Every vendor sells ranking. Here is what else they sell, and for how much.
We read five vendors’ own docs, pricing pages and terms and wrote down what we found, with the page for
every cell. Where a vendor does not publish something, the cell says so; it is not a “no”.
What only Patterna puts in the base plan: a holdout ledger that proves the uplift, a reason on every slot, and consent
modes that store no user id when the user said no.
Ten criteria, six products
Open a row for the sentence and the source behind each verdict.
Feature comparison, with the source for every cell
Criterionwhat we checked
Patternathis site
Recombee
AlgoliaRecommend
AmazonPersonalize
Google CloudVertex AI Search for commerce
Crossing MindsBeam
A control group that keeps seeing your own list, and the uplift with a confidence interval, reported by the product itself.
✓Yes
◐Partial
◐Partial
✕No
◐Partial
◐Partial
✓YesPatterna
Every placement runs a sticky holdout; the ledger reports the difference with a 95% interval, clustered on the user, in the base plan. patterna.nl
◐PartialRecombee
Admin UI A/B tests split traffic between a control scenario and variants and report significance and “probability to beat baseline”; no holdout without recommendations and no interval on the uplift is documented. recombee.com
◐PartialAlgoliaRecommend
Dashboard A/B tests with a credible interval exist for search indices and configurations; Recommend analytics compares periods, with no control group or lift for recommendations documented. algolia.com
✕NoAmazonPersonalize
A/B testing is described as a workflow you run with your own tool (CloudWatch Evidently is suggested); metric attribution reports event totals per recommender, no control group or interval. aws.amazon.com
◐PartialGoogle CloudVertex AI Search for commerce
Traffic is split with a third-party experiment platform; the console then shows lift with a 95% confidence interval for search and browse metrics. No equivalent for recommendations is published. cloud.google.com
◐PartialCrossing MindsBeam
A/B tests with a 50/50 split against a “custom” group are documented, with lift and a p-value in PoC and monthly reports; no holdout against your own list in a live panel, no interval. web.archive.org
The response says in words why each item is there.
✓Yes
✕No
✕No
◐Partial
✕No
✕No
✓YesPatterna
Each slot carries a reason key and a sentence in Dutch and English from the generator that won. patterna.nl
✕NoRecombee
Responses return item ids (optionally with properties); “Because you watched” rows are built client-side from a source item, no reason field. recombee.com
✕NoAlgoliaRecommend
A recommend hit carries objectID, ranking info and a 0–100 score; no reason field in the OpenAPI spec. algolia.com
◐PartialAmazonPersonalize
User-Personalization-v2 returns an enumerated `reason` (Promoted item, Exploration, Popular item); descriptive “themes” exist for batch Similar-Items only, in three non-EU regions. aws.amazon.com
✕NoGoogle CloudVertex AI Search for commerce
Predict results carry an id, optional product object and an optional score; no reason field. “Search explainability” is console-only and search-only. cloud.google.com
✕NoCrossing MindsBeam
The recommendation response is a list of item ids with warnings and the scenario that applied; no per-item reason. web.archive.org
A documented mode that stores no user id, and a published rule on personal data.
✓Yes
◐Partial
◐Partial
◐Partial
✓Yes
◐Partial
✓YesPatterna
personalized, cohort and non_profiled; non_profiled stores no user id. A personal field in a payload rejects the call. patterna.nl
◐PartialRecombee
Anonymous users use a session id as userId and are merged on login; a user id is always stored. Terms say personal data processing is not part of the service. recombee.com
◐PartialAlgoliaRecommend
A userToken is sent with every event; it must not contain PII and may be a per-session anonymous token. The DPA lists IP address and userToken as end-user identifiers. algolia.com
◐PartialAmazonPersonalize
Anonymous events with a sessionId are documented, but are not used for training until tied to a userId; the data-protection page warns against PII only in tags and free-text fields. aws.amazon.com
✓YesGoogle CloudVertex AI Search for commerce
visitorId must not contain PII; userId is optional and must be hashed; the implementation guide makes the customer responsible for redacting PII. cloud.google.com
◐PartialCrossing MindsBeam
Anonymous session recommendations are documented and the marketing says “without using personal data”, but the user catalog invites age, gender and location properties; no consent mode. web.archive.org
The customer can keep all data in EU regions.
✓Yes
✓Yes
◐Partial
✓Yes
✕No
–Not published
✓YesPatterna
Amsterdam only; a data processing agreement with every paid plan. patterna.nl
✓YesRecombee
Databases are created in a region; eu-west is one of four (US, Canada, Europe, Australia). recombee.com
◐PartialAlgoliaRecommend
Hosting locations US, UK and EU West on Grow plans; the DPA keeps end-user data in the EEA only when all services stay in-region, with an exception for IP addresses on a security watchlist. algolia.com
✓YesAmazonPersonalize
Available in eu-west-1 (Ireland) and eu-central-1 (Frankfurt); data stays in the region used. aws.amazon.com
✕NoGoogle CloudVertex AI Search for commerce
“A global service, independent of location”; endpoints use locations/global and no EU region is published for this product. cloud.google.com
–Not publishedCrossing MindsBeam
The privacy policy speaks of transfers to the United States under Privacy Shield; no region option is published. web.archive.org
Business users read results and manage lists without code.
✓Yes
◐Partial
✓Yes
✕No
◐Partial
✓Yes
✓YesPatterna
The panel opens in Simple mode: placements, slates and proof; Advanced adds flows, audiences, Lab and governance. patterna.nl
◐PartialRecombee
Admin UI with KPI console, scenarios and no-code widgets; custom rules are written in ReQL. recombee.com
✓YesAlgoliaRecommend
Models are trained in the dashboard; Recommend rules are if-then statements “without editing code”. algolia.com
✕NoAmazonPersonalize
AWS calls it the “developer console”; filters and promotions are configured there, no business-user UI is published. aws.amazon.com
◐PartialGoogle CloudVertex AI Search for commerce
A merchandising console with proposal and approval flow for controls, documented for search; for recommendations the only control is boost/bury. cloud.google.com
✓YesCrossing MindsBeam
Beam Studio: a QA dashboard, A/B testing, analytics, cohorts and KPI tracking for business users. web.archive.org
Editors pin, boost, cap or exclude items.
✓Yes
◐Partial
✓Yes
◐Partial
◐Partial
◐Partial
✓YesPatterna
Editors pin, cap per collection and set premium rules; a live change needs two approvals and every flow version rolls back. patterna.nl
◐PartialRecombee
Boosters, filters and constraints per scenario in the Admin UI; no pin-to-position is documented. recombee.com
✓YesAlgoliaRecommend
Recommend rules pin to a position, hide, boost, bury and filter; the number of rules depends on the plan. algolia.com
◐PartialAmazonPersonalize
Filters include or exclude items by expression; promotions reserve a share of slots for a category, positioned randomly. No pin or per-item boost. aws.amazon.com
◐PartialGoogle CloudVertex AI Search for commerce
Boost/bury for recommendations is pre-GA; pinning and filter controls are search-only. Per-request filters by tag or attribute. cloud.google.com
◐PartialCrossing MindsBeam
Filters and exclusions by rule; boost (“bonus/malus”) was marked “coming soon” and the docs say rules are typically configured by the vendor’s ML engineer. web.archive.org
A named engineer on the account as a product offer, not generic support.
✓Yes
◐Partial
◐Partial
✕No
✕No
✓Yes
✓YesPatterna
The Managed plan: a named ML engineer, weekly monitoring, monthly improvement cycle with its holdout result. patterna.nl
◐PartialRecombee
“Data Science Support” is a row on the Premium plan (from $4,499/mo); account manager and Slack channel from Pro. recombee.com
◐PartialAlgoliaRecommend
Professional services by the hour (advisory, solution architect, onboarding) at $275 an hour; no data-scientist offering. algolia.com
✕NoAmazonPersonalize
No engineer comes with the service; AWS sells consulting separately through Professional Services and partners. aws.amazon.com
✕NoGoogle CloudVertex AI Search for commerce
No engineer comes with the service; Google Cloud consulting is a separate engagement. cloud.google.com
✓YesCrossing MindsBeam
“A dedicated team comprised of at least 1 ML engineer, 1 data scientist, and your success manager”; every plan lists a dedicated AI engineer. web.archive.org
Start without talking to sales.
✓Yes
✓Yes
✓Yes
✓Yes
✓Yes
✕No
✓YesPatterna
60 days, one placement, shadow mode, a holdout report at the end; a form, no card. patterna.nl
✓YesRecombee
30-day unlimited trial, then a free plan with 100k requests a month. recombee.com
✓YesAlgoliaRecommend
Free plan, no card: 5k recommendation requests a month; Grow includes 10k before billing. algolia.com
✓YesAmazonPersonalize
Two months free from first use: 50k real-time requests a month on v2 recipes, 20 GB, 5M interactions for training. aws.amazon.com
✓YesGoogle CloudVertex AI Search for commerce
$600 in recommendation credits granted on signup, expiring after six months; setup is self-serve in the console. cloud.google.com
✕NoCrossing MindsBeam
A 30-day trial existed behind a “Get a demo” form; the team audits your data first. No self-serve signup. web.archive.org
A public price list.
✓Yes
✓Yes
✓Yes
✓Yes
✓Yes
✕No
✓YesPatterna
Five plans with prices, included volumes and overage rates on one page. patterna.nl
✓YesRecombee
Free, Standard $99, Pro $1,699, Premium from $4,499; Platinum on request. recombee.com
✓YesAlgoliaRecommend
Free, Grow and Grow Plus are published; Elevate is custom and annual. algolia.com
✓YesAmazonPersonalize
Per-GB ingestion, per-interaction training and $0.15 per 1,000 requests on v2 recipes. aws.amazon.com
✓YesGoogle CloudVertex AI Search for commerce
$0.27 per 1,000 predictions up to 20M a month, then $0.18 and $0.10; training $2.50 a node hour. cloud.google.com
✕NoCrossing MindsBeam
Three tiers, each “Get a Quote”, revenue-share by default. The one published number is the Shopify listing: $60,000 a year. apps.shopify.com
A REST API and official client libraries.
✓Yes
✓Yes
✓Yes
✓Yes
✓Yes
✓Yes
✓YesPatterna
REST API from one OpenAPI file; TypeScript, Go and Python SDKs. patterna.nl
✓YesRecombee
REST API; server SDKs for Java, Ruby, Node.js, PHP, Python, .NET, Go; client SDKs for JS, Kotlin, Swift. recombee.com
✓YesAlgoliaRecommend
REST API with a public OpenAPI spec; clients in C#, Dart, Go, Java, JavaScript, Kotlin, PHP, Python, Ruby, Scala, Swift. algolia.com
✓YesAmazonPersonalize
REST API and AWS SDKs for .NET, C++, Go, Java, JavaScript, Kotlin, PHP, Python, Ruby. aws.amazon.com
✓YesGoogle CloudVertex AI Search for commerce
REST API (retail.googleapis.com v2) and client libraries for C++, C#, Go, Java, Node.js, PHP, Python, Ruby. cloud.google.com
✓YesCrossing MindsBeam
REST API with JWT; client libraries for Python, JavaScript, Node, Java, .NET, Ruby, PHP on GitHub. github.com
Crossing Minds. crossingminds.com does not answer at the time of writing; its last archived homepage is a letter saying the team is joining OpenAI. The rows below read from archived vendor pages and the live Shopify listing, and describe the offer as it was published.
List price at 500k, 3M and 10M requests a month
Computed from each vendor’s published rates, request charges only. USD prices are shown as the vendor publishes them; Patterna is in EUR. Nothing is converted, so read the bars as list prices in two currencies, not as one scale. The table has the formula and what each number leaves out.
500k requests a month
Patterna€149
Recombee$379
Algolia$294
Amazon$75
Vertex$135
Crossing Minds$5,000*
3M requests a month
Patterna€799
Recombee$824
Algolia$1,794
Amazon$450
Vertex$810
Crossing Minds$5,000*
10M requests a month
Patterna€2,199
Recombee$2,699
Algolia$5,994
Amazon$1,500
Vertex$2,700
Crossing Minds$5,000*
Patterna, EUR (hatched) Others, USD as publishedEach panel has its own scale. Request charges only; what each number leaves out is in the table below. *Crossing Minds: the one published figure, $60,000 a year, flat.
Price formulas and exclusions per vendor
Vendor
500k
3M
10M
Formula
Not included
PatternaStarter €149 with 500k requests, Growth €799 with 3M, overage €0.30 and €0.20 per extra 1,000.
€149
€799
€2,199
the cheapest self-serve plan at that volume: Starter at 500k; Growth at 3M; Growth + 7,000 × €0.20 at 10M
Nothing: ledger, reasons, consent modes and EU hosting are in the base plan. VAT excluded. source
RecombeeStandard $99 with 100k requests, +$0.70 per extra unit of 1,000 ($0.25 once usage passes 1M); Pro $1,699 with 5M, +$0.20.
$379
$824
$2,699
500k: $99 + 400 × $0.70; 3M: $99 + 2,900 × $0.25 (the page says the unit drops to $0.25 past 1M; read as applying to all units, the cheaper reading); 10M: Pro $1,699 + 5,000 × $0.20
Interactions, active users and catalog items consume the same extra units, so a real bill is higher; add-ons (semantic search) unpriced. source
AlgoliaRecommendGrow: 10k recommendation requests a month included, then $0.60 per additional 1,000.
$294
$1,794
$5,994
(requests − 10,000) / 1,000 × $0.60
Search requests, records above 100k, crawls; Elevate volume discounts are quote-only. source
AmazonPersonalizev2 recipes: $0.15 per 1,000 recommendation requests, flat; plus training and ingestion.
$75
$450
$1,500
requests / 1,000 × $0.15
Training ($0.002 per 1,000 interactions), ingestion ($0.05/GB), and a minimum of 1 request per second billed per active campaign, which alone is about 2.6M requests ($390) a month, so the 500k figure is below the published floor. source
Google CloudVertex AI Search for commercePredictions $0.27 per 1,000 up to 20M a month; training and tuning $2.50 a node hour.
$135
$810
$2,700
requests / 1,000 × $0.27 (all three volumes sit in the first tier)
Training and tuning node hours (Google’s own example adds $450 a month for one daily-retrained model), log storage; no EU residency to buy. source
Crossing MindsBeamQuote-only, revenue share by default; the Shopify App Store listing is a flat $60,000 a year with a dedicated ML engineer.
$5,000
$5,000
$5,000
$60,000 a year ÷ 12, the only published figure, flat at every volume
Whether the flat fee has a request cap is not published; the vendor site is unreachable. source 1·source 2
Managed against managed
The one other offer with an engineer on the account is Crossing Minds. Here is what both say they include.
Patterna Managed
€2,900 a month · €4,900 setup once · 12-month term · €39,700 first year
Everything in Growth
A named ML engineer on your account
Custom algorithm for your catalogue
Weekly monitoring
Monthly improvement cycle, each with its holdout result
Quarterly roadmap call
10M requests a month
Stop rule: no uplift in the holdout after 90 days, you leave
EU hosting, data processing agreement, ledger export at any time
$60,000 a year on the Shopify listing · $5,000 a month · tiers otherwise “Get a quote”
“At least 1 ML engineer, 1 data scientist, and your success manager”
A/B test and proof of concept with lift and a p-value in reports
Beam Studio dashboard: QA, analytics, cohorts, KPI tracking
Revenue-share pricing by default; cost per recommendation set as fallback
Hosting region not published; 30-day trial behind a demo form
crossingminds.com does not answer at the time of writing; its last archived homepage is a letter saying the team is joining OpenAI. The rows below read from archived vendor pages and the live Shopify listing, and describe the offer as it was published.
Where the others are the better fit
Very high-volume commerce on an existing AWS or GCP commitment
Amazon Personalize’s $0.15 per 1,000 requests is below our Growth overage of €0.20 at any volume, and Vertex’s rate drops to $0.18 past 20M predictions a month; both count against a cloud commitment you already signed. You bring your own A/B tool, your training budget and your own engineer.
Shops already on Algolia search
If your catalog, rules and merchandisers already live in Algolia, Recommend adds a model to the same records and rules without a second vendor.
How this page was made
Method
We read each vendor’s own pricing pages and docs, every cell by hand, and linked the page. Request charges only, USD as published and EUR for us, nothing converted. We re-check them regularly.
Currency
USD prices are shown as the vendor publishes them; Patterna is in EUR. Nothing is converted, so read the bars as list prices in two currencies, not as one scale.
Verdicts
Yes, Partial, No, or Not published. “Not published” means we did not find it on the vendor’s pages; it is not a No. Partial means the capability exists with a documented limit.
Volumes
Request charges at 500k, 3M and 10M a month on the plan a buyer would pick for that volume; training, hosting, minimums and engineers are listed per vendor as not included.
Corrections
If a cell is wrong or out of date, tell us with the page that shows it and we fix it.