WysAI · Self-learning

Intent scoring that learns from your conversions

WysAI gives every site its own model. It scores visitors from day one, retrains after every 10 conversions, and only switches to a new model when it beats the default on visitors it hasn't seen. No configuration, no data science team.

Start free trial See the live demo

No cookies
GDPR-ready
No cross-site tracking

WysAI · intent model

Model in use

Trained on your conversions

Beat default on held-out visitors

1Collect
Conversions + non-converters, last 90 days
2Train
Logistic model over 5 behavior signals
3Test
Scored on the newest 25%, held back
4Adopt
Only if it ranks better (AUC) than default

Signal weights

DefaultYour site

Sessions
Pages viewed
Engaged time
Scroll depth
Clicks

Next retrain after 10 new conversionsIsolated to this site

Illustrative weights. Every site's model is different, and no accuracy figure is implied.

The problem

Most intent scores are static. Your visitors aren't.

Every site is different. A score that never learns from your own conversions misses what actually drives them.

One-size-fits-all scoring

Generic engagement rules treat a pricing-page visit on a SaaS site the same as a long read on a blog. What signals intent is different on every site.

Hand-tuned rules go stale

Rule-based scores need someone to pick thresholds and keep revisiting them as products, campaigns and seasons change.

Static models need data scientists

A model trained once and left alone drifts. Retraining, testing and deploying it usually takes a team most sites don't have.

How it works

How WysAI learns over time

Select a stage to see what WysAI does with your conversions.

Signals in the score

Sessionstuned to your sitePages viewedtuned to your siteEngaged timetuned to your siteScroll depthtuned to your siteClickstuned to your site

With at least 15 conversions and 30 non-converters in 90 days, WysAI trains a model for your site.

What happens

Compares visitors who converted with visitors who didn't
Only behavior before a conversion counts
Each signal is weighted by how well it separates the two

Weights stay close to the defaults until your data clearly says otherwise, so a few conversions can't swing the model.

Inside a training run

Learn, test, and only then adopt

Four steps repeat every time enough new conversions arrive.

Step 1 of 4

Start from default weights

Every site starts with the same default weights across sessions, pages viewed, engaged time, scroll depth and clicks, so scoring works from the first visit, before there's any conversion history.

Until your site has enough conversions, scores reflect general engagement rather than what converts for you.

Capabilities

Built to be trusted

One model per site

Each site gets its own logistic model. Your conversions never influence another site's weights, and theirs never affect yours.

Five behavioral signals

Sessions, pages viewed, engaged time, scroll depth and clicks, combined into a 0–1 intent score (high ≥ 0.6, medium ≥ 0.3).

Pre-conversion behavior only

Only what converters did before converting is used, so the conversion itself can't leak into the answer.

Tested before it's trusted

The newest 25% of visitors are held back. A new model is adopted only if it ranks them better (AUC) than the default weights.

Retrains every 10 conversions

New conversions trigger retraining on the last 90 days, so older behavior ages out and the weights follow your traffic.

Visible in your dashboard

Your dashboard shows the held-out test result for your site's latest model, so you can see whether learning is helping.

WysAI vs traditional approaches

Feature	WysAI	Rule-based	Static ML
Adapts to your site
Retrains on your conversions
Configuration	None	Manual rules	Model tuning
Follows recent behavior

What counts as a conversion?

WysAI learns from the events you flag as conversions: the outcomes that matter to your business.

E-commerce: purchases, add-to-cart, checkouts
SaaS: sign-ups, trial starts, activations
Content: newsletter sign-ups, downloads
Custom: any tracked event you flag as a conversion

Example scenario

A SaaS site where readers become buyers

Imagine a SaaS company whose sign-ups tend to come from visitors who read product pages to the end. Out of the box, WysAI weighs all five signals by default. After enough sign-ups, the trained model gives scroll depth and engaged time more weight. It's only adopted because it ranked the newest, held-back visitors better than the defaults.

Example scenario: illustrative, not a customer or measured result.

Weight change after training

Illustrative

Scroll depth▲ Higher
Engaged time▲ Higher
SessionsAbout the same
Pages viewed▼ Lower
Clicks▼ Lower

Privacy

Your data stays yours

Each site's model is fully isolated. Your conversion data never influences another site's model.

Model isolation

Each site gets its own model, trained only on its own visitors.

Behavior, not identity

Signals are sessions, pages, engaged time, scroll depth and clicks. No names or emails.

Anonymous visitor IDs

Visitors are recognized with a random ID stored in the browser. No cookies, same browser only.

FAQ

Frequently asked questions

How does WysAI learn from my conversions?

WysAI fits a per-site logistic model over five behavioral signals: sessions, pages viewed, engaged time, scroll depth and clicks. It compares what visitors who converted did before converting with visitors who didn't, and weights each signal by how well it separates the two.

How many conversions does it need before it starts learning?

Scoring works from day one with default weights. WysAI trains a model for your site once it has at least 15 conversions and 30 non-converters in the last 90 days, then retrains after every 10 new conversions.

What if the trained model is worse than the defaults?

Then it isn't used. Each new model is tested on the newest 25% of visitors, held back from training, and adopted only if it ranks them better (by AUC) than the default weights. Otherwise the defaults stay in place.

How accurate is WysAI?

There are no published accuracy numbers. How well scores predict depends on how many conversions your site records. Your dashboard shows the held-out test result for your own site.

What counts as a conversion?

Any tracked event you flag as a conversion, such as a purchase, sign-up, trial start, demo request or download.

Does my data improve other sites' models?

No. Every site's model is trained only on that site's visitors. Your conversion data never changes another site's weights, and other sites' data never changes yours.

Start learning from your conversions.

WysAI scores visitors from day one and retrains as your conversions come in. 90-day free trial, no credit card.

Start free trial See the live demo