Usha’s AI creative practice: from faster assets to testable full-funnel learning

01 Facts & context

Usha’s BLDC fan campaign in India

Amazon Ads presents Usha’s campaign for AeroEdge Plus BLDC fans as an example of combining generated imagery with advertising across the purchase journey. Its source notes identify Usha data from India in 2025. The account describes online product comparison, a substantial digital budget and creative variants adapted to household contexts. These are observations reported by the advertiser and platform, rather than an independently evaluated account of the wider Indian appliance market. The present article examines the operating choices suggested by that account, without treating its publication as a new campaign launched in September 2026.

The clearest reported outcomes are increased branded search, advertising-driven page views and return on advertising spend. The page reports a 32% improvement in ROAS. This is an attributed advertising metric, not a profit margin or an estimate of additional sales caused solely by generated images. Several activities changed together. The account does not provide the underlying customer records, complete spending history or a counterfactual for the whole program. Its brand-recall wording also combines percentages with an unclear baseline, so it should not be converted into an absolute recall rate. A useful reading begins by identifying what was observed and which explanations still need testing.

02 GSP analysis

From case results to experimental questions: separate concurrent changes

Usha changed AI imagery, regional contexts, audience targeting and budget allocation together. The reported outcome therefore concerns a combined program. IAB’s incrementality guidance calls for a match between method and question. To ask whether generated images outperform existing images, keep audience, placement, spend and prices as comparable as possible. To ask whether the entire new workflow merits investment, compare the bundle, but do not assign all benefits to a single tool. Both questions are useful; their evidence requirements differ.

GSP separates creative learning into content and process hypotheses. Content hypotheses ask which setting helps customers understand efficiency, comfort or product fit. Process hypotheses ask whether generation and review are faster, rework is lower and promising assets reach campaigns in time. Click-through rate can help screen creative, while net orders and returns test whether the message attracts suitable purchasers. Giving each stage an appropriate measure avoids using one sales percentage as simultaneous proof of creative quality, production efficiency and a stronger customer relationship.

03 GSP analysis

Full-funnel learning must allow time for brand effects

Meridian explains how brand equity can influence search and consideration before affecting purchase. This offers another reading of Usha’s search and sales results. More branded search can be an intermediate signal of demand formation, but an unsuitable model may give too much credit to performance advertising that captures that demand. Fans also have seasonal and weather-related demand. A quarter-on-quarter change should not be treated as directly comparable to a year-on-year target.

A brand can maintain two connected observation records. The shorter cycle examines message comprehension, visit quality and transaction friction. The longer cycle follows active search, purchase reasons, service feedback and repeated choice. Each comparison should state its time basis and media mix instead of selecting favorable figures across quarterly, annual and control-group comparisons. GSP sees AI creative’s strongest organizational contribution as reducing the cost of proposing and testing hypotheses, allowing unsuitable contexts to be rejected sooner. Whether that learning becomes durable brand value still requires observation across seasons and actual product use.

04 GSP analysis

Where product comparison is moving

The strategic question behind the creative tool is where customers compare alternatives. An established appliance brand may be familiar to shoppers while its product pages fail to explain energy use, room suitability, installation or service. Familiarity then creates an initial opportunity without completing the decision. GSP reads Usha’s account as a prompt to examine that gap: how effectively does an existing brand promise survive the move from broad advertising to a specific product comparison? This is an interpretation of the case, not proof that every buyer has abandoned stores or television.

Budget allocation should follow that examination. A brand that regards e-commerce mainly as a place to clear inventory will brief creative differently from one using product pages to explain long-term ownership value. The same image generator cannot resolve this disagreement. Before changing channel budgets, identify the questions customers cannot currently answer, the relevant product information, and the stage at which they leave. Search queries, service conversations and purchase feedback can provide complementary evidence; none alone describes the entire journey.

A useful counterfactual is to imagine buying the tool while leaving priorities unchanged. Production could become cheaper while product understanding remains weak. This separates a procurement benefit from a strategic benefit and gives the team a more precise question to investigate.

05 GSP analysis

Treat creative assets as testable variables

Generated imagery can make creative assets easier to vary, but variation only creates value when it tests a meaningful proposition. A different background may help customers understand room fit; a decorative change with no relationship to purchase needs may merely add production volume. Begin with a stated hypothesis about the information a customer needs. Then decide which visual element represents that hypothesis and which elements must remain stable so that the comparison can be interpreted.

For example, a proposed test might compare a quiet evening setting with a general household setting for the same fan. Product, price, audience and placement should be comparable. This is an illustrative design, not a test result disclosed by Usha. Before launch, check that the image represents the real product and does not imply unsupported performance. A visually persuasive rendering can create disappointed buyers if dimensions, installation conditions or product capabilities are inaccurate.

The operating asset is therefore the combination of a reusable brief, a controlled set of variants and a recorded result. Faster production can support this combination but cannot replace it. Measure review time and rejected outputs as well as images generated, because an expanding library of unusable material can consume the capacity that automation was intended to release.

06 GSP analysis

Connect search, visits and purchase objectives

Search, visits and purchases describe different parts of a customer’s progress. Usha’s account brings them together within one campaign narrative. That makes the narrative useful for asking how activities connect, but the coexistence of favorable measures does not establish that each caused the next. Branded search may rise because of awareness activity, seasonal interest, distribution changes or other communication. Visits can increase without sufficient product understanding, and attributed orders may include customers who would already have purchased.

A coordinated team should make these relationships explicit. The brand team can explain the intended customer response, the media team can describe who received which exposure, and the commerce team can document availability and transaction friction. Together they can identify where a proposed mechanism is supported and where it is only assumed. The public case does not establish Usha’s exact reporting structure or budget authority, so the organizational model discussed here is a GSP recommendation.

A practical shared review follows a customer question across the sequence: was the relevant message delivered, did it lead to useful comparison, and could the promised offer actually be bought? The objective is a coherent explanation with clear responsibilities, rather than one blended score that gives every team credit for the same order.

07 GSP analysis

Adapt creative to relevant regional contexts

Regional adaptation is useful when it reflects a real difference in customer needs. Housing conditions, climate, installation arrangements and household routines may change which product information matters most. Geographic labels alone do not establish those differences. A team should be able to state the observed need behind a regional version and identify the research or operating records that support it. Otherwise, generating more local-looking images can reinforce assumptions without improving relevance.

Consider a hypothetical comparison between customers asking about overnight noise and customers asking about airflow in a larger room. The creative brief would differ because the questions differ. Both versions still need accurate product information and an appropriate destination page. Search records can suggest a question, service teams can explain its practical meaning, and customer interviews can reveal whether the planned message answers it. This triangulation is more useful than treating an isolated click pattern as a complete portrait of an audience.

The learning cycle then runs from observed question to visual hypothesis, controlled exposure and a documented response. If results contradict the original assumption, revise the grouping rather than manufacturing more variants. Usha provides a starting point for this inquiry; the public account does not reveal enough audience-level data to prescribe another brand’s regional segmentation.

08 GSP analysis

Assign creative permissions and review responsibilities

When a media team can create variants directly, production permissions and review responsibilities need to be explicit. Some changes may fit within an approved template, while others alter a claim or the representation of the product. Treating every variation identically either delays routine work or leaves consequential changes without an accountable reviewer. The task is to distinguish these categories before campaign pressure encourages improvised decisions.

GSP proposes three connected responsibilities. The brand and product teams define accurate claims, visual principles and product constraints. The creative or growth operator produces variants, records which hypothesis each tests and removes unsuitable outputs. A campaign owner coordinates learning and budget changes across activities. These are suggested roles; they are not a description of Usha’s internal organization. Smaller businesses may combine roles, provided that responsibility remains visible and a person does not silently approve an unverified product claim.

The handoff should contain the approved product information, the intended audience question, the version history and the reason for accepting or rejecting an asset. This allows another colleague to understand the decision later. Without that record, apparent speed depends on individual memory and becomes difficult to sustain when people change or the number of products grows.

09 GSP analysis

Evaluate costs and alternative explanations

A continuing creative capability consumes more than a software subscription. Teams spend time preparing accurate product material, defining hypotheses, reviewing images, setting up comparisons and interpreting results. Campaign and analytical work may move between departments without disappearing. An honest cost comparison includes these activities and records whether automation actually reduces elapsed time or simply transfers work to another team.

The published outcome also has alternative explanations. Product availability, price, weather, competitive activity and overall media spending can influence sales and advertising returns. The platform case is not enough to assign a separate effect to each factor. GSP therefore treats generated creative as one plausible contributor within a broader operating program. It would be equally unsupported to declare the tool ineffective simply because the case lacks a full causal evaluation; the appropriate response is a better-defined local test.

Before moving from trial to recurring expenditure, compare a specified new workflow with a credible alternative. Record net orders after cancellations and returns, the relevant contribution from those orders and the full cost of producing and managing creative. A positive click response with excessive review work or disappointed customers may still be useful learning, but it does not yet justify expanding the program.

10 Suggested actions

Establish a learning rhythm before scaling

Implementation can begin with one product line and one customer question. Confirm the product information and the destination page, select a limited group of creative hypotheses, and assign a campaign owner. Decide in advance which conditions will be held comparable and what evidence will determine the next decision. This preparation makes a small test informative; buying a large collection of tools before specifying the question does not.

Next, establish a review rhythm appropriate to traffic and the purchase cycle. Each review should cover delivery, data completeness, customer response and production effort. Maintain an asset register linking every version to its hypothesis, approval status and observed result. Retire outdated or unsupported variants. Expansion to another region or product should have a reason, a responsible person and an explanation of which conclusions can reasonably transfer.

Finally, connect the proven workflow to broader media decisions. A successful image comparison does not automatically justify moving substantial budget between awareness and purchase activity. That decision requires evidence about the larger customer journey and the alternatives displaced by the move. The sequence is to establish useful learning, test whether it transfers and then change the scale of commitment. Different firms may move at different speeds because their product cycles and data are different.

11 Suggested actions

Measure production, customer response and business outcomes

Measurement should distinguish production performance, customer response and business outcomes. Production measures include time from a usable brief to an approved asset, review effort and the share of outputs requiring correction. Response measures include relevant visits, message understanding and product comparison behavior. Business measures include net orders, acquisition cost and contribution after the defined delivery and creative costs. ROAS is informative within this picture but does not substitute for it.

For a local experiment, compare equivalent products, periods and eligible audiences as closely as practical. Random assignment can strengthen interpretation when feasible. If the organization instead uses matched regions or historical observations, explain the remaining differences and the assumptions required. A four-to-eight-week window may be a practical starting proposal for some campaigns, but it is not a universal requirement or a finding from the Usha account. The observation period must allow enough purchasing behavior to occur.

Set adjustment conditions before results arrive. Investigate if generated variants attract visits but create confusion, if returns increase, or if review effort removes the production benefit. Record inconclusive outcomes as inconclusive. Repeatedly checking noisy results until one favorable percentage appears can turn ordinary variation into a misleading success story and make subsequent budget decisions less dependable.

12 Suggested actions

Conditions for useful replication

The approach is most useful when customers have identifiable product-comparison needs, the brand has reliable product information and the team can connect creative exposure with meaningful observations. It also requires someone able to act on what the test reveals. Those conditions matter more than an ambition to produce a large number of images. A small brand with a clear question and a manageable workflow may learn more than a larger one that has many tools but conflicting objectives.

Where basic information is missing, begin there. Unclear installation conditions need product or service clarification before a persuasive visual campaign. Fragmented measurement needs a common definition of orders and exposure before teams compare returns. Conflicting incentives need an agreed decision owner before budgets can move coherently. These are different constraints and should not all be assigned to the creative team to solve.

The transferable lesson is an operating discipline: make a specific promise understandable, test how customers respond, examine alternative explanations and retain the learning. Usha’s reported experience makes this discipline worth examining. It does not supply a guaranteed return, a universal media allocation or a ready-made organizational design. The quality of replication depends on how carefully another business adapts the mechanism to its own customers and delivery capacity.