Role & team shape

Priscilla Spicer leads Product Development, Sourcing & Design at Balsam Brands. 12+ years at Balsam. Team of 5 on the PD side specifically; the wider merchandising team she sits inside is 13–14. Based in Burlingame (the PD and merchandising team site).

She describes her team as one of the tenured ones — “I was there when it was only five of us doing all of this.”

Weekly time breakdown

Roughly 60% of her time on product design — designing SKUs, developing specs, iterating with vendors. The rest is in line-list updates, merchandising review, reviewing vendor samples, and meetings.

She carves out about 2 hours a week specifically to experiment with AI tools. Describes herself as ahead of her team on adoption and actively trying to pull the rest of PD along.

Annual volume / scope

Décor + greenery combined: 250–300 SKUs developed per year. Roughly 100 of those get dropped before launch based on sales signals from the prior season (Thanksgiving through ~December 15 is the read-through window).

Mix: some brand-new products, some line extensions (“two more additional red sizes to an existing red”).

Revenue context: décor and greenery are comparable in size, both significant chunks after trees. Some product lines run small on revenue but high on strategic/marketing value (e.g. PR and BH Studio Hawaii).

Process (her workflow)

Six repeating activities:

1. Trend capture. Team attends trade shows — Christmas markets (January), Christmas World. Takes thousands of iPhone photos. Uploads to Teams or PPT decks. Manual triage after.
2. Mood / theme boards. Manual work from the photo dumps — looking through images, noticing patterns (she called out a blue-and-white trend they built a theme around), building trend boards.
3. Ideation. From the trend boards, what products to develop. “We come in with images. We say: let’s develop a theme in blue and white.”
4. Line list (Excel). Every product in development lands here. Ongoing sheet with item numbers, costs, sourcing country, tariff implications (China vs Vietnam), sample stage (1st / 2nd / 3rd), notes per row. Also holds cross-functional requests (B2B, PR, BH Studio Hawaii) that don’t fit the website-only flow.
5. Vendor spec development. The majority of her time. For each SKU she hand-builds a PPT spec deck, sends to the vendor, gets back their mock-ups, iterates. She described this as “a lot of shipping things back and forth” — she wants to approve via email/picture before any physical samples get made because cross-border shipping is slow.
6. Prep for reviews & paginations. Weekly line-by-line review with the VP on the current line list, plus catalog pagination decisions (does this SKU have a home in the catalog, or is it online-only and therefore subject to MOQ gate).

Decision factors at the pipeline stage: catalog space availability, MOQ for online-only items, real-time sales data (Thanksgiving–Dec 15 crunch), and cross-functional requests (B2B September asks, PR ask-lists, BH Studio Hawaii custom orders).

Current AI usage

• ChatGPT image generation for spec-sheet concepts. Example: a wooden-train tree collar. She fed images of existing Balsam laser-cut decor to establish aesthetic, then prompted for a new piece. Works for a first pass. Falls apart on iteration — regenerations drift back to generic outputs; she can’t reliably edit or update a generated image. “It’s like you have to restart all over again.”
• Translating product specs for non-native-English vendors via ChatGPT. Puts what she wants in, gets an easy-to-read vendor-facing version out. Has helped.
• Testing Claude and Gemini for image generation as a side project.
• Describes herself and her team as roughly “here” on AI skill while the wider company is “here” — she’s actively trying to close the gap by prompting her team, sharing examples, pushing them past “I tried it once and it didn’t work.”

Blockers / pain points

From her PPT (slide 3), explicit in her own framing:

• Image generation quality. Inconsistent and not reliable enough for production-level concepting. No control over exact materials, proportions, details.
• Lack of workflow integration. AI not embedded into existing PD tools (PLM absence is particularly painful — PLM was put on hold). “Feels like a separate tool vs. part of daily workflow.”
• Prompting and skill gap. Team lacks training on how to use AI effectively for PD-specific tasks. Output varies widely by user skill.
• Trust and adoption. Skepticism in her team around accuracy, taste level, usefulness. Not yet seen as “must-use.”
• No standardized use cases. Inconsistent usage across team — some use for copy only, others not at all. No defined best practices or templates.
• Limited customization to brand. Outputs are generic without strong Balsam context. Need grounding in aesthetic, past SKUs, and standards.
• Data isolation. AI not connected to internal data (past SKUs, performance, pricing). Limits ability to give truly actionable recommendations.

Wish-list / what she’d do with more time

From her PPT (slide 2 + 4) and the call:

• Translate trends into SKUs. Turn trend inputs into concrete product concepts, materials, and price tiers. “If I had a magic wand, I’d dump all my photos in one place and it would come up with, oh, based on your photos, I see X, Y, Z concepts.”
• Brand-grounded image generation. Outputs that know Balsam’s aesthetic and its historical SKUs. Usable for production-level concepting, not just first-pass sketches.
• Faster iteration on concepts. Refine without restarting. Edit an image without losing the original.
• Simulate customer perspective. Pressure-test quality, style, positioning before committing to sample.
• Streamline internal alignment. Turn ideas into clear briefs, summaries, rationale for cross-functional teams.
• Reference internal assortment + historical data when using AI. Currently disconnected.
• More competitive intelligence. Would be “out in the market more” if she had time. Physical store visits, vendor insights from other clients, PR editor events. Interested in automated competitor research (Alex raised this in the call — pointed at Clint’s Reddit work and the BMW social-media example).
• More time designing. She’d keep the 60% and grow it. Design more products, closer to what customers actually want, with better feedback loops.

Cross-functional gaps she raised

The strongest cross-functional pain she described — in her own words:

• No unified company calendar. B2B asks for September delivery in the current month. “It needs to be on the boat next month. We don’t even have anything in development.” PD, inventory planning, merch, B2B, marketing, creative — each runs its own calendar. No shared cascade when upstream dates shift (Chinese New Year, licensor approvals).
• Marketing strategy isn’t shared back to PD. She doesn’t know marketing’s brief for the coming season. PD tells marketing what the big bets are; marketing doesn’t tell PD their strategy. Feels asymmetric.
• Unclear handoff once products leave PD. “Once we get the products out the door, I don’t know what happens from now to it’s actually on our site.”
• Multiple line lists across teams. Décor, greenery, trees, GE — each has its own spreadsheet. Tracking cross-team product development is ad-hoc.

Named colleagues she rely on

• Eric Price — R&D, trees. She leans on him for engineering specifics on stability, tree stands, branding. She doesn’t have an engineering degree and uses him as a translator.
• Russ Montano — Customer Service. She talks to him about CS reviews and complaints that drive product improvements. One example she gave: a perennially-popular SKU where she used review feedback to reduce the animation noise on a topper while keeping the light working.
• Kate Hollywood — worked with her team in the PLM evaluation phase before PLM was put on hold.
• Clint (SEO) — historically consulted on product names; she sees the value of tighter SEO/product connection.
• Swati Ayyar — at the PIF → PIM boundary. Priscilla explicitly flagged her as someone who should also be in the cohort.
• The creative team — she noted they’re “using mid journey and all these things” — further along on AI than PD is. Worth cross-pollinating.

Status

In conversation. Not confirmed. Her hesitation was about the time commitment — working through: “we were supposed to have PLM, we would have had to allocate time for that anyway.” Leaning toward participation.

Transcript reference

Granola notes: https://notes.granola.ai/t/85677746-459e-4851-a0da-70449b4aebf3

Meeting date: 2026-04-20

Artifact: the PPT she shared

She shared a 9-slide deck she had built for an internal AI discussion. Key contents:

• Slide 1: her team’s 6-activity workflow (the “Process” section above is drawn from here).
• Slide 2: “End Goals: How AI Should Support Product Development” — the 6-item wish-list above is drawn from this slide.
• Slide 3: “Current Blockers / Gaps” — the 7 blockers above are drawn verbatim from here.
• Slide 4: “What Would Unlock Value” — the unlocker list.
• Slide 5: example trend/mood board (blue & white theme).
• Slide 6: a spec sheet she generated with ChatGPT image-gen for a laser-cut wooden-train tree collar — the running example she used to describe image-gen limitations.
• Slide 7: a snippet of her line-list Excel.

Raw .pptx is not committed to the repo — it’s her internal PD material. Lives in Alex’s Downloads folder.