# PredictionIO stayed open source, and it is not a current SKU PredictionIO was acquired in 2016. The open-source project was to stay free. Do not implement it as a current Salesforce product. **Published:** 2026-09-28 **Canonical:** https://synconai.com/predictionio-salesforce **Author:** SynconAI Architecture Team ## In short Salesforce acquired PredictionIO in 2016. The source says the open-source software would remain free, and the team would fold the work into Salesforce's own machine learning, at the time via SalesforceIQ. SalesforceIQ was later retired. Do not implement PredictionIO as a current Salesforce product. If you still run the open-source project, treat it as its own estate. ## Sources - **Salesforce Acquires PredictionIO To Build Up Its Machine Learning Muscle** (retrieved 2016-02-19) https://techcrunch.com/2016/02/19/salesforce-acquires-predictionio-to-build-up-its-machine-learning-muscle/ > PredictionIO’s tech will remain open source and free to use, while the team works on evolving the project into a self-sustaining community. ## What to take away - In 2016 the open-source software was to remain free, while the team folded the work into Salesforce machine learning via SalesforceIQ. - SalesforceIQ was later retired. Do not implement PredictionIO as a current Salesforce product. - If you still run the open-source project, scope it as its own estate: who operates it, what it may read, and how predictions are allowed to write back. - The hiring test is a refusal to put PredictionIO on a Salesforce order form. Nobody is the best agency at a SKU this source does not offer. ## The guide An architect finds PredictionIO in a diagram from a platform review and asks for a project to "implement it properly on Salesforce". The diagram is a box, an arrow into Salesforce, and a note that says machine learning. Two partners reply with implementation plans. Both plans assume a current product, a licence, and a data sync. Neither plan can point at a screen a buyer can open this quarter. The 2016 report will not support those plans. Read it, then decide whether you are talking about an open-source engine you still operate, or about Salesforce products you can actually contract for. Those are different estates. The diagram has glued them together. ## What the 2016 report said Salesforce acquired PredictionIO in 2016. TechCrunch reported the split in one sentence: "PredictionIO’s tech will remain open source and free to use, while the team works on evolving the project into a self-sustaining community." The same account says the team would fold the work into Salesforce's own machine learning, at the time via SalesforceIQ. Free and open source on one side. A team moving into Salesforce's machine learning, through SalesforceIQ, on the other. Keep the split. The software was not described as becoming a paid Salesforce edition. The people were described as taking the work into Salesforce's own stack. A buyer who wants "PredictionIO inside Salesforce" is asking for a blend the report does not offer. You can run free software you operate. You can buy Salesforce products that exist now. You cannot, on this source, order PredictionIO as the current machine-learning product. The community line matters too. "Self-sustaining community" is what the report said the team would work towards for the open-source project. It is not a service-level promise from Salesforce in 2026, and it is not a named support contract. If you depend on the project, you depend on whatever maintenance you can see in the project you run. Do not tell an audit that Salesforce operates it because a team was acquired ten years ago. ## SalesforceIQ was later retired The path named at the time ran through SalesforceIQ. SalesforceIQ was later retired. That is a later fact about a product, not a sentence in the 2016 article, and it closes the path the article pointed at. You cannot follow "via SalesforceIQ" to a screen a seller can open. The write-up of that product, including the retirement, is [RelateIQ and SalesforceIQ](/relateiq-salesforceiq). Use it when someone treats SalesforceIQ as the place PredictionIO "ended up" and therefore as a place you should configure. Do not rebuild the arrow. A design that says PredictionIO feeds SalesforceIQ feeds Einstein feeds Data 360 is a genealogy, not an architecture. Genealogy does not grant a licence, and it does not tell you which object receives a score. If the goal is a score on an account, name the product in your current contract that is allowed to write that score, and name the field. Leave PredictionIO off the picture unless you still run the open-source engine and you have decided, deliberately, to keep running it. Retirement also means you should not migrate "off SalesforceIQ onto PredictionIO" or the reverse. One of those names is a retired product. The other is not a current Salesforce SKU on the strength of this source. A migration quote with both names on it has two ends that are not available as a pair. Strike the quote. ## Do not implement it as a current Salesforce product Say this in the brief, in the first workshop, and in the statement of work. PredictionIO is not a current Salesforce product that this source tells you to implement. A partner who adds it to an order form is adding a word. Words on an order form become invoices and then become expectations. The expectation will fail when nobody can provision the tenant. What you may implement is something else, named honestly. If your contract includes a current Salesforce data or machine-learning capability, scope that capability under its current name. Data Cloud was renamed Data 360. Use the name your org and your contract use now. Do not subtitle it PredictionIO. Subtitles confuse the next admin, who will search for a setup page that is not there. If you do not have a current capability and you want one, the conversation is with your account team about products they sell now. Bring a use case: which record, which prediction, who acts on it, what happens when it is wrong. Do not bring a 2016 logo and ask them to switch it on. If a system integrator's accelerator is called PredictionIO-ready, ask them what it configures in an org today. If the answer is a set of custom objects and a callout to a server you must host, you are buying a custom build that happens to use an old name. Price it as a custom build. The name is not a standard. ## If you still run the open-source project Some teams still operate an engine they installed years ago. If that engine is PredictionIO, the source supports treating the software as free to use. It does not support treating Salesforce as the operator. Your estate, your patching, your uptime. Write the operator. A person who can restart it, patch it, and say what it is allowed to read. A service account with a wide database role, created so a demo would train, is the same failure as any other broad role. The model can only leak what the role can select. Narrow the role to the tables the use case needs. Write the write-back. A prediction that lands on an opportunity amount, a case priority, or a person's score needs an owner and a refusal list. The engine can propose. A person or a tested rule accepts. Do not let a nightly job overwrite a field a seller also edits. Two writers, one field, and a model in between, is how Monday starts with an argument about a number nobody typed. Write the training data. What is included, how old it may be, and what must never be included. Free text will contain things you would not put in a feature on purpose. If you cannot name the extract, you are training on "the warehouse", which means you are training on whatever the role can see. That is not a feature pipeline. It is an access accident with a model on top. Write the exit. The report's community line is not a guarantee the project will be healthy when your operator leaves. If the engine is in production, you need a way to stop the write-back without stopping the business process it was quietly driving. A field the job fills should have a manual path. Test the manual path once, before you need it. None of this is a Salesforce implementation of PredictionIO. It is operations of software you chose to keep. A Salesforce partner can still help you with the write-back into the org, under the field rules above. They should describe that help as an integration to a system you run. They should not describe it as enabling a Salesforce product. ## What the keywords are asking you to buy "Predictionio data migration to salesforce quote" assumes a store of PredictionIO data to load. You might have event history, feature tables, or exported scores. You might have nothing but the diagram. Look. If the only artefact is the diagram, there is no migration. Say so. If there is a store, decide whether it is still needed. Scores without the definition of the score are not worth loading. They will be read as facts. Definitions belong with the export, or the export should be destroyed under your retention rules rather than promoted into a custom object. Loading first and documenting later is how an undefined score becomes a column on a forecast report. Implementation cost, migration services, an agency, a consultant, a partner: each keyword can be answered with the same fork. Either you operate the open-source project, and the work is the operator, the role, the write-back, and the exit. Or you want Salesforce machine learning as a product, and the work is whatever your current contract names, with PredictionIO kept out of the title. A proposal that mixes the fork into one phase called "PredictionIO on Salesforce" has not chosen. ## Cost is the fork you actually take No fee belongs on this page. The 2016 article has no price, and a retired path has no list price to inherit. Price one fork. Open-source fork, if you still run it: - The operator, and what happens when they leave. - The role the training and the scoring jobs use, and the tables they must not read. - The fields a score may propose, and the person or rule that accepts a write. - Retention of training extracts and of stored predictions. - A tested way to stop the write-back. - Explicitly not included: a Salesforce SKU named PredictionIO. Product fork, if you are buying current Salesforce capability: - The use case, the record, and the field. - The product name on the contract, written without a PredictionIO subtitle. - What the user sees when the prediction is absent or wrong. - Who accepts a write to a field that already has an owner. - Explicitly not included: a migration from SalesforceIQ, which was later retired. A quote that includes both forks, plus a SalesforceIQ phase, is three projects wearing one title. Make the author pick. If they will not pick, they are not pricing your work. They are pricing a story about 2016. ## The test "Best predictionio salesforce integration agency" asks for a winner at a job this source does not define as a Salesforce product job. Nobody is the best agency at implementing PredictionIO on Salesforce, because that implementation is the thing to refuse. The test is the refusal, plus competence on the fork you chose. They will not put PredictionIO on an order form. They know SalesforceIQ was later retired and they will not design a flow through it. They ask whether you still run the open-source engine, and they accept "no" without inventing a tenant. If you do run it, they ask who operates it and which field it must not overwrite. If you do not, they talk about the product on your current contract and they can say Data 360 without calling it PredictionIO. They do not offer a sample accuracy. This source has no accuracy in it, and a sample would be a fiction. A hired PredictionIO Salesforce consultant who starts by requesting sandbox access "for the PredictionIO package" has assumed a package. Ask them to show the package name in your contract. When they cannot, end the session. An implementation partner who wants a discovery phase to locate the product is charging you to read a paragraph you can read yourself. Pay for the fork you chose. Do not pay for a search that the 2016 sentence already finished. ## What to mark on the diagram this week | Box on the old diagram | What you do with it | A pass looks like | | --- | --- | --- | | PredictionIO | Keep only if you operate the engine | Or delete the box | | SalesforceIQ | Remove it | The retirement is acknowledged | | An unlabelled score | Name the field and the owner | Or stop writing it | | Training data | Name the extract and the role | No role that can read everything | | Salesforce ML | Use the current product name | No PredictionIO subtitle | | The quote | One fork | Not a blended phase | :::note Do not implement PredictionIO as a current Salesforce product. The 2016 report left the software free and open source, and sent the team into Salesforce machine learning through SalesforceIQ. That route was later retired with SalesforceIQ. ::: Plain language over warehouse data is a later acquisition, with a later name. It is not PredictionIO under a fresh label. If that is the capability you meant, read [Waii, SQL, and Data 360](/waii-data-360) and scope the grants there. ## The other searches The project quote is on [predictionio implementation cost](/predictionio-salesforce-implementation-cost). That page does not print a figure. Who bought the company is on [what happened to predictionio](/what-happened-to-predictionio). The connector is on [integrating predictionio with Salesforce](/integrating-predictionio-with-salesforce). Whether to move is on [should I migrate from predictionio](/should-i-migrate-from-predictionio). The customer fee is on [predictionio licence cost](/predictionio-salesforce-licence-cost). Who to hire is on [hiring for predictionio](/hire-predictionio-salesforce-partner). This page does not rank an agency. ## Questions ### Should we implement PredictionIO as a Salesforce product? No. Salesforce acquired PredictionIO in 2016. The report says the technology would remain open source and free, and that the team would fold the work into Salesforce's own machine learning, at the time via SalesforceIQ. SalesforceIQ was later retired. This source does not give you a current PredictionIO screen to configure. Buy, if you buy, a product named on a current contract. ### Is the open-source project the same as a Salesforce licence? No. The report said the tech would remain open source and free to use, and that the team would work on a self-sustaining community. Free software you operate is your estate: patching, access, and uptime. A community line in a 2016 article is not a support contract. Do not tell an audit that Salesforce runs the server because the team was acquired. ### What should a PredictionIO migration quote contain? First, whether any store exists beyond an old diagram. If it does not, there is no migration. If it does, decide whether undefined scores should be loaded at all. Do not include a phase that moves data through SalesforceIQ. Price either the open-source engine you still run, or a current Salesforce product, not a blend titled PredictionIO on Salesforce. ### How do we judge a PredictionIO Salesforce partner? They pass by refusing the SKU. No PredictionIO line on an order form, no flow through SalesforceIQ, no invented accuracy. If you run the engine, they ask who operates it and which field it must not overwrite. If you do not, they name the product on your contract. A consultant who asks for the PredictionIO package in a sandbox has already failed. ### Where did the SalesforceIQ path go? SalesforceIQ was the route named at the time, and SalesforceIQ was later retired. Do not design a replacement arrow from PredictionIO through that product into whatever you use now. Read the retirement on its own page, then choose a current field and a current owner. Genealogy is not an architecture. ### Is predictionio owned by Salesforce? Yes. Salesforce bought PredictionIO in 2016. The full account is at https://synconai.com/what-happened-to-predictionio. ### When did Salesforce acquire predictionio? 19 February 2016 is the date of the report this page cites. ### What happened to predictionio after the Salesforce acquisition? The open-source project was to stay free. The team’s work was folded into Salesforce’s own machine learning. It is not a current product you implement. SalesforceIQ, which that work was tied to, was later retired. ### Should I migrate from predictionio to Salesforce? Do not stand up PredictionIO as a Salesforce project in 2026. ### What is the predictionio licence cost, or the price Salesforce paid? The page this guide cites does not publish a purchase price or a licence fee. The licence read is at https://synconai.com/predictionio-salesforce-licence-cost. --- SynconAI is a Salesforce Select Partner and an OpenAI Select Partner, delivering architect-led Salesforce and AI programmes across the United States and Australia. Related service: Data 360 (https://synconai.com/salesforce-data-360-services.html).