Intelligence Suite

Brands and distributors use Salsify as a source of truth for the ecommerce product data and assets they need to sell across online retailers. As generative AI became more prevalent, we saw an opportunity to automate time-consuming manual work and help customers work more efficiently.

The business wanted to bring AI capabilities to market quickly and establish Salsify as a leader in the space. Engineering had already demonstrated what was possible through an internal proof of concept (POC), and our team was tasked with turning it into a production-ready MVP. I led the UX strategy, discovery, and design required to create a customer-ready experience.

The result was Intelligence Suite, which embedded AI capabilities into Workflow, Salsify’s tool for automating product data processes through repeatable steps, rules, and human review. Customers could generate, extract, and optimize content at scale while reviewing results before committing changes to their data.

The POC proved the technology worked, but the experience reflected its technical architecture rather than customers’ mental models. Configuration was considerably longer than existing Workflows, with dense settings, technical terminology, and controls whose effects weren’t always obvious. Output quality also depended heavily on the prompts customers wrote—another hurdle for users who weren’t AI or Workflow experts.

We couldn’t redesign Workflow from scratch. Intelligence Suite had to fit its existing patterns and technical constraints without disrupting established customer processes.

The stakes were high: inaccurate AI outputs could affect product data at scale, introducing customer trust, financial, and regulatory risks.

Before deciding what belonged in the MVP, I needed to understand the system myself. Configuration changed by AI task (such as text generation or image analysis), the product data or assets involved, and dependencies among settings. Even after exploring the POC, some relationships remained unclear.

I mapped these scenarios to identify what users needed to configure and where we could simplify the experience.

With roughly five months to launch, we couldn’t assume every POC capability belonged in the MVP. I led a detailed review with product and engineering, evaluating configuration options one by one and drawing on feedback from customer-facing teams.

The instinct was to offer maximum flexibility, but several advanced controls were difficult to explain, even internally, and it wasn’t clear when customers would benefit from changing the defaults. We chose a simpler starting experience, deferring options whose complexity outweighed their immediate value. This was applicable for sections like debugging responses, response formats and context keys, and defining custom review behavior other than reviewing for suggestions or directly editing.

Once we’d clarified the MVP scope, I reorganized configuration within Workflow’s existing UI patterns to follow the order in which customers would complete their tasks.

The result was a more focused configuration experience that preserved necessary flexibility without exposing every technical detail up front.

Simpler configuration wasn’t enough. Because AI could change large volumes of product data, customers needed ways to understand recommendations and control when changes became final.

Even a clearer setup couldn’t solve a fundamental challenge: most customers weren’t experienced prompt writers. Starting from a blank field would slow adoption.

For the MVP, our team turned 13 high-value use cases (previously identified through earlier customer and stakeholder input that went into creating the POC) into pre-packaged AI Accelerators. We created, refined, and tested prompt templates so customers could begin with useful configurations rather than build workflows from scratch. We also designed prompt writing best practices, enablement on how to use each accelerator, and how to ensure users took advantage of the functionality.

One example was content optimization. Customers with small content teams—or lengthy review processes—could use an Accelerator to adapt existing marketing copy for seasonal campaigns, shopping events, or different audiences instead of repeatedly rewriting it by hand.

The goal was to give customers a faster path to value while helping them learn how to build and adapt their own AI workflows over time.

The MVP launch began an ongoing cycle of discovery and prioritization. Image generation was emerging as a likely next investment: customers and Sales teams were asking for it, and our team’s model testing was producing more promising results.

Ahead of our annual customer conference, I facilitated a user-story mapping workshop with product and engineering to surface dependencies, align on the end-to-end experience, and decide what we could prototype for the next milestone.

We shared future concepts with hundreds of customers at the conference, alongside continued customer interviews and feedback from customer-facing teams. That input helped us create 6 additional image-generation accelerators around tasks customers were otherwise handling manually or outsourcing to agencies: tasks like creating lifestyle images, image variants, hero images for multipacks and highlighting attributes, and infographics.

We also had to adapt the roadmap as AI capabilities and requirements evolved, including unplanned compliance work related to the EU AI Act to add visible watermarking and filters for AI-generated images.

Intelligence Suite became Salsify’s highest-selling module to date. Customers also reported meaningful time savings from AI-powered workflows, giving them more time for higher-priority work.

Our usage dashboard suggested that most customers weren’t using Accelerators at all, which seemed unlikely. To find out what was really happening, I analyzed customer configuration and prompt data using Claude and internal data tables to see how customers were applying AI in practice.

Approximately 80% of workflows aligned with our 19 existing Accelerators. This supported our initial selection of use cases while revealing emerging patterns to monitor for potential future templates.

The analysis also pointed to a gap in how usage data was being captured, which I flagged to our data team for further investigation.

Embedding AI into Workflow delivered value quickly, but configuration still required more time and product knowledge than I wanted. Based on my own experience and recurring snippets of customer and stakeholder feedback, I suspect the first successful setup remained a major hurdle.

I’d explore a more guided, hands-on experience that walks customers through creating their first high-value AI workflow, then helps them confidently customize and build their own.