No history, no signal
Outbound-trained models need a long send history to work. For a brand-new contact, they fall back to a global default and guess.
Wave · Channel Affinity
Stop marketing on your history. Start marketing on theirs.
Every ESP optimizes channel mix from the emails you have already sent. Wave trains on how each person actually arrives, their UTMs, referrers, and AI-assistant referrals, and predicts the channel they will answer, even for brand-new contacts.
Quick answer
Channel Affinity is a Wave feature that predicts each person's preferred marketing channel from inbound signals, how they arrived and which referrers and AI assistants sent them, rather than from your outbound send history. It produces two predictions per contact, a preferred inbound channel and a preferred promotional channel. Content format preference is a separate Wave product, Engagement Channel.
Key capabilities
- Inbound and promotional channel prediction per contact
- Trained on inbound signals, not send history
- AI-assistant referral classification
- Per-tenant classification you define and control
- Daily scoring with confidence grades
- HubSpot writeback with audit and rollback
- Per-tenant kill switch
Last updated July 2026
The problem
Channel preference guessed from past sends is blind where it matters most.
Every major channel optimizer, Klaviyo, Braze, Bloomreach, and Salesforce Einstein STO, trains on outbound engagement history. That model has nothing to say about a net-new prospect or an early-funnel contact, exactly where B2B teams struggle most.
AI referrals vanish into direct
Buyers now start in ChatGPT, Perplexity, and Gemini, then arrive on your site. Most stacks bucket that traffic as direct or none and lose the signal entirely.
One channel guess, not two
Where someone consumes content and where they respond to an offer are different questions. Treating them as one number sends the right message on the wrong channel.
How Wave does it
Train on inbound arrivals, predict two channels per person.
Wave reads how every person comes to you and scores the channel they will actually engage, then writes the prediction to the contact record your campaigns already run on.
- 01
Capture every inbound arrival
Wave records each channel touch, UTM source and medium, document referrer, and AI-assistant referral from ChatGPT, Perplexity, Claude, Gemini, and Copilot, as a first-class signal.
- 02
Classify against a canonical taxonomy
Wave maps each touch to a channel in a per-tenant taxonomy you define and control. Classification is accurate and cost-efficient, and you decide how every channel, including each AI assistant, is recognized.
- 03
Predict two channel axes
Wave scores a preferred inbound channel, how each person arrives and discovers you, and a preferred promotional channel, where they respond to an offer, with a confidence grade per person.
- 04
Write it back to your stack
Both predictions sync to HubSpot as contact properties, refreshed daily, governed by a per-tenant kill switch. Your existing sequences route on them with no data engineering.
Where it fits
One axis of Wave's person-level intelligence layer.
Channel Affinity is one part of Wave's person-level data layer. A sibling product, Engagement Channel, predicts the content format each person prefers, so Channel Affinity stays focused on where they arrive and where they respond.
Your sources
Wave
HubSpot
Wave reads from your existing systems through native connectors, runs the model per tenant so your data never trains on anyone else's, and writes two channel predictions back as contact properties. Every writeback is logged with previous value, new value, and the model version that produced it. Rollback is a single audit row.
Works alongside Motion Traction, Buying Groups, The full Wave platform.
Why Wave is different
The only channel model trained on inbound, not outbound.
The competitive line is explicit. Every incumbent learns from what you sent. Wave learns from how people came to you, which is the signal that actually exists for new and early-funnel contacts.
Most tools
Wave
Klaviyo, Braze, Bloomreach, and Einstein STO train on outbound send and open history.
Wave trains on inbound arrivals, so it has signal for net-new and early-funnel contacts.
AI-assistant traffic is bucketed as direct or none and lost.
Wave classifies ChatGPT, Perplexity, Claude, Gemini, and Copilot referrals as first-class channels.
A single channel score per contact.
Two predictions per contact: how they arrive and discover you, and where they respond to offers.
Your data trains a shared, cross-customer model.
Wave trains one model per tenant. Your data never enters a shared model.
FAQ
Questions buyers ask about Channel Affinity.
What is Channel Affinity in Wave?
Channel Affinity is a Wave feature that predicts each person's preferred marketing channel from inbound signals, how they arrived and which referrers and AI assistants sent them, rather than from your outbound send history. It produces two predictions per contact, a preferred inbound channel and a preferred promotional channel, and syncs them to your CRM.
How is Wave different from Klaviyo, Braze, or Einstein STO?
Those tools train on outbound engagement history, what you sent and what was opened. Wave trains on inbound arrival signals like UTMs, referrers, and AI-assistant traffic. That means Wave has signal for net-new and early-funnel contacts where outbound-trained models fall back to a default.
What are the two channel predictions?
Wave predicts a preferred inbound channel, how a person arrives and discovers you, and a preferred promotional channel, where they respond to an offer. Content format preference, such as webinar versus blog post, is a separate Wave product called Engagement Channel.
Does Wave track AI-assistant referrals?
Yes. Referrals from ChatGPT, Perplexity, Claude, Gemini, and Copilot are classified as first-class channels in Wave's taxonomy, instead of being lost in a direct or none bucket like most analytics stacks.
Where do the channel predictions end up?
Wave writes both predictions back to HubSpot as contact properties, refreshed daily. Your existing sequences and workflows can filter and route on them with no data pipeline or engineering work.
Does my data train a shared model?
No. Wave trains one model per tenant. Your contact and engagement data never enters a shared or cross-customer model, and every prediction is stamped with the model version that produced it.
Can I control or pause the writeback?
Yes. Channel Affinity writeback is governed by a per-tenant, per-property kill switch and defaults to off until you opt in, so you can score without writing, then enable writeback when you are ready.
How do I see Channel Affinity on my own contacts?
Book a 20-minute walkthrough. We will configure a tenant against your stack and show the two channel predictions Wave would produce on your real inbound signal.
See it on your data
See your channels predicted from real inbound signal.
Book a 20-minute walkthrough. We will run Channel Affinity against your stack and show the two channel predictions Wave would produce for your contacts.
Request a demo Download the one-pager (PDF)