Describe the integration questions that determine real total cost.

Endsleigh County Library Consortium shares one catalog system across a dozen small branches. Ansel Petrosky is the systems librarian evaluating Cataly, a vendor that auto-generates subject tags and metadata for new acquisitions. A peer consortium's own costly exit from a similar vendor is what changed how Ansel asked his questions.

How to answer this, stage by stage

Nobody is scoring whether you can list integration questions. They're scoring whether you can rank them by which one you can't take back.

Let's learn

Say a library consortium wants an AI tool that reads a new book's cover and table of contents and writes its subject tags automatically, instead of a librarian typing them in by hand.

Before Cataly, cataloguing a new item took a librarian about twenty minutes: scan the cover, write the subject tags, enter the record, file it in the shared system. Across all twelve branches, that's roughly eighteen thousand new items a year, plus a backlog of thirty four hundred waiting to be catalogued.

This is the job Cataly is meant to speed up. Nothing about it names what happens if the consortium ever wants to leave.

Cataly's pitch cuts that to about three minutes an item, and its license reads fifteen thousand dollars a year for the whole consortium, a number that looked easy to approve on its own.

At its worst, signing on a license number alone, with no answer on data portability, risks the consortium discovering years later that leaving costs more than joining ever did, at a moment when leaving is exactly what it needs to do.

What I would leave alone: a tool the consortium could swap out in an afternoon, like a simple room-booking widget for meeting spaces, doesn't need this level of scrutiny. The caution here is for decisions that are genuinely hard to reverse, not every purchase.

The lesson: a license fee is the smallest number in the decision. The real cost lives in the questions nobody thought to price until it was almost too late to ask them cheaply.

Now here is the same thing as a story

The short version above is what you'd say defending this evaluation to Endsleigh's consortium board. Read this one for how the real questions actually got asked.

Ansel Petrosky had managed Endsleigh's shared catalog system for six years, patient enough to read a vendor contract's whole appendix before anyone else on the board bothered to open it.

Most of Cataly's pitch deck was about the left door. The real risk was hiding behind the right one.

Cataly's sales team led every conversation with the license number and the time saved per item. Both were real. Neither one mentioned what happened to two hundred forty thousand existing catalog records once they moved into Cataly's own tagging system.

Word reached Ansel from a systems librarian at a neighboring county consortium, who'd adopted a similar AI cataloguing vendor two years earlier and later needed to switch to a new shared catalog system entirely. Getting their own records back out, in a format the new system could read, cost them ninety five thousand dollars, more than six years of the original license fee combined.

Ansel realized he'd been pricing Cataly's decisions in exactly the wrong order.

Ansel asked Cataly one question before any other: could Endsleigh export its tagged records back into standard MARC format, and at what cost. The vendor confirmed it was possible, through a paid migration service, quoted at fifty eight thousand dollars, worse than the forty two thousand dollar cost of migrating in, since Cataly's own export tool only handled a small fraction of its proprietary tag types cleanly.

He had considered simply trusting Cataly's promise that "standard export is always available" without a real quote attached. He pushed for the number instead, since a promise with no price behind it isn't a real answer, it's a sales line.

Four questions, none of them on the pricing page. All four decided the real cost.

Ansel also found a recurring cost buried in the integration details: Cataly charged per record for its ongoing sync with Endsleigh's shared system, and the vendor's own roadmap suggested a full re-sync of the whole catalog roughly every two years as its tagging model improved, a cost that never appeared on the license page at all.

Endsleigh ran a five hundred record pilot before committing the other two hundred thirty nine thousand five hundred.

Endsleigh ran a small pilot first, migrating five hundred records to learn the true per-record labor cost, then negotiated a capped exit-migration price into the contract itself, before signing anything for the full two hundred forty thousand.

ORDER, in five movesNot a checklist of vendor questions. ORDER is what tells you which one to price first.

This order isn't a preference. It follows directly from what's hardest to undo.

The recap, one line per letter: outcome is naming the real total cost including the exit, reversibility is the exportability question that's nearly impossible to undo, dependency is needing the technical answer before pricing labor, evidence is a small pilot revealing the real numbers cheaply, and rank is exit cost first, then migration, sync fees, and ownership.

And if you want to be sure it really works, try it somewhere elseSame five letters, a veterinary clinic network instead of a library consortium. A different dependency breaks the second ranking.

Foxhollow Veterinary Group runs eight clinics sharing one patient record system. Rosalind Kemp, its operations director, was evaluating VitaTag, a vendor offering AI-generated summaries and diagnostic tags for patient visit notes. Mapped onto ORDER: outcome is the real cost of the tool over its contract life, not just its per-visit fee. Reversibility is again the exportability question, whether VitaTag's generated diagnostic tags can move to a different veterinary record system later in a standard format. Dependency here breaks differently: Foxhollow couldn't price migration labor at all until it knew whether VitaTag's tags mapped to the clinic's existing diagnostic-code standard, a mapping VitaTag itself hadn't fully built yet, which meant the true integration cost was unknowable until VitaTag finished its own unfinished work. Evidence is asking VitaTag to run its tagging on twenty real patient records first, to see how many tags needed manual correction. Rank puts the mapping-dependency question first here, ahead of exit cost, since nothing else could even be estimated until that one technical answer existed.

Foxhollow faced the same two doors Endsleigh did, just with a diagnostic code standard instead of MARC.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "price the exit before you price the entry, since that's the decision you can't take back," and stop.
Cost: there's no time to negotiate a capped exit fee before a decision is due. Say so honestly, and at minimum get the vendor's real export cost in writing before signing, rather than skipping the question.
The model gets better, for real: if Cataly's next release ships a clean, free, standards-based export tool, that's still worth verifying on a real sample before trusting the license number alone.

Where people run it wrong.
They compare vendors by license fee alone, without asking what happens to their own existing data.
They accept a vendor's word on data portability without a number attached to it.
They price migration-in labor from a demo, instead of running a small pilot at their own real volume first.

How to use it live. The moment someone asks about integration costs, ask back: which of these decisions can we actually undo later, and which one locks us in? Price the one that locks you in first.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0