You are viewing a free preview of this lesson.
Subscribe to unlock all 13 lessons in this course and every other course on LearningBro.
Spec mapping: AQA 7138 Unit 3.1.3 — Marketing Management (refer to the official AQA specification document for exact wording). This lesson develops the role of artificial intelligence in marketing at A-Level depth — predictive segmentation, generative content, programmatic media buying, dynamic pricing, recommendation engines, sentiment analysis, conversational AI — and the ethics dimension that any serious 15-mark Evaluate response must engage with. AI in marketing is one of the disruptive-technology themes the new 7138 specification (first-teach September 2026) makes explicit; the 15-mark Evaluate at the end of this lesson is the discriminator on this batch and exists to demonstrate Annex 8 sophisticated-concept deployment as the route to Top-band marks.
Connects to:
Definition: AI in marketing is the application of machine-learning models, large language models, computer vision and related techniques to automate, personalise or optimise marketing decisions — including segmentation, content generation, media-buying, pricing, recommendation, sentiment analysis and customer service.
The A-Level move on AI is to refuse the cosmetic version (a chatbot bolted onto a homepage is not an AI marketing strategy) and to treat AI as the re-engineering of marketing-decision infrastructure. The defining strategic characteristic of AI in marketing is that the marketing-mix decisions which used to be made by a brand manager in a planning room are increasingly made by an algorithm in real time, with implications for control, accountability, cost structure and competitive advantage that an examiner expects you to engage with.
Five structural shifts distinguish AI-driven marketing from the prior digital era:
The candidate who treats AI as "another marketing channel" signals lower-band understanding. The candidate who treats AI as a re-engineering of the marketing-decision infrastructure (with its own cost, control, ethics and competitive-advantage characteristics) is engaging with the question at A-Level depth.
Definition: Predictive segmentation uses ML models to identify micro-segments and propensity-to-buy patterns from behavioural and demographic data, going beyond the four canonical bases (geographic / demographic / psychographic / behavioural — see sister lesson order 5) to capture subtle multi-variable interactions that traditional segmentation misses.
The two operational forms are propensity modelling (predicting which existing customers are most likely to buy a given product) and lookalike audience generation (identifying new prospects whose behavioural pattern resembles existing high-value customers). Both are routinely operationalised through paid-social ad platforms and CRM-integrated marketing automation.
The strategic upside is sharper marketing-spend allocation; the strategic risk is opacity (the model identifies a segment but the marketer cannot always articulate why the segment matters) and algorithmic bias (if the training data over-represents certain demographic groups, the model's predictions inherit and amplify that bias).
Definition: Generative AI produces marketing copy, images, video and ad-creative on demand — through large language models (LLMs) for text, diffusion models for image, and increasingly video-generation models for short-form video. The output can be templated at scale (product-description variants for an e-commerce catalogue) or bespoke (personalised email subject lines for individual customers).
The structural trade-off is cost-quality-control: unit cost falls dramatically (a 200-word product description that took a copywriter 30 minutes can be generated in seconds), but average quality is more variable, brand-voice consistency is harder to enforce, and brand-safety incidents become possible (hallucinated product claims, off-brand imagery, factual errors that propagate at scale). The successful pattern emerging in practice is human-in-the-loop workflows — AI generates a draft, a marketer reviews and approves — which captures most of the cost gain while preserving brand-safety control.
Definition: Programmatic media buying automates the purchase of digital advertising inventory through real-time bidding (RTB) on demand-side platforms (DSPs) and supply-side platforms (SSPs). The bid decision — whether to bid on a particular ad impression, and how much to bid — is made by an algorithm in the milliseconds between a user requesting a webpage and the page loading.
| Element | Function |
|---|---|
| DSP (Demand-Side Platform) | The buyer's tool; sets bid rules, audience targeting, frequency caps |
| SSP (Supply-Side Platform) | The publisher's tool; auctions inventory to the highest bidder |
| RTB (Real-Time Bidding) | The auction protocol; sub-100ms bid decisions |
| DMP / CDP (Data / Customer Data Platform) | The audience-data backbone; first-party data is the privileged asset post-cookie-deprecation |
| Attribution model | The arithmetic that assigns conversion credit across the customer journey |
The post-cookie pivot is the defining contemporary shift: the deprecation of third-party cookies (Chrome's phased removal during 2024–2025) has forced advertisers onto first-party data (their own logged-in customer interactions, loyalty programmes and consented opt-ins) as the primary targeting signal. AI in this context is the layer that makes sparse first-party data work at scale — modelling propensity, building lookalikes, optimising frequency.
Definition: Dynamic pricing uses algorithms to set prices that vary with demand, supply, competitor pricing, time-of-day, customer characteristics or other inputs. Hotel-chain room rates, airline ticket prices, ride-share surge pricing and e-commerce daily-deal pricing are all dynamic-pricing applications.
Dynamic pricing is operationally powerful but ethically and reputationally sensitive. The line between dynamic pricing (price responds to market signals, applied uniformly to all customers at a given moment) and discriminatory pricing (different customers see different prices simultaneously based on inferred willingness-to-pay) is the line between standard practice and a brand-damaging consumer-trust incident. The A-Level evaluation move on dynamic pricing is to engage with the customer-trust constraint — perceived unfairness can destroy more value than the dynamic-pricing uplift creates.
Definition: A recommendation engine suggests products, content or actions to a customer based on their behavioural history, similar customers' behaviour (collaborative filtering), product attribute similarity (content-based filtering), or hybrid combinations.
The archetypal cases are Amazon's "customers who bought this also bought…", Netflix's content recommendations and Spotify's Discover Weekly. Recommendation drives a material share of total engagement and revenue on these platforms — frequently cited industry estimates put the share above one-third for the leading platforms (figure indicative, not from any specific business case).
For most A-Level businesses the recommendation-engine question is whether a recommendation layer can be added to an existing e-commerce or content offer and what marketing-mix uplift it produces. The cost is moderate (a third-party recommendation API can be integrated in weeks); the upside is conversion-rate and basket-size uplift attributable through standard A/B testing.
Definition: Sentiment analysis uses natural-language processing to classify the emotional tone of customer-facing text — social media posts, product reviews, customer-service transcripts — as positive, negative or neutral, and to identify trending themes within negative or positive clusters.
The use-cases are brand monitoring (early detection of reputation incidents), crisis detection (identifying when a complaint thread is escalating before it goes viral), competitive intelligence (tracking sentiment around competitor brands and product launches), and product-feedback synthesis (extracting feature requests and complaint patterns from large review corpora).
Definition: Conversational AI covers chatbots, voice assistants, voice-search optimisation and AI-driven customer-service interfaces. The current generation (LLM-backed assistants) is materially more capable than the rule-based chatbots of the 2010s, but the brand-safety and hallucination risks are also materially higher.
Voice-search optimisation is the SEO discipline adapted for voice-query patterns (longer, more conversational, often local-intent queries delivered to Alexa / Google Assistant). Conversational commerce — completing a purchase inside a messaging interface — is the leading edge.
The Annex 8 quantified concept return on marketing spend (#c6) is the diagnostic that turns AI-marketing investment from a fashion question into a financial-discipline question. Annex 7 formula 28 is the operational form:
Return on marketing spend (%) = (Profit from marketing activities ÷ Amount of marketing spend) × 100 (Annex 7 formula 28 — provided in the exam formula sheet)
Worked example. A hypothetical mid-market e-commerce business spends £180k on a 6-month AI-personalisation programme (recommendation engine integration, propensity-model build, additional analytics headcount). Attribution analysis shows the AI-personalisation work generated £540k of incremental gross profit over the period — incremental sales of £1.5m at a 36% gross margin.
ROMS: (£540k ÷ £180k) × 100 = 300%.
Figures fabricated for illustrative purposes; not affiliated with any actual business.
That figure looks compelling — but the A-Level evaluative move is to interrogate the attribution. Three questions an examiner expects:
ROMS deployed without these qualifications is decorative; ROMS deployed with them is the analytical move that earns Annex 8 credit at 15-mark tariff.
flowchart TD
Strategy["AI-marketing<br/>strategy decision"] --> Build["Build vs Buy"]
Build --> Internal["Internal capability<br/>(data team, ML engineers)"]
Build --> Vendor["Third-party platforms<br/>(SaaS personalisation, generative tooling)"]
Internal --> Data["First-party data infrastructure<br/>(CDP, data warehouse)"]
Vendor --> Data
Data --> Operate["Operating cost<br/>(compute, licensing, headcount)"]
Operate --> Measure["Measurement layer<br/>(attribution, ROMS, A/B)"]
Measure --> Iterate["Iterate or kill"]
Iterate -. quarterly review .-> Strategy
style Strategy fill:#1d4ed8,color:#fff
style Measure fill:#15803d,color:#fff
style Iterate fill:#a16207,color:#fff
The structural point this diagram makes: AI-marketing investment is dominated by fixed and front-loaded costs (data infrastructure, model build, capability headcount) with recurring operating costs (compute, vendor licensing) that scale with usage. This cost-structure profile differs sharply from traditional marketing spend (predominantly variable: media-buy costs scale linearly with audience reached) and changes the break-even arithmetic — high fixed-cost models pay off if and only if the personalisation uplift accumulates over multiple campaign cycles.
The AO4 evaluative move on any 15-mark AI-marketing question must engage with the ethics dimension. Six specific issues an examiner expects you to recognise:
| Ethics issue | Mechanism | Marketing-strategy implication |
|---|---|---|
| Algorithmic bias | Training data over-represents certain groups; model outputs inherit and amplify the bias (race / gender / age in audience targeting) | Audience-targeting can become unlawfully discriminatory; reputational risk if exposed |
| Transparency | UK GDPR Article 22 and the EU AI Act require explanation of automated decisions affecting consumers | Operational obligation to document model logic; constraint on opacity-by-default systems |
| Data privacy | First-party data accumulation creates large datasets of personal information | UK GDPR consent, data-minimisation and purpose-limitation principles bind data use |
| Dark patterns | UX choices that nudge consumers into decisions they would not otherwise make (forced consent, hidden subscription rollovers) | Increasingly subject to regulatory action; reputational damage when surfaced |
| Hallucination and misrepresentation | LLM-generated content can produce false product claims, fake reviews, misleading imagery | Brand-safety risk; potential breach of ASA / CAP Code on misleading advertising |
| Workforce displacement | Generative AI substitutes for copywriter, designer, customer-service headcount | Stakeholder-impact question; ESG-reporting implications under the broader corporate-responsibility frame |
The platform-wide A-Level convention is that a 15-mark Evaluate on AI in marketing that ignores the ethics dimension caps at Stronger. A response that engages with at least one ethics issue as a constraint that reshapes the recommendation (not as a closing-paragraph afterthought) reaches Top-band.
Lumen & Loop is a hypothetical mid-market UK e-commerce business selling sustainable home textiles (bedding, towels, lamps) founded in 2017 in Manchester. Revenue grew from £2.1m in 2022 to £14.6m in 2025, primarily through Instagram-led brand marketing and a loyal repeat-purchase base of 28–48 year-old design-conscious customers. Gross margin is 54%; net margin is 7.4%; the brand is widely featured in design and lifestyle press. The board is weighing how to deploy a £1.8m marketing-investment budget over 2026–2028. Two strategic options are on the table.
Option A — Heavy AI-personalisation investment. Build a first-party customer-data platform, integrate a recommendation engine, deploy propensity-modelling for email and paid-social targeting, and pilot generative AI for product-description copy at catalogue scale. Forecast outcomes: a 22% conversion-rate uplift on personalised journeys, a 14% average-order-value uplift on recommendation-served customers, and an estimated 36% return on marketing spend (ROMS) blended across the AI-driven channels. £1.4m of the £1.8m budget is front-loaded fixed cost (platform build, model development, vendor licensing first-year). The remaining £400k is operating spend across the three-year horizon.
Option B — Heavy human-centred brand-creative investment. Commission a sustained brand-storytelling campaign with award-winning creative agency partners; expand the in-house creative team; invest in higher-production-value content (photography, short-form film, longer-form documentary) for owned and paid channels; partner with named designers and craft-makers on co-branded product launches. Forecast outcomes: a 9% brand-awareness uplift in the target demographic, a 17% improvement in unaided brand recall, an estimated 22% ROMS, but a much harder direct-attribution chain to specific revenue uplifts. £600k of the £1.8m is front-loaded creative production; £1.2m is recurring campaign-and-content spend across the three-year horizon.
The founders hold 62% of the equity; institutional investors hold 38%. The board is split — the CFO favours Option A on ROMS grounds; the CMO favours Option B on brand-equity grounds; the founder-CEO is undecided.
Figures and company are fabricated for illustrative purposes; not affiliated with any actual business.
Evaluate which of the two marketing-investment strategies (Option A — heavy AI personalisation; Option B — heavy human-centred brand creative) is most appropriate for Lumen & Loop over the 2026–2028 horizon. (15 marks)
| AO | What the question rewards | Mark weighting on this 15-mark item |
|---|---|---|
| AO1 | Knowledge of AI-marketing use-cases, ROMS arithmetic, brand-equity vs direct-response marketing trade-off, ethics dimension | ~3 marks |
| AO2 | Application to Lumen & Loop — sustainable-textiles positioning, design-conscious target segment, £1.8m budget split, board-level split | ~3 marks |
| AO3 | Analytical chain — because the ROMS forecast is sensitive to attribution-model choice therefore Option A's 36% figure is not directly comparable to Option B's 22% on like-for-like terms; because Option A's £1.4m front-loaded fixed cost concentrates risk into one decision point therefore the cash-flow profile differs from Option B's smoother spend | ~5 marks |
| AO4 | Evaluative judgement — a defensible recommendation with explicit deployment of at least two Annex 8 sophisticated concepts and acknowledgment of the ethics dimension as a binding constraint | ~4 marks |
The 15-mark Evaluate is the discriminator question on the 7138 paper. Top-band 15/15 visibly deploys at least two Annex 8 sophisticated concepts, applies them diagnostically (not as ornament), engages with the ethics dimension as a constraint on the recommendation, and resolves to a structured judgement that names the conditions under which it could be wrong.
Option A looks attractive because the ROMS forecast (36%) is higher than Option B (22%), and the conversion-rate uplift (22%) and average-order-value uplift (14%) would translate directly into revenue growth. AI personalisation also has the advantage of being measurable through standard A/B testing, which makes the marketing-spend justification easier to defend to the board.
Subscribe to continue reading
Get full access to this lesson and all 13 lessons in this course.