Use caseRecords management software
Two years after launch, most of Corvane's 140 counties still run version 8. Each upgrade is custom, and no county can stop issuing permits.
Every county's data is inventoried before anything movesEach move is rehearsed on the county's real recordsVersion 8 stays live until the county signs
1.9 million records over 41 types, 11 custom fields, and 214 records whose file is already missing on the old share.
source · Westbrook database and file share, read in placeMissing before the move · the county records it
Two years ago, Corvane shipped version 10 of its records management software for local governments. Version 10 runs on a new document repository. Version 8 keeps its documents in a database and a file share, on servers in each county's building. Corvane still maintains both, and every upgrade is run by whichever consultant is free. The new release is ready; the installed base is not moving.
Corvane turns the move into a program. Every county's version 8 is inventoried in place, one versioned mapping carries every county with its own fields on top, and each county gets a test run on its real records, reconciled document by document. The county's key users check their own records, and the county signs before version 8 is switched off.
One rule runs the whole program: the county's data is known before the plan is written, every move is rehearsed and reconciled, and the switch happens only on the county's signature. The three acts read the base, rehearse Westbrook, then close its move.
Ingrid Solberg, Corvane's migration lead, starts with what is actually out there in 140 county buildings.
With read-only access each county grants, the operations engine analyzes the version 8 database and file share where they sit. It counts records per type, measures volumes, lists the custom fields each county added, and finds files the database references but the share does not hold.
Tomasz Nowak, VP of product, groups counties into waves with Ingrid, on the volumes the inventory measured. Wave 1 takes six small counties with few custom fields, to prove the method. Carrick, with 27 custom fields, gets a wave of its own. No wave lands on a county's fiscal year-end or on local elections.
Version 8 and version 10 do not store a permit application the same way. Ingrid writes the standard mapping once: old record types to new document types, old status codes to workflow states, old references kept as a searchable property. Each county's custom fields go on top. Every version of the mapping is kept.
Four weeks before its window, Westbrook County gets a test run on its real data while version 8 keeps running.
Ingrid runs mapping 4.2 on a sample first, then on all of Westbrook. The operations engine extracts, transforms and loads the records into Westbrook's new space on version 10, which no user sees yet. It runs at night to spare the county's servers, pauses during business hours and resumes from where it stopped.
The reconciliation report compares what was expected with what arrived, one line per record, with the version 8 and version 10 identifiers side by side. Westbrook's report names 214 records whose file was already missing on the old share, and 38 old fax scans converted for the new version.
Rosa Delgado leads permit administration at Westbrook. She picks 40 records herself: a recent application, a 1998 appeal, a protected tree permit, a file with large plans. She opens each one in version 10, in the browser, compares it with version 8 on her second screen and ticks it off.
The test run is clean. Westbrook moves in its agreed window, and the program turns to the next wave.
On Friday, September 10 at 5:00 p.m., Corvane sets version 8 to read-only at Westbrook. The operations engine synchronizes only what changed since the test run, with mapping 4.3, then reconciles again. Daniel Hughes, Westbrook's IT manager, opens version 10 to users on Saturday afternoon.
Westbrook's staff work in version 10 from Monday. Version 8 stays intact and can take users back. After two weeks with no blocking issue, Daniel approves the acceptance task in Corvane's customer process. Only then are the old servers retired with the county's IT.
Tomasz follows the program across every wave: counties accepted, counties in test run, anomalies waiting on a county, and the next windows. Each wave's trace exports as it is. He uses it to decide when version 8 can stop receiving updates, a date counties hear well in advance.
Each of Westbrook's 1.9 million records gets one of three results, and each result sends someone to do something different.
Count, file and properties line up between version 8 and version 10.
Next: the next batch, or the next step, can start.The file was already absent on the old share. The report says so.
Next: the county looks for it, or records the gap in its own files.A record failed a control on the way. It is listed with both identifiers.
Next: only the affected batch reruns, once the cause is fixed.It switches no county over. It retires no server. It drops no field. Across the nine steps on this page, the switch and the acceptance are the county's, made by named people on their own records.
The operations engine reads, transforms, loads and reconciles. The control is in the sequence: inventory, test run, key-user check, cutover, signature.
Westbrook County alone produced these counts. Another installed base, with other versions and other customers, will produce its own.
Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.
An installed base moves when the vendor's head of product, its migration lead and each customer's IT manager are ready to sign. This is what moves each of them:
You maintain two products because customers will not move.
Every upgrade is rebuilt by hand by whoever is free.
You cannot lose a record, and you cannot stop service for a week.
With one small customer and the inventory. Read their old installation in place and compare what it really holds with what your upgrade guide assumes.
Then write the standard mapping on that customer, and run it on a sample of their records before any window is set.
See it live
Share the data model of your old and new versions and one customer's volumes. We will plan a first wave with you and run a test on real data.