Rainbow CDO David Cost on Judging AI by Real Retail Results

Every AI investment clears the same revenue gate

Cost does not treat AI as a special case. Any digital investment faces the same test. Does it add traffic? Does it lift conversion rate? Does it raise average order value? Those are the variables that drive an eCommerce business. The only other thing he will consider is whether the tool substantially cuts expenses, since a cost reduction can pay off like new revenue. He applies that frame to everything, and AI clears it or it does not.

A voice-AI vendor’s numbers did not survive Rainbow’s own data

That discipline exposed a vendor claim last year. A voice AI provider said its agent could resolve 40 percent of incoming customer service calls without a human. Rainbow supervised the rollout and let it ramp. The vendor duly reported 40 percent. Yet Rainbow holds twelve years of agent data, and a real 40 percent resolution should have cut agent hours by a similar share. The team could barely see a difference.

The reason turned out to be a measurement gap. “The only thing we’re able to determine at the end is the AI vendor was measuring tickets,” Cost said. A call that ended with the voice agent counted as closed. What the vendor could not see was the unhappy customer calling straight back to reach a human. So the two sides read the same deployment very differently. The lesson hardened an existing rule. Rainbow now insists on a proof-of-concept before any long-term contract, and it maps out how it will prove ROI before a test even begins.

Push is the one customer channel Rainbow fully controls

Asked for a north-star metric, Cost offered something less obvious. Rainbow runs native apps on iOS and Android, because a real slice of its customers prefer an app to the web. Apps also unlock push notifications, and customers have to opt in. So the team tracks opt-in and opt-out rates as a read on the relationship. A shopper who visits once is one thing. A shopper who returns, downloads the app, and then grants push has taken the relationship up a level.

Push matters because Rainbow owns it outright. With email, Google, Yahoo, Microsoft, and Apple sit in the middle and decide what reaches the inbox. With SMS, the carriers control volume, timing, and price, and can raise that price at will. “No one interferes with push,” Cost said. Once a customer opts in, Rainbow can message them freely. In a Shopify world, apps are cheap to run, so he sees little excuse to skip that direct line.

Generative AI finally makes true personalization possible

In a rapid-fire round, Cost called AI genuinely transformational, and he was specific about why. Conversion rate optimization has stalled. “We’ve hit a ceiling for what we can do with conversion rate optimization,” he said. Online retail sells pictures, since shoppers cannot touch the product. Generative image tools let a team create alternative product images, test one set against another, pick a winner, and repeat until the gains run out. Over a longer horizon, each shopper could see a different set of images tuned to their taste.

He sees the clearest proof in messaging. Rainbow works with Attentive on SMS that writes a unique message for every person. Each one is shaped by what a shopper viewed, how often they visited, and what they added to cart. “It is scary good,” Cost said. The owners of the business once received one and asked who had written it by hand. For years the industry called its work personalization when it was really just segmentation. This, he argues, is the first time the word fits.

Transformation succeeds on org design before tools

The assumption Cost sees fail most often is a focus on software over structure. “They focus on the tool rather than the organizational design,” he said. What decides a project is how you arrange people, incentives, and KPIs around it. Rainbow learned this while shifting from a physical photo studio, with models and retouching, to AI-generated imagery. “Trying to get the organizational pieces correct has taken more time than actually incorporating the technology,” he said. The move rewrote job descriptions, and it demanded a team flexible enough to absorb that.

Rainbow is building AI for the customer the industry overlooks

Coverage favors premium DTC over value retail and the multicultural urban communities Rainbow sells to. He backs the gap with numbers. About a third of Americans hold a college degree, so two-thirds do not. The average Google query runs three words, while the average ChatGPT query runs thirty-two. He doubts that most of the shoppers will write long prompts or turn into expert prompt engineers. So the fixation on agentic shopping and LLMs replacing search misses how most people live. “They’re missing the way the majority of Americans live and work,” he said. The room fights over the top fifth while ignoring the rest.

That view shapes how Rainbow reads AI itself. The team sorts experiments into two buckets. One tries to build a system that replaces a person. The other gives a talented person more reach. Every replacement test failed to pay for itself through headcount. “On the augmentation side, those have all been winners,” Cost said. Generative image tools now free creatives from the budgets and logistics that once capped them, so they can pursue ideas they could never fund before. He is skeptical of the loudest job-loss forecasts. He points to a KPMG survey of 500 CEOs, where only 10 percent planned to cut headcount over their AI use. His advice: learn the tools, play with them early, and treat the moment the way early adopters once treated the spreadsheet.