy. Risk management, and parts of auditing, legal advice, and industry analysis. These are services where the budget and the outcome purchase already exist, where the supplier is separated from the business, and where the pricing rests on an outdated foundation. No organisational change, no retraining, no legacy systems in the way. Just a contract that can be terminated.

The largest expense item remains internal, in the payroll. But it is furthest away, and part of it is already used to service the external advisors. When a risk advisor asks for policy documents, financial information, or employee data, it is internal employees who spend hours gathering it. Switching suppliers to an AI-native alternative that pulls the data itself not only frees up external money. It also frees up internal time, which today is billed invisibly over the advisor's service.

On external services, the return is fast and measurable, and it hits the accounts in the current quarter. When an AI agent can deliver more service, faster and more consistently, the implementation is not a transformation. It is a change of supplier, and it is the easiest AI implementation a CFO can carry out.

The objection is predictable: if it is so frictionless, why has it not happened already? The answer is trust and liability. When the CFO pays his advisor, he is not just buying the analysis. He is buying someone to call when something goes wrong, and someone to place the liability on.

It is a valid objection, and it can be answered. An AI-native alternative can both deliver the analysis and assume liability contractually through documented methods, professional indemnity insurance, and the same regulatory framework as a traditional advisor. Once that is in place, there is nothing left to justify the old price.

Take risk management as an example. A risk advisor maps exposures, compares across companies, negotiates terms, and on top of that tries to sell consultancy hours on "special" projects. Most of it is pure intelligence work: collecting, comparing, filling out. The classic advisor sells their time; an AI-native alternative sells the result.

We have analysed the risk profile of nearly 1,000 Danish companies with our AI agent vera. Nine out of ten had coverage gaps that their existing advisor had not caught.

The point is not that risk advisory is a special case. The same pattern applies every time a company pays for analysis, documentation, or comparison from the outside. In auditing and law, regulation protects some parts, the auditor's statement, and the lawyer's signature, but the intelligence layer next to the signature is open to the same shift. The signature remains; the fee for the intelligence work does not.

To any CFO who wants to test the point on Monday morning: pull the list of outsourced advisors and set up two columns. What you pay them, and what the work actually consists of. If the work is analysis, comparison, documentation, or reporting, that is the contract you should terminate first.

The board is already asking; next year they will ask why you did not take the easy path.