ACAI for SBOs: AI-Driven Ad Creation

• ACAI for SBOs is a generative AI toolkit that leverages multimodal structured inputs to help small business owners create brand-aligned advertisements.
• It employs a three-panel interface—Branding, Audience/Goals, and Inspiration Board—to synthesize detailed ‘super prompts’ for advanced MLLMs, ensuring tailored ad generation.
• Empirical studies in London and Manchester demonstrate that ACAI enhances user agency and reduces prompt-related friction, leading to improved creative control and brand specificity.

ACAI for SBOs refers to the use of AI Co-Creation for Advertising and Inspiration, a generative AI-powered tool suite devised explicitly for small business owners (SBOs) to address deficits in design experience, resource constraints, and promptability required for high-quality, brand-aligned advertisement creation. ACAI leverages multimodal LLMs (MLLMs) and structured interfaces to bridge the skill gap between creative vision and domain-specific execution, with empirical assessments demonstrating substantial improvements in user agency, brand control, and creative accessibility (Karnatak et al., 9 Mar 2025, Karnatak et al., 19 Apr 2025).

1. Conceptual Overview and Motivation

ACAI targets SBOs who are typically novice designers and frequently lack technical vocabulary or resources required for effective prompt-based generative advertising workflows. The system utilizes a multimodal interface anchored by three structured panels—Branding (business assets and identity), Audience and Goals (campaign purpose, emotional targeting, market segmentation), and Inspiration Board (reference images, stylistic cues)—to synthesize “super prompts” for the underlying MLLM engine, such as Gemini 1.5 Pro (Karnatak et al., 9 Mar 2025, Karnatak et al., 19 Apr 2025). This configuration lowers the cognitive burden of prompt formulation while enhancing transparency and brand alignment in output generation.

2. Structured Multimodal Interface

The core technical innovation in ACAI is the panel-based input paradigm:

• Branding Asset Panel: Users supply logos, color palettes, typography specifications, and brand values. These elements are either uploaded or selected from pre-existing assets.
• Audience/Goals Panel: SBOs articulate advertising objectives, refine emotional tone, and designate target audience segments.
• Inspiration Board: Facilitates multimodal input by permitting image uploads and real-time stylistic descriptor extraction using MLLMs. Images are analyzed to segment contextually relevant visual features, outputting natural language descriptors (e.g., "weathered and covered in patches of moss, giving a sense of life and softness"). Dynamic sliders allow adjustment of attributes such as depth of field and lighting, supporting iterative refinement.

Inputs from these panels are programmatically synthesized into a “super prompt” and passed to the MLLM for ad brief generation. This approach foregrounds user-defined context and supports granular control over the final advertising outcome.

3. User Agency and Creative Control

Empirical studies involving SBOs in London and Manchester (n=16 and n=6, respectively) demonstrate that ACAI’s structured scaffolding enhances user agency, reduces prompt-related friction, and improves brand specificity in generated advertisements (Karnatak et al., 9 Mar 2025, Karnatak et al., 19 Apr 2025). Users report increased transparency regarding the influence of individual input elements—such as branding assets and audience segmentation—on AI outputs. Requests for input weighting mechanisms highlight a desire for finer granularity of control. Notably, the multimodal inspiration board enables SBOs with clear creative intent but limited design jargon to effectively communicate stylistic preferences, fostering a collaborative, co-creation loop rather than unidirectional automation.

4. Technical Architecture and Output Generation

ACAI’s architecture follows a three-layered structure:

• User Input Layer: Aggregates information from the three panels.
• Processing Layer: Consolidates inputs into a unified “super prompt.” Formally, this is represented as $$SuperPrompt = f(\text{Branding\_Text}, \text{Audience\_Objectives}, \text{Inspiration\_Descriptors})$$, where $f(\cdot)$ encodes the prompt engineering synthesis.
• Output Generation Layer: The MLLM produces an ad brief, typically comprising fields such as Summary, Background, and Foreground, reflecting both textual and stylistic requirements. Output text is editable, supporting user-initiated revision cycles.

The system is implemented on advanced MLLMs (Gemini 1.5 Pro), exploiting capabilities for simultaneous analysis of textual and visual assets and enabling attribute modulation in real time.

5. Contextual Intelligence, Adaptive Interaction, and Data Management

The evolution of ACAI is projected along three dimensions:

• Contextual Intelligence: Incorporates a contextual memory mechanism storing brand assets, prior design choices, and user feedback. Preference learning and bi-directional feedback loops allow the system to adaptively refine AI-generated suggestions. Temporal weighting employs a decay function $w(t) = e^{-\lambda t}$, where $\lambda$ is a tunable decay rate based on reinforcement of user inputs (Karnatak et al., 9 Mar 2025). This mechanism ensures that more recent or frequently affirmed brand directions predominate, while supporting long-term personalization.
• Adaptive Interaction: ACAI proposes fluid interaction modes adjusting the level of AI support according to user skill, design complexity, and real-time engagement signals (including potential multimodal sensing, e.g., eye tracking or voice cues).
• Data Management: Emphasizes transparent data provenance and local storage, addressing SBO concerns about privacy, ownership, and update control. Users can review and modify the brand-contextual data that feeds into the generative pipeline.

6. Impact on Human-Computer Interaction and Generative AI Design

ACAI advances HCI research by:

• Demonstrating that structured multimodal prompting can significantly reduce the “gulf of envisioning,” enabling conversion of abstract business and creative intuition into robust, machine-interpretable form (Karnatak et al., 19 Apr 2025).
• Challenging dominant text-only paradigms in generative AI, highlighting the inadequacy of free-form prompts for novice users in domain-specific contexts (e.g., advertising), and emphasizing the viability and desirability of co-creative frameworks.
• Supporting inclusive design practices and democratization of advanced generative capabilities for traditionally underserved populations (i.e., SBOs with limited design experience and resources).

7. Future Directions and Limitations

Future enhancements of ACAI articulated in recent research include:

• Integration of real-time feedback loops, continuously updating AI outputs as users adjust panel inputs.
• Transition toward fully automated visual ad generation, bypassing current limitations in multimodal prompting and human–AI handoff.
• More sophisticated adaptive interfaces, potentially leveraging physiological sensing for dynamic assistance.
• Comprehensive personalization and contextual adaptation, predicated on persistent contextual memory and evolving user preference profiles.

A plausible implication is that developing interfaces which foreground user-defined context and provide transparent, iterative control mechanisms could materially improve the utility and adoption of generative AI tools among SBOs and other novice creative professionals.

Table: Key Features of ACAI for SBOs

Interface Panel	Function	Technical Mechanism
Branding	Asset upload and brand value specification	Direct structured input
Audience/Goals	Advertising objective and market segmentation	Text-based targeted input
Inspiration Board	Upload, analysis, and attribute extraction from images	Multimodal LLM and sliders

In summary, ACAI for SBOs represents a validated, structured, multimodal approach to generative advertising that bridges the gap between conceptual brand vision and executable AI-driven content creation, with a technical foundation and empirical support for increased agency and creative control among novice business users (Karnatak et al., 9 Mar 2025, Karnatak et al., 19 Apr 2025).