How it works
The engine behind the personalization.
Two things make Fulcrum different from a campaign-based recommendation tool: everything runs all the time, and everything rebuilds itself every night.
- 8
- Algorithms, live from day one Every customer
- 24h
- From new product to live recommendation Nightly recompute
- 0
- Empty recommendation slots Waterfall fallback
- 0
- Pixels to maintain Server-side integration
Coverage
Every algorithm. Every tier. Day one.
No paying more to access more recommendation types. You get the full suite immediately, on every surface that matters.
Live across product pages, cart, category, home, and search.
Same coverage whether you are just getting started or running at full scale.
Pricing scales with traffic, never with how much of the product you are allowed to use.
- Similar Products
- Frequently Bought Together
- Customers Also Bought
- Viewed Together
- Recently Viewed
- Top Sellers
- Abandon Cart Triggers
- Replenishment
Freshness
Rebuilt from scratch, every single night.
Most recommendation relationships are not refreshed on a campaign schedule. They are recomputed in full, overnight, from your latest catalog and purchase data.
New products can appear in recommendations within 24 hours of being added.
Out-of-stock items drop out of recommendations by the next morning.
Price changes automatically reshape what counts as similar, and nobody has to re-run anything.
Powered by a nightly Dataflow pipeline across your full catalog and purchase graph.
Trust
The holdout is the guarantee.
A small, fixed 10% of your traffic never sees personalization - for the entire length of your contract. Comparing that group against everyone else, at the end of your 12-month term, is the entire basis for your guaranteed number.
Runs continuously for the full contract term, not a one-time test.
Calculated once, at the end of the term, using the conservative lower bound of the result.
This is the number behind your guarantee. See the guarantee .
Getting live
Getting live is the easy part.
Connection, configuration, and design review all happen before you ever see a change on your site, so what goes live is already approved, tested, and ready.
-
Connect
A first-party, server-side integration on Shopify Plus or BigCommerce. No pixels, nothing new for your team to maintain.
-
Configure
We apply the full algorithm suite to your catalog and purchase history. Every recommendation type, live from day one.
-
Review and approve
You see exactly where and how recommendations will appear across your site before anything goes live.
-
Go live, then stay current
Every surface activates together, and the entire suite rebuilds itself overnight from there on, so what shoppers see always reflects your latest catalog.
Common questions
How the measurement actually works.
How is the guarantee measured?
It is based on holdout lift: the difference in revenue per user between personalized traffic and a fixed 10% randomized holdout that never sees personalization. We calculate it once, at the end of your 12-month term, at a 95% confidence level, and guarantee the conservative lower bound.
What exactly are you counting?
We measure one thing: the difference in revenue per user between your holdout group and everyone else over the full term. Your guaranteed figure is the conservative lower bound of that range, not a raw total.
How do you know the holdout is trustworthy - couldn't you just call it 'significant' whenever you want?
There is no interim evaluation. We calculate the lift exactly once, at the end of the 12-month term, at a 95% confidence level. A result qualifies when its lower bound is above zero.
Does recommendation freshness cost extra?
No. Pricing is based on monthly page views - a cost-to-serve metric - which is separate from the outcome guarantee. The nightly recompute and full algorithm suite are included at every tier.
What's the difference between what you show me and what you guarantee?
There is one number now. The holdout lift IS the guarantee basis - the holdout is not a separate check on some other figure.
See it running on your own catalog.
The 45-day pilot runs at full fidelity: same algorithms, same nightly recompute, same validation, with nothing owed.
Start Your Free Pilot