DEV Community

harsha tiwari
harsha tiwari

Posted on

Seven things that break XP to Sitecore AI migration Estimates- Part 2

Part 2 of the series. Custom code with nowhere to go, personalization nobody uses, the CDN nobody asked about, and the parallel run that is always missing.

In my last article I covered the first three gaps I keep finding in XP to SitecoreAI migration estimates: site count, rendering variants and content volumes. If you haven't read it, start there, as the same approach applies here. Connect to the XP CM with Sitecore PowerShell Extensions, pull the real numbers, and only then estimate.

Let's look at the remaining four.

Custom pipelines and computed fields with nowhere to go

This is the one that catches people who know XP well. In SitecoreAI there is no CD tier for you to run, no xDB, no delivery side Solr that you own, and no request time pipeline on delivery. Any customization that lived in those places has to be redesigned, moved to the front end, or dropped.

To find custom pipeline processors (anything outside the Sitecore.* namespace):

$config = [Sitecore.Configuration.Factory]::GetConfiguration()
$config.SelectNodes("//pipelines//processor[@type]") |
Where-Object { $.type -notlike "Sitecore.*" } |
Select-Object @{ n = "Pipeline"; e = { $
.ParentNode.Name } },
@{ n = "Type"; e = { $_.type } }

To find computed index fields:

$cs = [Sitecore.Configuration.Factory]::GetConfigNode("contentSearch")
$cs.SelectNodes(".//fields[@hint='raw:AddComputedIndexField']/field") |
Select-Object @{ n = "Field"; e = { $.fieldName } },
@{ n = "Type"; e = { $
.InnerText.Trim() } }

Treat every row as a line item. For example:

1. A httpRequestBegin processor doing redirects may move to the front end or to the CDN.
2. A computed field feeding site search needs a redesign for Sitecore Search, where content comes in through crawlers or the ingestion API, not your own indexing code. On one prototype I worked on, the search redesign was the real effort. The pages were the easy part.
3. XP era lines like CD hardening, Solr upgrades and xDB sizing should come out of the estimate. I still see them copied from the last XP project.
Enter fullscreen mode Exit fullscreen mode

While you are here, confirm the Sitecore Search tier. An entry level appears to be included with the SitecoreAI subscription, but it can vary by deal. Get it confirmed by the account team before adding or removing a search workstream.

Personalization that nobody checked

xDB, Experience Profile, EXM and the XP rules engine do not carry over. So the question "do we need CDP and Personalize?" comes up in every migration. Very often the honest answer is that personalization was bought but is barely used.

Check the layout fields for personalization rules:

Get-ChildItem -Path "master:/sitecore/content/" -Recurse |
Where-Object {
$["Renderings"] -match "<rls" -or $["__Final Renderings"] -match "<rls"
} |
Select-Object Name, ItemPath

If it returns a handful of pages, personalization is a small rebuild or a phase two conversation. If it returns hundreds, it is a proper scope item with a licensing question attached.

There are alsodata questions which is not technical. What happens to historical xDB contacts and interactions? Keep, export or purge? That is a decision for the client's data protection officer. Raise it early, as the answer decides whether you need a data export task at all.

Which CDN is actually in front?

In a RFP we need to know the CDN, The CDN in front of an XP site often holds redirect rules, geo logic and header rewrites, and all of that needs a new home.

Check this using :

Resolve-DnsName www.example.com | Select-Object Name, Type, NameHost

(Invoke-WebRequest -Uri "https://www.example.com" -Method Head).Headers

A CNAME ending in edgekey.net or akamaiedge.net points to Akamai. Headers like Server, Via, X-Cache or CF-Ray usually tell you the rest. Then ask the client what rules live there and add a line for moving them.

The parallel run, and the estimate's own sanity checks

Every phased migration keeps XP running for some time as a fallback. That means hosting, licences, monitoring and someone who can still deploy to it. In every estimate version I reviewed this year, this cost was missing. Not underpriced, missing.

And if someone wants an AI productivity saving in the estimate, show it as a range, mark it as an assumption and tie it to a pilot sprint. You will have to defend that number in the room.

To sum up the series, before sending an XP to SitecoreAI estimate:

1. Confirm the real site list in writing.
2. Count rendering variants and Scriban templates, and drop unused renderings.
3. Pull item counts per language, media size and unreferenced media.
4. List custom pipeline processors and computed index fields, and confirm the Search tier.
5. Check if personalization is actually used, and ask the DPO about xDB data.
6. Find out which CDN is in front and what rules it holds.
7. Price the parallel run, and check QA percentage and contingency.
Enter fullscreen mode Exit fullscreen mode

Half a day with SPE and a few awkward questions. That is the difference between an estimate we can defen.

Top comments (0)