Ask buyers how long their personalization engine took to implement and few give a complete start-to-finish timeline. The clearest reported durations concern different milestones: six months to build one personalized landing-page test in-house, and about a year and a half for a phased rollout across page types after a very long pilot; a first working placement is not a finished rollout. The long waits come after signing, and they differ in kind: one large specialty retailer was still actively setting up its replacement engine more than a year in, while one brand was still not using recommendations it had bought about twenty months earlier. Two things are specific to these engines: go-live does not establish value, and engines can go live one placement at a time, sometimes behind a test. This page draws on standalone personalization engines, suite modules and in-house builds; recommendation features bundled into site search or Shopify apps are not covered. The companion reads are why companies switch personalization engines and what personalization-engine buyers wish they'd known, where the ratings live.
What a personalization implementation spends its time on
The product feed and the tags come first
Before an engine can recommend anything it needs the catalog and the behaviour data, and buyers describe that plumbing as the first delay. One team: “It was we couldn't get the tagging set up, we couldn't get the product feed set up, and it took forever to get going.” Another says “it takes a lot of resources in order for us to implement the A/B tests and get the feeds ready for the product recommendation.” And the feed work does not end at go-live: a retailer with recommendations already live says “not everything's in place yet from a feed perspective to give us quite the power that we could get from it.” Why the data and content come first is on what personalization-engine buyers wish they'd known; as with CDPs, the data is the long pole.
Named in this context: Monetate · Dynamic Yield
Every placement joins the engineering queue
An engine that needs a developer for each change runs at the speed of the development calendar. At one retailer, “Our merchandisers don't know how to code, so it needs to be sent to our tech team, but that has to be booked into a PI planning session.” The result: “So doing something as simple as adding a new recommendation carousel can end up taking several months because we can't upload it to the website if it doesn't match stylistically.” Another buyer says “there's a little bit of a misnomer I think in the industry that you don't need a developer,” and a third finds the server-side version of its tool “really difficult to do it.” As with A/B testing tools, the pace is set by engineering time.
Named in this context: Dynamic Yield · Monetate · Adobe Target
Onboarding can be harder than it was sold
Several buyers describe set-up taking longer than the sale suggested. One was told “upfront in the sales process that it's no problem, it's no-brainer, they do it all the time,” and some months into the engine has one carousel live. A large specialty retailer still actively setting up its engine more than a year in finds that “everything needs to be done custom,” a story told below. Of a suite's website engine, another buyer says it “took a lot of time to implement and did not live up to the business case that Salesforce had.” These are onboarding complaints from individual buyers, and fast launches appear in these interviews too; the complaint form of integration effort is on top complaints about personalization engines.
Named in this context: Monetate · Salesforce
Some purchased engines never reach active use
Not every long wait is a long build. Some buyers bought personalization products and did not put them into use. A specialty ecommerce brand bought a recommendations product, had not implemented it at ten months and was still not using it at about twenty, a story told below. Another company “bought the tool and paid for it for a year, but we couldn't really build a business case to use personalization” with it. These measure the time between buying a product and using it, not time spent continuously implementing it, and the interviews do not give a single cause for either. Why under-used engines are a standing risk is on what personalization-engine buyers wish they'd known.
Named in this context: Optimizely · VWO
The build-or-buy cycle comes before the build
Some companies spend years deciding between building and buying. A consumer marketplace had “looked at Optimizely and Dynamic Yield as solutions and not pulled the trigger, as recently as last year,” built one personalized landing-page test itself instead, and says “it took us six months to build that”; a year on, it was running a vendor's engine. A large retailer piloted its own engine for three or four years before replacing its vendor, a story on why companies switch personalization engines. As with loyalty programs, the clock starts before the build.
Named in this context: Dynamic Yield · Optimizely
The program and the contract around it set the date
An engine is often bought as part of a larger move, and its date follows that program. A national specialty retailer consolidating onto one suite planned one-to-one personalization and testing as “something which we really want to unlock in next nine to 12 months,” and about six months on said of the tool, “We just onboarded it a month back, like it went in live.” A multi-brand company plans personalization alongside a new commerce platform and data platform: “All of that will be over the course of twelve to sixteen months.” And the contract sets the exit: “I don't anticipate that we would sign another deal with Monetate once our current term expires.” As with CMS migrations, the suite sets the order, and as with BI tools, the contract sets the window.
Named in this context: Adobe Target · Monetate
How long does a personalization engine implementation take?
These interviews do not establish a typical duration, and few buyers give a complete start-to-finish timeline. Two buyers report durations for finished but different kinds of work: six months to build one in-house landing-page personalization test, and about a year and a half for a tool's rollout across page types after a very long pilot, with the full rollout coming in the last six months of that. Two more can only be approximated by comparing interviews with the same person: a tool live about six months into a nine-to-twelve-month plan, and recommendations live about three months after a project began, with feed work still outstanding. Those measure one test, a phased rollout and two first go-lives, so they should not be combined into a typical range, and a first live placement is not a finished implementation. The longest clocks belong to work that did not finish: a replacement engine still being actively set up more than a year in, and a different kind of wait, recommendations still unused about twenty months after purchase and a module paid for over a year and never used.
When does a personalization engine start producing results?
These interviews do not establish a typical period, and go-live did not settle it. A subscription apparel business, after a first season of poor output, says “My team and I had to scramble to get the product groups up in the way that they should,” and was told by its vendor that “it takes about 90 days for the AI system to kind of learn and produce” — the vendor's estimate, relayed by the buyer, not an observed result. A fashion brand ran a pilot “starting in May of last year, and we've just not really seen the results.” Another retailer was “only six, seven months into that,” with one carousel live. A team with a newly live engine described being “in a sort of bedding-in period with those.” Model learning, gathering enough evidence to judge performance, and reaching a result the business accepts are different milestones, and these accounts do not separate them: a pilot without results after nine months may reflect the traffic, the product data or the set-up as much as the model. This rests on six companies and is not a rate; the brands with low traffic or small catalogs gave volume as their own explanation, as with A/B tests, which wait on traffic.
Should you roll personalization out across the whole site at once?
In several accounts buyers did not. A large online retailer says its tool “started with a very long pilot and not every part of the website has had BlockBuilder from the beginning,” moving from the homepage to landing pages and then product pages, with “A full rollout probably just in the last six months.” An apparel retailer went live with recommendations while “in phase one of our deployment.” Another retailer went live on a subset test of one assortment and then expanded. Once live, each new placement can take months of its own, as with the carousel in the engineering queue above. This rests on six companies, two of them describing plans or told in aggregate, so it is the order these accounts describe, not a rule.
Do personalization projects stay on schedule?
These interviews give too few plan-versus-outcome comparisons to answer. One tool went live inside its planned window, though the milestone was the tool rather than the use cases it was bought for; one personalized section launched about six months later than planned, a story told below; and one recommendations project arrived later and narrower than an earlier plan, by a different executive's account. Before the build, others were pushed, for budget or because no replacement had been scoped. That cannot show how often personalization projects finish early, on time or late.
Due for a stack physical?
Book mine →The stories behind the timelines
The counter-current: the go-lives that came quickly
The quick ones here share a shape: one tool on one surface, with a suite or an existing team behind it. A tool went live about six months into a nine-to-twelve-month plan; recommendations went live about three months after a project began; and a pilot was validated within about two months. None was a whole program, and the quickest was live in phase one with feeds still missing. They say nothing about which vendor is faster — the same engines appear on the quick and the slow sides here — and recommendation features bundled into site search or Shopify apps are not covered, so none of this says personalization projects are fast or slow in general.
What this means for your personalization plan
In these interviews the length of a personalization implementation is set after signing, by the feed and tagging work, the engineering queue, the rollout across placements and whether anyone puts a purchased feature to use, more than by the vendor's installation. Our recommendations, not practices buyers describe: get the product feed and tags working for the planned placements before you set a launch date; book the engineering time for the first placements before you sign, and agree how a merchandiser can make a change without a development cycle; name the owner and the resources that will put a purchased feature into use; agree with the vendor what working means and when to judge it, before go-live; date each placement and its validation as its own milestone; and put the contract end and any suite program on the plan. For why teams move or stay, see why companies switch personalization engines.
Common questions
How long does a personalization engine implementation take?
These interviews do not establish a typical duration, and few buyers give a complete start-to-finish timeline. The two who give a number describe different kinds of work: six months to build one personalized landing-page test in-house, and about a year and a half for a phased rollout across page types after a long pilot. The quicker go-lives were first placements of single tools, not finished implementations. The long waits come after signing: one retailer was still actively setting up its replacement engine more than a year in.
When does a personalization engine start producing results?
Not necessarily at go-live, and these interviews do not establish a typical period. One vendor told a buyer its model needed about 90 days to learn, which is the vendor's estimate rather than an observed result; one pilot had not shown results nine months in, for reasons the interviews cannot isolate; and another team was six or seven months in with one carousel live. Model learning, validating performance and reaching a business result are different milestones. Agreeing with the vendor what working means, and when to judge it, before go-live is our recommendation, not a practice buyers describe.
Should you roll personalization out across the whole site at once?
In several accounts buyers did not. They went live on one placement or page type, sometimes behind a test, and added the next: one retailer moved from a long pilot to the homepage, then landing pages, then product pages. That is the order these accounts describe, not a rule. Dating each placement as its own milestone is our recommendation, not a practice buyers describe.
Why do personalization projects stall after signing?
Buyers point to the product feed, tagging and data plumbing; an engineering queue that every new placement has to join; and onboarding that turned out harder than it was sold. A different kind of wait is a product bought and never put into use: one brand was not using its recommendations about twenty months after purchase, and one buyer paid for a personalization module for a year without using it. That measures purchase to use, not continuous implementation work.
Do personalization projects stay on schedule?
Too few buyers compared a personalization project with its plan to say. Where they did, one tool went live inside its planned window, one personalized section launched about six months later than planned, and other projects were pushed before they began, for budget or because no replacement had been scoped. That cannot show how often personalization projects finish early, on time or late.
A physical for your tech stack
Peer-powered: built from hundreds of verified buyer interviews, and fitted to the tools you already run. You get a shortlist for your stack and your next steps.
Book my stack physicalFree · no password · your answers are anonymized before they inform research like this.