Take a company that bought a general AI assistant for every employee in the spring. By the July close the usage report looks healthy. Sales drafts emails with it. The FP&A analyst tidies variance commentary. Accounts payable still codes invoices by hand, the controller still chases the same reconciliations on day three, and nobody in the leadership meeting can name a process that now runs differently.
That is a common shape for enterprise AI adoption in 2026. The licences are real, and so are the individual time savings. What has not moved is the way the company works, because a tool that knows nothing about the company has no way to change how it operates.
Why a chatbot for everyone changes so little
A general assistant learns from public text. It does not know what your company means by an active customer, which cost centre owns a contract, or what the approval limit is for a capital purchase in Germany. Ask it something that depends on any of that and it will either decline or produce an answer that sounds right and is not.
So people sensibly use it where company knowledge does not matter: rewording, summarising, first drafts. Useful, and invisible at the level of the process. The close calendar, the payment run and the forecast cycle look as they did before.
The surveys point the same way. Gartner's 2025 survey of 183 CFOs and other senior finance leaders found "59% of finance leaders reporting the use of AI in their finance function", with "91% of respondents reporting low or moderate impact initially". In June 2026 Gartner reported from the same survey that "84% of finance organizations have implemented or are planning to implement AI, yet only 7% report a high or very high impact". Use is wide. Impact, so far, is narrow.
What enterprise AI adoption needs
Four things separate a rollout that changes work from one that adds a browser tab.
| Element | What it means | What happens without it | Usual owner |
|---|---|---|---|
| Company context | The tool draws on your own definitions, policies, chart of accounts and approved data, and sees only what the user may see | Answers are generic, or confidently wrong about your business | Finance and IT, with each policy owner |
| Use cases by role | For each role, a short list of tasks where AI helps, with inputs, expected output and reviewer written down | People experiment alone and the gains stay with individuals | Function heads |
| Guardrails | Rules on which data goes into which tool, what needs human review and what is logged | Either nothing is allowed and use stalls, or everything is and risk grows quietly | CFO, with IT security and legal |
| Measurement | A baseline for the process and a record of what changed after | Login counts stand in for results | Finance |
Company context. This is the layer most rollouts skip, and the next issue in this series, AI With Company Context: The Layer Most Rollouts Skip, covers it in full. In short, the tool needs to read the company's own approved material, show where each answer came from, and respect the permissions of the person asking.
Use cases by role. "Use AI more" is not an instruction anyone can act on. "Draft the first pass of the monthly flux commentary from the trial balance and last month's commentary, for the FP&A lead to edit" is. Write a handful of these for each role, starting with high-volume, rules-heavy work. An earlier issue on scoring candidate tasks sets out one way to choose.
Guardrails. Someone has to decide which classes of data can go into which tools, which outputs need a named reviewer, and what gets logged. Those decisions belong in a short written policy before use spreads past the first teams.
Measurement. Measure the process rather than the tool. Cycle time, exception rates, rework and hours put to other use tell you whether work changed. Usage dashboards tell you only that people opened the application.
Why finance usually ends up owning it
AI runs on data, and in most companies finance owns the data people trust: the ledger, the customer and vendor masters, the cost centre structure, the delegation of authority. Finance also runs the controls that AI will sit beside, from segregation of duties to review sign-offs. When a tool drafts a customer credit note or proposes a journal, it is finance's controls that have to hold.
Deloitte's CFO Signals survey for the second quarter of 2026 asked 200 North American CFOs at companies with at least US$1 billion in revenue. It found that "93% say their organizations use AI across key operations" and that "One in five (19%) CFOs say they have greatest responsibility for AI governance". Balancing speed and risk was the top challenge, cited by 59%.
Not every CFO needs to chair AI governance. Every CFO does need a formal voice in it, because decisions about data access, review and evidence land on finance whoever runs the committee.
The same survey found 96% of these CFOs somewhat or very confident in their company's AI governance framework. Confidence is a fair starting point. The practical test is narrower: for any AI-assisted step, can you show who could see what, who reviewed the output, and where the record sits?
A rollout sequence by role
Choose the role and the process. Pick one with a lot of repeat work and clear rules, such as invoice coding or bank matching.
List the context that process needs. The coding guide, the approval limits, the vendor master, last quarter's exceptions. Note where each lives and who owns it.
Decide the review and the record first. Name who checks the output, against what, and where the evidence is kept, before anyone uses the tool on live work.
Take a baseline. A month of cycle time, exception rate and rework, measured the way you will measure it afterwards.
Run, measure and fix. The first month will expose gaps in the context and the rules. Close them before moving to the next role.
What enterprise AI adoption will not fix
It cannot repair a process nobody has designed. If the approval matrix lives in three people's heads, an AI tool cannot follow it, and neither can a new hire.
Nor will it settle accountability. A board, an auditor or a lender will still hold a named person responsible for the number, the payment or the disclosure. Decide who that person is for each AI-assisted step before the tool goes live on that step.
Frequently asked questions
Why does giving employees a general AI chatbot not change much?
A general chatbot does not know the company's definitions, policies or data, so people use it for drafting and summarising rather than for the processes that drive cost and risk. The time savings stay with individuals and the way work gets done stays the same.
What does enterprise AI adoption require?
Enterprise AI adoption requires company context, use cases defined for each role, guardrails on data and review, and measurement of the process before and after. Licences and general training on their own do not change how work is done.
Should the CFO own AI governance?
Not always, but the CFO should have a formal say, because AI touches the data, approvals and controls finance runs. In Deloitte's Q2 2026 CFO Signals survey, 19% of CFOs said they have greatest responsibility for AI governance.
How do you measure whether AI adoption is working?
Measure the process, not tool usage: cycle time, exception rates, rework and hours put to other use, against a baseline taken before the change. Usage figures show only that people opened the tool.
Where should a company start with AI beyond a chatbot?
Start with one role and one high-volume, rules-heavy process, give the tool the company context it needs, set a named review step, and measure the result before extending to the next role.
Where CFOLogic fits
We work with CFOs on company-wide AI adoption as an approach: defining use cases by role, supplying company context and setting guardrails, on top of the systems the company already runs. The tools we design and build are read-only by default, and a person signs off anything that leaves the team. More on how we work is on the CFO Success Partners page.