Measure what changed, one fact at a time, before the project closes

Chapter 14 classified every site by its evidence of use before it joined a wave, and Chapter 9 made decommissioning wait for the verification record. Chapter 38 switches the farm off: the most visible result, and the least informative. This chapter reads the rest, each against its before, and says what each cannot establish.

What changed, beyond the farm being gone, and what does each measure establish?

The decision this chapter leads to is whether the project closes, or opens a named remediation phase with its own owners, taken on the measures read one at a time, never on a total.

Which projects this chapter is for

A move from SharePoint Server 2016 or 2019, copied into SharePoint Online (Chapters 8 and 9) or attached to Subscription Edition (Chapter 23), whose last wave is past cutover and hypercare (Chapter 37). It needs what earlier chapters archived: Chapter 14's exports, Chapter 11's search reports, Chapter 5's measured figures and Chapter 9's verification record.

Read each measure over a window that starts when hypercare ends, so the copy's writes and the first weeks' questions fall outside it, and spans the business cycle Chapter 14 wrote down.

Seven measures, and the before each one needs

Adoption: people with people, writes with writes

Online, the site usage page counts a site's unique viewers, "the total number of individual visitors to the site, regardless of how often they've visited", over 7 or 30 days, with daily trends and visits for 90; its owners can export 90 days (site usage, checked on 7 November 2026). The admin centre's site usage report lists every site's files, active files (those "saved, synced, modified, or shared"), page views and last activity date (site usage report); its activity report counts files viewed, edited, synced and shared, and pages visited, per user (activity report). Both cover the last 7, 30, 90 or 180 days and "typically become available within 24 to 72 hours" (usage reports overview, all three checked on 7 November 2026).

On Subscription Edition, Popularity Trends shows views and unique users, daily for 14 days and monthly for three years, the monthly figure being "the sum of unique users per day within the month" (view usage reports), once a farm administrator configures usage collection on the new farm (plan for site maintenance, both checked on 7 November 2026). Take the before from Chapter 14's exports, never from the new farm.

Three pairs compare:

• People with people. Chapter 14's distinct names per site per month from the IIS logs, what is not a person set aside, against the 30-day unique viewers online or the new farm's IIS logs.
• Writes with writes. Chapter 14's last write by a person, against active files, each read as yes or no for the window.
• Popularity Trends with itself, on Subscription Edition: summed daily counts match no other instrument.

Even the closest pair is two instruments: compare direction per site, not difference, and where Chapter 12 flattened a site collection, compare content, not address. Nor is the last activity date a last write: a person opening a file or viewing a page moves it (site usage report, above).

A drop in use is not a failure, and a rise is not a success. Chapter 26 left sites behind, Chapter 27 archived others, and a site read at year end waits for year end. Each change is a question for the site's owner, asked with both readings; the answer joins the record.

In the admin centre, "by default, all reports hide user information such as usernames, display names, groups, and sites", site addresses included. A Global Administrator who clears "Conceal user, group, and site names in all reports" reveals them in every report, and "showing identifiable user information is a logged event" (usage reports overview, above). Matching Chapter 14's sites needs those addresses, and the switch reveals people with them, so the owner of data protection decides, as Chapter 14 asked; otherwise read the site usage page, which counts people without naming them. What usage reports establish covers their other limits. Keep the files: the longest window is 180 days, so a year online exists only if somebody exported every month.

Search: the same reports, site collection by site collection

Chapter 11 exported the farm's number of queries, top queries, abandoned queries (those with "low click-through"), no-result queries and query rule usage (search diagnostics, checked on 7 November 2026). Subscription Edition's new Search service application produces the same reports, meaningful only after its full crawl. Online, each site collection's Insights page shows query volume and the top, abandoned and no-result queries, by day for 31 days or by month for 12, and a query joins the top list only when "searched at least three times with a click on a result" (search usage reports, checked on 7 November 2026). The tenant report adds searches in Teams, Outlook and Windows (Microsoft Search usage reports, checked on 7 November 2026), so it matches no farm report.

We would read the no-result queries first. A query that found something on the farm and finds nothing now is a finding, whatever the cause: content left behind, a permission, the index, or a query rule not yet rebuilt. Abandoned queries come next. Neither report says whether anybody found what they needed.

Time to reach content needs a test

We know of no Microsoft report that measures how long a person takes to reach a document. IIS records each request's time taken, which "does not reflect time across the network" (IIS log settings); online, the Page Diagnostics for SharePoint tool reports a page load time shaped by "network latency, the performance of the computer" and the browser (Page Diagnostics, both checked on 7 November 2026). Neither says whether anybody found the document.

So the measure is a test: tasks each owner chooses ("find last year's audit report"), done by the people who do them, timed, outcome recorded, before cutover and after hypercare. Chapter 5 kept "hours saved by better search" out of the business case unless a measurement existed; this is where one starts. A test first run after cutover has no before: it opens a series.

Tickets: like with like, or no comparison

If Chapter 5 measured support from the ticket system, run its query again, with the same categories, over equal windows before cutover and after hypercare. Hypercare's weeks measure the change itself, so a rise in tickets in the first weeks is not a verdict.

Count per category, never in total: missing content, access, how-to, platform. Check each missing file against Chapter 8's verification record, which says whether it existed and moved; fewer how-to questions do not offset more missing files. Service health keeps "the history of any service advisories and incidents that have affected your tenant in the past 30 days" (service health, checked on 7 November 2026): record it monthly, so a ticket raised during a Microsoft incident counts as the platform's. On Subscription Edition the platform is yours: its incidents sit beside Chapter 42's update log, and IIS logs each request's HTTP status, so "not found" responses compare per site, old farm against new (IIS log settings, above). Online we know of no Microsoft report of the errors people meet: they reach the record as tickets.

If the farm's tickets were never categorised by cause or site, there is no before, and we would not reconstruct one from memory: the after starts a series.

Three lists, each line with a name

Infrastructure retired is Chapter 38's record read against Chapter 2's inventory: servers, databases, service accounts, certificates, DNS entries and backup jobs, each by name. Servers switched off are a fact; a saving is not, until somebody holds the costs. Return on investment stays out for that reason: it needs cost data IT rarely holds in full, and financial precision must not exceed evidential precision (Chapter 5).

Risks removed come from Chapter 5's known risks and Chapter 7's farms out of support, each with its evidence and the owner who accepts it as gone. SharePoint Server 2016 and 2019 reached the end of support on 14 July 2026 (ending support in 2026, checked on 7 November 2026); that risk ends when the last server running either is switched off, not at cutover. On Subscription Edition it changes shape rather than ending, and Chapter 42's calendar keeps it away. New risks join the same list, none netted against an old one.

Capabilities in use are the benefits, evidenced or unpriced, that Chapter 5's case named, each with the date of its first use in production, by whom, confirmed by its owner. Configured and never used is not yet a result.

Close, or open a remediation phase

Each measure ends in one row, in one of three states:

As Chapter 6 said of governance, unknown is an answer: a row not established is not a quiet pass. The decider Chapter 1 named takes the decision, on rows the validators confirmed. The project closes when every finding is fixed or accepted in writing, and every row not established has an owner and a date in routine operation (Chapter 39). Otherwise it opens a remediation phase: named, with its own owners, a scope that is the open findings and nothing else, and an end that is the last of them closed.

Seven measures, read separately, are seven statements with seven owners. Added into one figure, they let the retired servers pay for the search nobody can use.

Out of scope

Hypercare is Chapter 37, decommissioning Chapter 38, governance after the migration Chapter 39 and lessons learned Chapter 41. Return on investment, prices, licensing, Purview auditing, satisfaction surveys and adoption programmes are not covered.

Before you continue

• Every measure's window starts after hypercare and spans the business cycle Chapter 14 wrote down
• Adoption is compared per site, like with like, and every change carries its owner's explanation
• The owner of data protection decided whether reports show names, and the exports are kept monthly
• No-result and abandoned queries were read against Chapter 11's exports, and each lost result is a finding with an owner
• Tickets are counted per category over equal windows, platform incidents apart, and no before was reconstructed
• Time to reach content comes from a timed task test, or is recorded as not established
• Infrastructure retired, risks removed and capabilities in use are lists with a name on every line
• Every measure is established, not established with an owner and a date, or a finding, and no figure combines them

Download chapter 40 (PDF)Same text, same tables, same sources as this page.