Getting a Sceptical Finance Director to Approve AI Spend

You have run the pilot, the team likes it, and the purchase is stuck with finance. The requests coming back are for numbers you do not have and cannot honestly invent.

A sceptical finance director is almost never objecting to AI. They are objecting to three specific gaps: they cannot see where the saving lands, they cannot see who carries the risk when it goes wrong, and they cannot see how the company gets out. Answer those three and the meeting changes character.

Objection one: the saving is unclear

"It saves time" is not a finance answer. Time saved becomes money only when it turns into headcount you did not hire, overtime you did not pay, or work you now do that you were previously turning away.

Pick one of those three and be specific about which. If nobody leaves and nobody is hired, say so, and argue on capacity instead: the same team handles more work, or handles the same work with fewer errors that cost money to fix.

The strongest evidence comes from your own pilot, not from a vendor. Before the pilot, record how long a specific recurring task takes and how often it needs rework. After, record the same two numbers. Two measured tasks beat a page of general claims. If the seat price is what is being questioned, convert it into what you pay per approved output, which is a figure finance can hold against the work itself.

Be honest about what the pilot does not prove. A pilot with your keenest people is not a forecast for the whole team. Say that out loud in the meeting, because a finance director who spots an unstated assumption stops believing the ones you did state.

Objection two: the liability is unclear

The second question is who is responsible when an AI output goes out wrong and a customer, a regulator or an employee is affected.

The answer needs three parts. First, the control: does a human approve output before it reaches anyone outside the team? In SicherOne, a human approves agent output before it ships, which means the accountable party is the named approver, not an unattended system. Second, the record: can you show afterwards what the agent produced, what the human changed, and who signed it off? Third, the contract: what does the vendor agreement actually say about responsibility, data handling and indemnity?

Do not answer the third part from memory. Read the terms and bring the relevant clauses. Finance directors are used to people describing contracts optimistically.

Where data location is the sticking point, a self hosted private model changes the shape of the conversation, because the sensitive material stays inside infrastructure you already control and already had approved. Either way, walk in with the specific questions about where your prompts are stored already answered.

Objection three: the exit is unclear

This is the objection most teams do not prepare for, and it is often the real one. The fear is not that the tool fails. It is that in two years the tool is load bearing, the price goes up, and there is no way to leave.

Three answers help.

• Where does the data live and can you get it out? If the records are exportable in a usable format, the switching cost is bounded.
• Can you buy less rather than all or nothing? SicherOne is sold per seat with modules separable, so a department can be removed without dismantling the whole thing, which makes buying modules one at a time rather than all at once a real option. Being able to shrink is worth more to a finance director than being able to grow.
• What breaks if you switch it off tomorrow? Answer this honestly at purchase, and you have set the expectation that dependency is a thing you are managing rather than a thing that crept up.

Bring a smaller number than they expect

The instinct is to make the business case as large as possible. It works against you. A large claimed benefit invites scrutiny of every assumption, and one weak assumption sinks the whole case.

A modest claim you can defend line by line survives the meeting. If the honest position is that year one roughly pays for itself and the benefit arrives in year two, say that. Finance people are not allergic to a slow payback. They are allergic to being surprised in month eight.

Give them a decision, not a pitch

End with something specific to approve: a seat count, a module list, a start date, a named review point. Vague enthusiasm produces vague delays.

Put a review date in the proposal yourself, with the criteria written down in advance. Offering to be measured is the single most effective way to reduce resistance, because it moves the decision from permanent commitment to a bounded one. Nobody wants to be the person who approved something with no way back.

Expect the second meeting

You will rarely get a yes first time, and that is not failure. Take the objections away in writing, answer them with evidence you can source, and come back. A finance director who tests a proposal properly is the same person who will defend the spend when someone else questions it later.