Marketing
Marketing consists of identifying the customers a firm wants to reach, understanding what those customers want, producing content and campaigns that reach them, and measuring what works. The work combines strategic thinking (positioning, brand, message), creative production (copy, imagery, video, experiences), analytical work (segmentation, attribution, measurement), and operational execution (campaign management, channel operations, vendor coordination). AI has been absorbed into marketing practice at a pace and depth that exceeds most other professional domains, in part because marketing's outputs are measurable and the feedback loops allow AI-driven approaches to demonstrate their value quickly.
Content production is the most visible AI application in marketing. Copy for ads, emails, landing pages, blog posts, social media posts, and other channels is produced substantially by large language models, with human marketers providing strategic direction and editorial judgment. Image generation tools produce marketing imagery from text descriptions. Video generation tools produce short-form video content from scripts or prompts. The quality has improved substantially with recent model generations, and the economics have shifted such that producing marketing content at scale is substantially less expensive than it was several years ago. The marketer's role shifts from producing the content to directing the production and selecting from what is produced.
Personalisation and targeting use machine learning extensively. A marketing system that decides which message to show which customer at which moment draws on models that learn from customer behaviour, purchase history, demographic signals, and interaction patterns to identify the content most likely to produce the desired outcome. These systems have been in production for years in large consumer-facing operations (e-commerce recommendations, streaming service recommendations, programmatic advertising) and have extended into smaller operations through marketing platforms that deliver personalisation as a service.
Customer analytics and segmentation rely on machine learning to identify patterns in customer behaviour that inform marketing strategy. Unsupervised learning produces customer segments based on behavioural similarity rather than predefined categories. Supervised learning predicts outcomes of interest (likelihood to purchase, likelihood to churn, lifetime value). The analytical outputs inform marketing decisions (which segments to target, which offers to make, which customers to prioritise for retention) that marketing leaders make based on the combination of the analytical insight and their strategic judgment about the business.
Campaign analysis and attribution use both traditional statistics and machine learning to understand what marketing activity produced what business outcome. This is substantively harder than it sounds, because customers interact with many marketing touches across channels before converting, and attributing the conversion to specific touches requires modelling the full customer journey rather than simply assigning credit to the last touch. AI-driven attribution models produce more defensible views of marketing effectiveness than simpler models, and they inform budget allocation decisions across channels and campaigns.
Brand monitoring and social listening use AI to track mentions of a firm, its products, and its competitors across social media, news, and online forums. Sentiment analysis classifies mentions by tone. Entity recognition identifies what specifically is being discussed. Trend detection identifies emerging topics that merit attention. The outputs inform brand management, crisis response, and competitive intelligence.
The limits of AI in marketing reflect the creative and strategic nature of the work at its most substantive. The AI-generated content that performs well typically involves human direction at the strategic level (what message, for what audience, toward what outcome) and human selection at the editorial level (which of the AI-produced alternatives is actually good). Pure AI-generated content without human direction tends toward generic, undifferentiated output that does not build brand or produce distinctive positioning. The underlying challenge of marketing, which is understanding what customers actually want and producing experiences that actually serve them, remains human work that AI supports rather than replaces.
A concrete example illustrates the production workflow. A marketing manager at a mid-sized firm is launching a product campaign across email, paid social, and the company blog, with campaign launch scheduled for the following Monday. In the AI-augmented workflow, the manager starts Monday with a strategy document (what the campaign is selling, to whom, against what positioning, toward what measurable outcome). That strategy document is the input to every subsequent AI-assisted step. A large language model produces a set of email subject lines to test, three variations of the email body copy, several versions of the social media posts, and a blog post outline. The manager selects the strongest options, refines them against the firm's voice guidelines, and sends the finalists to the creative team for imagery. Image generation tools produce candidate images from brief descriptions, which the creative director selects and refines. The campaign assembly that traditionally took a week of copywriter and designer time compresses into two days of AI-assisted production followed by human review and selection. The campaign launches on schedule. The strategic work, the brand judgment, the selection among alternatives, and the integration of the campaign with the firm's broader marketing programme remain human work. The production volume that was the bottleneck in the traditional workflow is no longer the bottleneck.
Business Operations
Business operations covers the internal functions that keep a firm running: finance operations, human resources, information technology, procurement, legal operations, facilities, and the administrative processes that tie them together. The work is heterogeneous across functions, and the common pattern is that operations work involves substantial document processing, structured workflow, and routine decisioning at scale. These characteristics have made business operations an early and deep adopter of automation historically and of AI more recently.
Internal document processing is a high-volume AI application across operations functions. Invoices arrive and need to be matched against purchase orders, approved, coded to accounts, and paid. Expense reports arrive and need to be validated against policy, approved, and reimbursed. Contracts come in and need to be reviewed, approved, and tracked. Vendor onboarding documents arrive and need to be verified. All of these processes involve extracting structured information from documents, applying business rules, and triggering downstream workflow. AI document extraction combined with rule-based AI for decisioning and automation for workflow execution produces systems that handle these processes with limited human intervention, with humans reviewing exceptions and edge cases.
Human resources has adopted AI across the employee lifecycle. Candidate sourcing uses AI to identify potential candidates from resume databases and professional networks. Resume screening uses AI to assess fit between candidates and open positions. Interview scheduling uses AI-driven calendar tools. Onboarding uses conversational AI to answer common new-employee questions. Performance management uses AI to synthesise performance data from multiple sources. Learning and development uses AI to recommend training content based on role and career goals. Employee experience uses AI to provide internal service portals. The specific applications vary across firms, and the underlying pattern is that routine HR work is increasingly handled by AI-augmented systems while human HR professionals focus on employee relations, policy development, strategic workforce planning, and sensitive individual matters.
Information technology operations use AI extensively in monitoring (identifying anomalies in system behaviour), incident response (diagnosing problems from logs and user reports), security (detecting threats from network and endpoint data), and help desk operations (handling routine user requests through AI-powered self-service). Software development itself has been substantially reshaped by AI coding assistants that suggest code completions, explain existing code, and help debug problems. The productivity effects are substantial for individual developers, and firms have been restructuring software engineering work around AI-augmented approaches.
Procurement uses AI for spend analysis (identifying patterns in purchasing that suggest opportunities for consolidation or renegotiation), supplier risk monitoring (tracking signals that indicate supplier financial distress or compliance issues), contract management (tracking obligations across supplier contracts), and category management (analysing purchasing categories to identify optimisation opportunities). The analytical work is substantive, and the administrative work of procurement operations (requisition processing, purchase order generation, invoice matching) is heavily automated.
Finance operations uses AI across accounts payable (invoice processing as described above), accounts receivable (collections prioritisation, dispute handling), general ledger (anomaly detection in transactions, close process management), and reporting (automated generation of management reports, variance analysis). The finance function has been an aggressive adopter of AI because the work is high-volume, rule-driven, and produces measurable efficiency gains when automated.
The limits of AI in business operations are functional and change-management related rather than technical. The technology handles the work well. Organisations adopting AI in operations often find the harder problems are redesigning business processes to take advantage of the technology, managing the workforce transitions that result from automation, and maintaining the judgment and exception-handling capacity that the remaining human work requires. The AI tools themselves do what they claim to do. The organisational challenge is using them well in a way that produces the efficiency gains without compromising the controls and judgment that keep operations functioning.
A concrete example illustrates how the layers combine in a routine operations function. An invoice arrives in the accounts payable inbox of a mid-sized firm at 9am on Tuesday. In the AI-augmented workflow, an automation system routes the invoice to a document extraction tool that reads the PDF, extracts the vendor name, invoice number, line items, amounts, and payment terms, and produces a structured record. A machine learning system matches the invoice against open purchase orders, finds the corresponding order, and confirms the line items match what was ordered. A rule-based system applies the firm's approval policies, routing the invoice to the appropriate approver based on the amount, the vendor category, and the cost centre. The approver receives a notification, reviews the invoice against their own records, and approves it through a single click. The automation system schedules the payment, updates the general ledger, and closes the invoice record. The entire process takes fifteen minutes of elapsed time and roughly ninety seconds of human attention. The equivalent process in a non-AI-augmented operation would involve a clerk opening the invoice, manually keying its contents into the accounting system, manually matching it against the purchase order, routing it physically or digitally for approval, and manually scheduling the payment once approved, with the work consuming ten to fifteen minutes of human time per invoice. At scale across thousands of invoices per month, the efficiency gain is substantial. The human work that remains is focused on exceptions (invoices that do not match purchase orders, vendors that flag compliance concerns, amounts that exceed routine approval limits) where the judgment that catches problems remains valuable.