CWX
Talk to us
A woman in a tweed jacket smiling in an office, with a colleague holding glasses in the foreground.
Solutions · Personalization

Higher conversion. Longer relationships.

Models trained on your own data, turning customer signals into insight.

The customers most likely to buy again are the ones you already know. We put what you know about each of them to work, so every offer, product and conversation fits, measured in the numbers your teams already report.

Published work in B2B marketing and retail, including an early in-store pilot.

Talk to us about conversion See the results
Client results

Relevance clients can measure.

Each result below comes from the client's published case study.

Figures as published in each client's case study.

Why it matters

Every customer sees the same page, and every campaign starts from instinct.

Growth from existing customers is a data problem before it is a marketing problem.

The signals are already in the business: what each customer browses, buys, clicks and ignores. The structure keeps them apart. Customer data sits with one team, creative with another, the store with a third. Each does its part well, and the customer still sees something generic.

None of that is a lack of ideas or effort. It is the shape of the work. We've seen conversion and retention move when three things are built to work together: signals about each customer, models that choose for them, and a metric the business already trusts to say whether it worked.

  • Same page for everyone
  • Creative chosen by feel
  • Variants slow to produce
  • Answers spread across systems
What you're hearing

The words that come before this work.

  • We need more from the customers we already have.

    Growth has to come from deeper relationships, not only from new accounts.

  • Every customer sees the same thing.

    The data to tailor the experience exists, but it is not connected to the moment of choice.

  • We're bringing marketing in-house.

    The team needs its own data, models and measures, not another agency report.

  • Show me it in revenue, not in a dashboard.

    Leaders want a result in a number finance already tracks.

How value is earned

From a use case to a number the P&L can feel.

Use case

Recommend a product. Pick the best creative. Answer a shopper's question.

tied to an outcome before anything is built

Outcome

Conversion rate. Repeat purchase. Revenue per customer. Time to sale.

What we measure

  • Click-through and conversion on the moment we change
  • Repeat purchase and retention over the period the business reports
  • Revenue per customer against a matched baseline
  • Effort behind each campaign, so the gain is not bought with more work
  • Time to sale where staff serve the customer

Each measure is agreed with finance before the first experiment, so marketing and the CFO read the result the same way.

What we build

Four parts, one loop.

Personalization works when the signal, the model, the content and the measure are built as one loop rather than four projects.

  • Customer signals

    Behavior, preferences and past response, brought into one shared view of each customer, with consent and access rules attached.

  • Models that choose

    Recommendation and propensity models that pick the product, offer or next step most likely to land.

  • Creative chosen by performance

    Agents that learn from past click-through and image features to suggest the asset most likely to work.

  • Answers at the point of sale

    Assistants that give store and service staff the answer a customer needs, from every system, in one place.

  1. Signals
  2. Models
  3. Content
  4. Experience
  5. Measure

every result trains the next choice

BigQuery and BQML, Vertex AI, recommendation models, retrieval workflows.

What we've seen

Creative chosen by performance, not by feel.

A product and marketing team reviewing creative performance.17%increase in CTR20%less consulting time

A marketing platform with 1k+ global customers.

The challengeSelecting and optimizing creative for account-based marketing was manual, slow to market and subjective.What we builtA dual-agent system that learns from historical CTR data and image features to suggest assets autonomously.What changedClick-through rose 17%, and consulting time fell 20%.AI Engineering · Data AI/MLRead the Demandbase story

Results as published on each case. Levi's is an early pilot.

Control

Relevant to each customer. Responsible with every record.

  1. 01

    Consent travels with the data.

    Customer data carries its classification and permitted uses, so a model only sees what it is allowed to use.

  2. 02

    Brand stays in your hands.

    Creative and copy guardrails are set by your brand owners before anything is generated or chosen.

  3. 03

    A fair comparison.

    Every change is measured against a baseline the business agreed, so a lift is a lift and not a seasonal accident.

  4. 04

    People stay in the loop.

    Store and service staff use the assistant's answer; they still own the conversation.

Who's in the room

Your people and ours, one outcome.

You bring
  • the CMO or growth lead who owns the number
  • marketing, merchandising or store leaders
  • brand owners
  • customer and performance data
  • a finance sign-off on the baseline
CWX brings
  • Data AI/ML and AI Engineering
  • the Transformation Office for AI enablement across marketing and store teams
  • evaluations and runbooks
Daniel JoanesSr. Director Data and AI

Leads Data AI/ML: data strategy, one shared meaning for your data, and the models built on it.

Data AI/ML
Industries

Where we work.

Retail

Every shopper sees what fits them, and an early in-store pilot puts answers in stylists' hands.

ProofHuckberry · Levi's (early pilot)

Technology and software

Creative chosen by performance across account-based campaigns.

ProofDemandbase

Working in another industry.Talk to us

How we report it

Read in the metric you already trust.

  • Realized

    It met or exceeded the agreed threshold against its baseline, and your finance team has agreed the readout.

  • Realized with feedback

    It moved the outcome but has not yet reached the threshold. The team names the change it will make and the date of the next readout. If the threshold is not met by then, it is retired.

  • Retired

    It did not move the outcome within its time box, or its stop rule was triggered. The owner makes the call, and the work stops on purpose.

Every experiment reports against the conversion or retention measure agreed with your finance team. A pilot stays labelled as a pilot until its readout says otherwise.

Questions

Frequently asked questions

Something else on your mind? Ask us directly

  • Where your governance says it should. We design around your access rules and consent, not around ours.

Get started

Grow the customers you already have.

Tell us the customer moment and the number you report on it. We will show you what could move it, and how we would prove it.

Start the conversationTalk to us about conversionGo furtherRead the Demandbase story

Prefer email? sales@cwx.tech