Data & AI for consumer brands
I ran Data & AI at Jones Road Beauty. Now I’ll run it for yours.
At a nine-figure beauty brand I built the warehouse, the forecasting, the reporting, and the AI tools the team uses every day. I take a few clients at a time.
Certified expert — Polar Analytics · HQ
What I built at Jones Road Beauty
A nine-figure omnichannel beauty brand. Everything below runs in production — the team uses it every day.
Dashboard screenshot
Demand planning & forecasting
Replaced the incumbent planning vendor with a planner built around the team’s own process — live data, edits made right in the tool, reasons saved. A fraction of the vendor’s cost.
Dashboard screenshot
The GTM hub — every launch on one page
Briefs, claims, channel plans, and dates for every launch, current and in one place for the whole marketing team. Status stopped being something you chase people for.
Dashboard screenshot
Alerts that go to one person, with a reason
Slackbots that watch the data and speak up — results vs. plan, contracts about to expire, discount codes near their cap. Each one lands with the person who can act on it, and says why it matters.
Dashboard screenshot
Retail reporting + the daily analyst routine
Point-of-sale, e-commerce, and warehouse data reconciled and delivered every morning — the manual pull that used to eat an analyst’s day now runs itself.
Dashboard screenshot
JRB OS — 35 tools, one front door
The company’s accumulated spreadsheet sprawl, replaced with a single sign-on portal the whole team actually opens.
Dashboard screenshot
The skills library with plugins
A library of reusable AI skills and plugins the team runs on demand — no ticket, no engineer, no waiting.
Also in production: retention and LTV deep-dives · a monthly sales forecaster · vision tools for shade matching and artwork QA · the reconciliation checks that tie retail doors to e-commerce to the cent
You know the symptoms
Four people pull revenue. Four different numbers.
Every report is somebody’s Monday morning.
The forecast lives in one workbook one person can open.
Thirty bookmarked tabs. Nobody knows which is current.
Every one of these has a price. Hours of your best people’s week spent moving numbers between tabs. Decisions that wait until someone can pull the data. The overbuy nobody catches until the warehouse is full, the launch you read wrong for a month, the channel that bled all quarter because the report that would have caught it was never built.
The numbers that looked fine
Some of what I’ve found in production, inside data everyone trusted:
A “reach” metric inflated almost 6× — in the acquisition math for a year.
A silent 1,000-row cap. The charts looked normal, just small.
Shipments triple-counted by status logs. It looked like growth, so nobody audited it.
A traffic sensor offline for weeks, feeding zeros into a conversion rate.
Two definitions of “sales,” 10% apart, used in the same meeting.
A dashboard rendering stale numbers under today’s date.
None of these threw an error. Someone made a spend or inventory decision on every one.
Every one I find becomes a permanent, automated test in your warehouse. The system gets harder to fool every month I’m in it.
Understand first, build second
Most data and AI work fails one of two ways: a strategy deck nobody builds, or a pipeline nobody opens. The fix for both is the same, and it isn’t technical.
Before I build anything, I map how the work actually happens — who touches each number, where it comes from, and who owns each step. About half of what I find shouldn’t be automated at all; it should be deleted. Automating a broken process just makes it break faster.
It’s slower for the first two weeks and faster every week after.
What gets built
The data foundation first, then the AI layer on top of it. All of it ran in production before I ever sold it.
The data foundation
Warehouse builds — Polar Analytics, Snowflake, or BigQuery
Pipelines for Shopify, Amazon, Klaviyo, your ERP, and retail POS
Demand, inventory, and revenue forecasting
Automated daily and weekly reporting into Slack and email
Board packs generated straight from the warehouse
The AI layer
Skills libraries your team runs without you
Custom MCP servers over your systems and vendors
Retrieval assistants in your brand’s own voice
Vision tools — shade matching, artwork QA
Flagship analyses — customer journeys and hero-product economics
Decision queues — alerts with an owner, a reason, and the dollars stated
What happens after it’s delivered
Everything I build follows the same path, so you keep more than a deliverable.
1 · An answer
I run the analysis once. You get the decision, not a deck.
2 · A button
It becomes a skill your team runs on demand — same method, no ticket, no waiting on me.
3 · A signal
The ones worth watching run continuously and alert the one person who owns the call, with the dollars stated. An alert without an owner is spam.
Skills live in a versioned library — when the method improves, everyone’s copy improves.
Two analyses with a decision at the end
Most analysis ends in a chart. These end in rules your team can act on Monday. Both come from work that changed how a nine-figure brand spends.
The Customer Journey Diagnostic
Three years of your order history, read properly: which entry products create second orders, where the journey breaks, and a repeat-purchase definition your finance team will sign.
The Hero Product Deep-Dive
At Jones Road, the product everyone assumed was the retention engine wasn’t — customers who entered through it came back at double the rate of any other entry point. It was retaining the customer, not the product. One SKU, the full workup, and the merchandising rules that fall out.
Delivered as living web reports your team can edit and share — not a PDF that dies in a drive.
Polar Analytics
Certified expert
End-to-end implementation: connectors, custom metrics, dashboards, and the API layer underneath.
HQ
Certified expert
White-glove integration of HQ as your company’s knowledge layer.
In their words
[Placeholder — Cody’s quote about the work goes here. Replace before launch.]
Cody Plofker
Former CEO, Jones Road Beauty
[Placeholder — David’s quote about the work goes here. Replace before launch.]
David Dokes
Co-founder, Polar Analytics
The first ninety days
Every engagement starts with a written roadmap that comes out of the Diagnostic — what we build first depends on what you have and what’s costing you the most. Most first quarters land three things:
One set of numbers
The warehouse live, your sources connected, and metric definitions agreed, written down, and reconciled until they tie to the dollar.
Reporting that runs itself
Daily and weekly reporting into Slack and email, dashboards for each team, and a board pack that assembles itself.
Manual work, automated
Your most repetitive processes automated, and your internal tools behind one login.
At day ninety we set the next roadmap, or I hand everything over, documented. Month to month after the first quarter.
How we’d work together
The Diagnostic
2 weeks
Two weeks to map every source, report, and spreadsheet you have, sit with the people who use them, find where your numbers disagree, and write the 90-day roadmap. Credited in full against your first month.
The Engagement
3-month minimum
Unlimited requests — anything can go on the queue, any time
Two things in flight at a time; you set the order
A written update every Friday, a call every two weeks
No hourly billing, ever. Re-roadmapped quarterly.
I work with consumer brands doing $20M+. What an engagement costs depends on scope and what’s already in place — the Diagnostic is where we work that out.
Who you’d be working with
I’m Ben Rosenwald. I was Director of Data & AI at Jones Road Beauty, where I built everything on this site — the warehouse, the reporting, the forecasting, and the internal platform the company ran on.
I’m a systems thinker before I’m an engineer. The interesting problem is rarely the query; it’s the twelve manual steps around it, and the fact that the person doing them never asked for a dashboard. I take a few clients at a time, deliberately.
Start with the Diagnostic
Two weeks to map what you have and where the numbers disagree. Credited against your first month.