What is AI by business function?
AI by business function and use case
AI by business function means looking at where artificial intelligence actually does useful work inside an organisation, function by function: customer service, operations, marketing, sales, finance administration and HR administration. The pattern across official surveys is consistent. AI lands first on repetitive, text-heavy, checkable tasks. Marketing, sales and administration tend to adopt earliest, while value is real but uneven and almost always needs human review.
Reviewed by Jackie, Head of Learning & Development, Levellers · Last reviewed 8 June 2026
What this means
This is a hub page. It gives you the map of how AI is used across the main business functions and routes you to a dedicated article for each one. It is about adoption and practical use, not about AI regulation.
The simplest way to think about it: AI is not equally useful everywhere. It is strongest where work is repetitive, language-based and easy to check, and weakest where work depends on physical presence, hard judgement or accountability that cannot be delegated to a machine. That is why the same few functions show up first in survey after survey.
For a small or mid-sized organisation, the practical question is not "should we use AI" but "which function should we start with, and how much checking does it need". This page helps you answer that, then sends you to the function articles for the detail.
Why it matters
Most organisations are still early. The US Census Bureau's Business Trends and Outlook Survey reported a national AI use rate of 19.8% of firms, rising to about 32% on an employment-weighted basis, with adoption expected to reach 22% within six months. In the UK, the Department for Science, Innovation and Technology found that roughly 1 in 6 businesses currently use at least one AI technology. Eurostat reported that 20.0% of EU enterprises with 10 or more employees used AI technologies in 2025, up 6.5 percentage points from 13.5% in 2024.
Commercial surveys report much higher headline figures: McKinsey's 2025 State of AI survey found 88% of organisations report regular AI use in at least one business function, up from 78% a year earlier. But the same survey found just 39% report any EBIT impact at the enterprise level, and only about 6% qualify as high performers seeing meaningful financial returns. So the value is real but concentrated and uneven. Knowing which functions pay off first, and which look stronger than they are, is the difference between a useful pilot and wasted effort. This hub gives you that map and links down to each function so you can prioritise deliberately rather than copying whatever is loudest in the market.
How it works
The common pattern: repetitive, text-heavy, checkable work goes first
Across the major official datasets, the same functions lead. The Census working paper on AI diffusion found that among firms using AI, the most common function is sales and marketing (52% of adopting firms), followed by strategy and business development (45%), IT (41%) and research and development (40%); customer service, finance and accounting, and HR sit in the mid-tier. Eurostat found that 34.70% of AI-using enterprises used AI for marketing or sales, the single most common purpose, ahead of business administration (31.05%) and accounting or finance (23.1%), with logistics the least common (6.08%). The UK DSIT survey found the business areas most likely to be using or planning AI were marketing (72%), administration (72%) and IT (64%).
The logic behind that ordering is durable even as tools change. AI is strong where the task is language-based (drafting, summarising, classifying), repetitive (the same shape of work many times), and checkable (a human can quickly verify whether the result is right). It is weak where work is physical, depends on accountable judgement, or where an error is costly and hard to catch. Use that test, not the tool of the month, to decide where AI fits.
Customer service
This is one of the most studied functions. A controlled study of 5,179 customer support agents found that access to a generative AI assistant increased productivity, measured as issues resolved per hour, by 15% on average, with the largest gains (around 30%) for less skilled and less experienced staff and little gain for the most experienced. AI helps draft replies, summarise tickets, surface knowledge and triage queries. The boundary is that customers notice when a relationship is reduced to a script, and wrong answers create refunds and complaints, so human handling of complex or sensitive cases stays essential. See ai-for-customer-service.
Operations
Operations covers forecasting, scheduling, quality checks, document handling and process automation. Adoption is solid but practical: the UK DSIT survey put operations at 53% among businesses using or planning AI. The durable fit is in structured, data-heavy back-office tasks where patterns repeat. The boundary is data quality and integration; fragmented systems and messy data limit what AI can do, and physical or safety-critical steps still need human authorisation. See ai-for-operations.
Marketing
Marketing is consistently the fastest-moving function. It is text-heavy and creative, which suits drafting, ideation, content variants, summarising research and basic analysis. The risk is also highest here on a per-task basis: studies of marketers report that a large share encounter AI errors regularly, and fabricated statistics or claims can ship publicly and damage credibility. Marketing is a natural starting point precisely because it is low-stakes to pilot, provided a human reviews everything before it goes out. See ai-for-marketing-teams.
Sales
In sales the durable uses are lead scoring, pipeline forecasting, CRM data capture, call summarisation and drafting personalised outreach. The UK DSIT survey put sales at 49% among businesses using or planning AI. The pattern is that AI handles the data work and the first draft while the person owns the relationship, the negotiation and the judgement. Value depends heavily on clean CRM data; poor inputs produce confident but unreliable scores and forecasts. See ai-for-sales-operations.
Finance administration
Finance administration is a strong fit for AI on the bookkeeping and processing layer: invoice capture, coding, bank reconciliation, expense categorisation and anomaly flagging. The work is rules-based and checkable, and errors are caught at reconciliation. Eurostat found about 23.1% of AI-using enterprises applied AI in accounting, controlling or finance management. The boundary is firm: AI assists, but accountability for the numbers, controls and compliance stays with named people, and historical data quality determines whether the system is reliable. See ai-for-finance-administration.
HR administration
HR administration is text-heavy (drafting job descriptions, policies, summaries, answering routine staff queries) and so suits AI for the administrative layer. It also carries the sharpest legal boundary of any function covered here. Where AI screens, ranks or scores candidates or employees, equality law applies regardless of whether a vendor supplied the tool, and the employer remains liable for discriminatory effects. Human review of any high-impact decision is not optional. Adoption is correspondingly lower: the UK DSIT survey put HR at 26% among businesses using or planning AI. See ai-for-hr-administration.
How to prioritise across functions
For a small or mid-sized organisation, start with one function where the work is repetitive and checkable, the cost of an error is manageable, and the data already exists in usable form. Marketing drafts, customer service replies, finance processing and sales admin are common low-risk entry points. Avoid starting in a function where errors are costly and hard to catch, or where decisions are legally sensitive (notably HR screening). The Census data shows most adopters stay narrow: about 57% use AI in three or fewer functions. That is a reasonable model to copy. Master one function, measure it, then expand. The broader sequencing belongs to the adoption journey itself; see ai-adoption, ai-for-small-business, ai-productivity and ai-workflow-assessment, and route governance questions to ai-governance.
Examples
Customer support productivity. In a study of 5,179 customer support agents at a large software firm, access to a generative AI assistant increased issues resolved per hour by 15% on average, with the gain reaching around 30% for the least experienced agents and minimal change for the most experienced. The AI worked by surfacing the practices of the best agents to everyone else. This is the clearest documented example of AI lifting a whole function, and it shows the augmentation pattern: the human stayed in the conversation.
Cross-function adoption ordering. The US Census Bureau's analysis of AI-adopting firms found sales and marketing leading at 52%, then strategy and business development at 45%, IT at 41% and research and development at 40%, with customer service, finance and accounting and HR in the mid-tier. This is real reported behaviour across a nationally representative sample, not a vendor projection, and it is the empirical basis for the "marketing and admin first" pattern.
Augmentation over replacement. The same Census research found that about 44% of AI-using firms use AI to supplement a task done by an employee, while only about 2% reported any AI-related fall in headcount. Most firms reported no employment change. The dominant real-world pattern is AI assisting people, not removing them.
Common misunderstandings
"AI is useful everywhere equally." It is not. The evidence is consistent that value concentrates in language-based, repetitive, checkable work and thins out elsewhere. Treating every function the same wastes effort.
"High adoption means high returns." McKinsey's 2025 survey found 88% of organisations report regular AI use in at least one function, but just 39% report any EBIT impact at the enterprise level and only about 6% are high performers seeing meaningful returns. Using AI and capturing value are different things.
"AI replaces the function." In the most representative data, AI overwhelmingly augments tasks rather than substituting them, and headcount reductions attributed to AI are rare. The realistic model is a support layer, not a replacement.
"You should roll it out across the whole business at once." Most adopters deliberately stay narrow, using AI in three or fewer functions. Depth in one function beats shallow use across many.
"The tool is the decision." The durable choice is which task and how much review, not which product. Tools change quickly; the logic of where AI fits does not.
Risks and boundaries
The recurring boundary across every function is accuracy and accountability. Generative tools produce fluent text that can be wrong, and the error is often invisible until someone checks. Studies of marketers report frequent AI errors and instances of fabricated content being published. UK government research found that most businesses using AI keep significant human oversight, with about two-thirds reporting significant checking of AI output and only a small minority reporting none. That is the right default.
Two functions carry sharper boundaries. In HR administration, where AI is used to screen, rank or score people, anti-discrimination law applies regardless of whether the tool came from a vendor, and the employer stays liable for discriminatory effects; a human must own high-impact decisions. In finance administration, accountability for the numbers and for compliance cannot be delegated to a model. Data quality is a cross-cutting constraint: fragmented or inconsistent data produces confident but unreliable results in sales scoring, forecasting and finance processing alike. None of this is regulatory advice; for the governance angle see ai-governance, and keep this hub focused on use by function.
What to do next
First, pick one function to start, using the test of repetitive, text-heavy, checkable work with a manageable cost of error. For most small and mid-sized organisations that means marketing drafts, customer service support, finance processing or sales admin, not HR screening.
Second, define what good looks like before you start: a baseline (time taken, error rate, volume handled) and a target, so you can tell whether the pilot worked. UK government research found that most adopters report productivity gains but most have not yet seen a revenue change, so measure the thing you actually expect to move.
Third, build human review into the workflow from day one, sized to the risk. Light review for internal drafting, mandatory sign-off for anything customer-facing, financial or personnel-related.
Fourth, stay narrow on purpose. Get one function working and measured before expanding, mirroring the majority of adopters who use AI in three or fewer functions.
Fifth, route the wider questions correctly. Use ai-adoption and ai-for-small-business for the overall journey, ai-workflow-assessment to find the right tasks, ai-productivity for measurement, and ai-governance for oversight. Then go deep in the relevant function article. Benchmarks that should change your plan: if a function shows no measurable gain after a fair pilot, stop and pick another; if error rates in review stay high, the task is not yet a good fit; if a use touches hiring, pay or compliance, raise the review bar before scaling.
Explore individual entries: AI for customer service, AI for finance administration, AI for HR administration, AI for marketing teams, AI for operations, AI-assisted sales operations, AI for small business.
Have a question or a suggestion, or want to understand how we research and review these guides? Read about our editorial standards and how to reach us.
FAQs
Which business function should adopt AI first?
Start where work is repetitive, language-based and easy to check, and where an error is cheap to catch. In practice that is usually marketing content, customer service support, finance processing or sales admin. Official surveys consistently show marketing, sales and administration adopting earliest for exactly these reasons.
Where is AI weakest by function?
Where work is physical, depends on accountable human judgement, or where errors are costly and hard to detect. Logistics is the least common AI use in Eurostat's data, and functions like HR screening carry legal boundaries that limit how far automation can responsibly go.
Does AI replace whole functions?
Rarely, on current evidence. The most representative data shows AI mostly augments tasks rather than substituting them, and AI-related headcount reductions are reported by only a small share of firms. The realistic model is a support layer that makes existing staff more productive.
How many functions should we use AI in?
Start with one and expand deliberately. Most adopters use AI in three or fewer functions, and depth in a single function generally beats shallow use spread across many. Master and measure one before adding the next.
Why do adoption figures vary so much between surveys?
They measure different things. National statistics bodies like the US Census Bureau, the UK ONS and Eurostat ask narrow questions about recent reported use and find figures around 18% to 23%. Commercial surveys reporting 78% or 88% use broader definitions that include experimentation. Lower official figures are usually the better benchmark.
How much human review does AI work need?
Enough to catch the errors the task can produce. UK government research found most adopters keep significant human oversight. A practical rule: light review for internal drafts, mandatory sign-off for anything customer-facing, financial or personnel-related.
Is this page about AI regulation?
No. This is a hub about how AI is used across business functions and where it helps. Regulatory and oversight questions belong in the governance material; see ai-governance.
Where do I go for detail on my function?
Follow the dedicated article: ai-for-customer-service, ai-for-operations, ai-for-marketing-teams, ai-for-sales-operations, ai-for-finance-administration or ai-for-hr-administration. This hub is the map; those are the detailed children.