How Routeflow's Growth Team Gets 3× More Creative From Every Sprint

It is 4:47 p.m. on a Thursday sprint close and Priya Mehta is scrolling back through a ChatGPT conversation that is now 34 messages long, looking for the version of the blog cover image she used three weeks ago. It isn't there — or rather, it's in there somewhere, sandwiched between a thread about homepage copy and a request for a LinkedIn caption, already indistinguishable from the six other blue-accented images the chat history shows. She needed it as a reference for the campaign that's going live Monday. She starts the prompt again from scratch.

Mehta is Head of Growth at Routeflow, a route-optimisation software company based in London. The team is twelve people. The marketing function is three: Mehta, a content lead, and a part-time designer who is booked exclusively for product illustrations and sales collateral. Paid social creative, blog covers, and LinkedIn posts run through whoever has time, and for the past eighteen months that workflow ran almost entirely through ChatGPT.

The problem with using a chat interface for production image work is that chat was never designed to be a production workflow. Images live in threads with no library, no version history, and no way to return to a specific asset without finding the specific conversation it appeared in. Brand consistency depends entirely on how precisely you reconstruct the original prompt — a constraint that scales poorly across a three-person team where not everyone writes prompts the same way.

The result. Priya Mehta switched Routeflow's growth team to Playyy's AI Image Generator, AI Image Editor, and Background Remover. Ad creative output rose from 4 to 12 variants per two-week sprint. Per-image production time fell from 18 minutes to 6. Brand inconsistency — the most persistent complaint from the sales team about the team's self-produced creative — was eliminated within the first sprint.

The Real Cost of ChatGPT Image Generation for a Marketing Team

The issues Routeflow experienced with ChatGPT's image generation were not about quality. The outputs were good enough for paid social, often better than what a stock photo search would have returned. The issues were structural.

First, there was no asset library. Every image lived in a chat thread. Finding a specific creative required remembering which conversation it appeared in. Recreating a specific look required reconstructing the original prompt from memory, which produced a result that was similar but never identical.

Second, there was no brand kit. ChatGPT has no persistent memory of brand colours, tone, or visual style between sessions. Each new conversation started from zero. The result was that Routeflow's blog covers, LinkedIn posts, and paid social images shared a general aesthetic — "professional, tech-adjacent" — but not a consistent one. The accent colour the team called "Routeflow blue" appeared in six slightly different values across the quarter's creative output.

Third, there was no spec control on export. ChatGPT exports images as generic JPEGs at whatever resolution the model produces. LinkedIn expects 1200×628. Google Display wants 300×250 and 728×90. Getting from a ChatGPT output to a platform-ready asset required opening a second tool — usually Canva or Figma — to crop and resize. For a team producing four creative variants per sprint, that second step added fifteen minutes per image.

According to a 2025 Sprout Social report, 73% of B2B marketing teams identify creative production speed as a primary bottleneck in campaign execution. For a three-person growth function without a dedicated designer, every minute of friction in the image workflow is a direct constraint on how much the team can test.

Why Playyy's Workflow Is Different

The shift Mehta noticed immediately when she opened Playyy was not the image quality — it was the architecture. Playyy is not a conversation interface that produces images. It's a production environment where images are assets you manage, edit, and export to spec.

The first sprint on Playyy looked different from minute one. Mehta uploaded Routeflow's brand hex values into the brand kit. Every generation session that followed defaulted to those values — no prompt engineering required to maintain consistency. When the content lead generated a blog cover independently, it matched the paid social creative Mehta had produced the day before. That had not happened once in eighteen months of ChatGPT use.

The AI Image Generator produces images into a persistent library, not a chat thread. Returning to a specific asset means opening the library, not scrolling through a conversation. For a team producing creative across a two-week sprint, the library alone eliminates the retrieval problem that cost Mehta's team an estimated 20 minutes per sprint just in asset archaeology.

In our experience with SaaS growth teams transitioning from chat-based image tools, the brand kit and library are the two changes that produce the most immediate measurable impact. Quality differences between AI image generators are often marginal at the output level. Workflow differences — persistent brand context, organised assets, spec-controlled export — compound across every image the team produces.

From 4 Variants to 12: The Edit-Elements Workflow

The output volume change came from a specific workflow shift. Under the ChatGPT model, producing a variant meant regenerating from a new prompt. Each prompt produced a different image — different composition, different lighting, different background, sometimes different colour interpretation. Four variants meant four full generation passes, each starting from scratch, each requiring a separate export-and-resize step.

On Playyy, a variant means using the AI Image Editor to make targeted changes to an existing base image. Generate once, then edit specific elements with a prompt: change the background from a neutral grey to a dark navy. Shift the accent colour from the default teal to Routeflow blue. Swap the foreground subject against a clean white field using Background Remover for a version that works on a coloured ad background.

Each edit takes a focused prompt and produces a result in 30 to 60 seconds. The composition, lighting, and core subject of the image stay intact. Only the targeted element changes. From one strong base image, Mehta's team now reliably produces six to eight variants — background swaps, colour shifts, subject isolation — before the creative starts to diverge meaningfully from the source.

For a two-week sprint with two base images and two campaign angles, twelve variants is now a floor, not a ceiling.

What the Numbers Look Like After One Quarter

Routeflow ran three full two-week sprints on Playyy before Mehta pulled the numbers for her quarterly marketing review.

Creative variant output moved from 4 per sprint to a consistent 11 to 13. The per-image production time — measured across all three team members who contribute to creative — dropped from 18 minutes to 6. The 18-minute figure included the prompt iteration, the export from ChatGPT, and the resize-and-crop step in a second tool. The 6-minute figure includes generation, one edit pass using the image editor, and a spec-controlled export directly to the platform format.

The brand consistency issue resolved itself. With a shared brand kit, every team member's outputs defaulted to the same colour values. The sales team stopped flagging inconsistent creative in the Monday channel review — which had been a standing item for six months.

The campaign volume increase had a downstream effect Mehta had not anticipated: the team now has enough creative to run genuine A/B tests on a consistent basis. At four variants per sprint, every variant had to run to get any data. At twelve, the team can split-test two approaches against each other and retire underperformers mid-sprint without running out of creative.

For more on the broader shift from single-asset production to variant-based creative testing, see Campaign Visual Creation From Brief to Every Channel and Marketing Creatives.

What the Workflow Looks Like Now

Sprint planning now includes a creative brief that specifies two base image directions and a target variant count — typically twelve across both directions. Mehta generates both base images on Monday morning, usually in under 30 minutes combined. The content lead runs the edit-element passes for background and colour variants through the week as posts are scheduled. Background Remover handles the cases where a subject needs to run on a coloured ad background without a competing environment behind it.

Export is the last step and the one that previously required a second tool. Platform-spec export — 1200×628 for LinkedIn, 1080×1080 for feed, the correct dimensions for Google Display — happens inside Playyy. The resize step that used to take five minutes per image is gone.

The Thursday sprint-close image archaeology session — the one that started this story — no longer exists. The asset from three weeks ago is in the library, tagged to the sprint it was produced in, available in 30 seconds. The prompt that generated it is attached. The brand context that shaped it is already set for the next generation.

For a twelve-person company with a three-person growth function, that time goes directly back into the work.