Software Integration Requirements: A Buyer's Checklist

Turn this article into takeaways for your work.

Each assistant summarizes the article only for you and suggests best practices for your work.

"Software integration requirements" sounds like a technical detail you hand to IT after the contract is signed. Treated that way, it's how a team ends up with a "native integration" that only pushes new records and never syncs an update, or an API that technically exists but sits behind a pricing tier nobody budgeted for. Both show up after signature, not during the demo.

This guide is the checklist that catches them first. Pair it with the SaaS vendor evaluation scorecard once you're scoring finalists, and with how to run a software trial once you're ready to test the connection instead of taking the vendor's word for it.

Key Facts: software integration requirements

  • The average enterprise now runs 957 applications, up from 897 a year earlier, and only 27% of them are actually integrated together, according to Salesforce's 2026 Connectivity Benchmark Report.
  • The average company uses 101 apps, per Okta's 2025 Businesses at Work report, which is a lot of surface area for a single sync to quietly fail on without anyone noticing.
  • 93% of API teams report ongoing collaboration blockers, and 55% specifically point to inconsistent, outdated, or missing documentation as the cause, according to Postman's 2025 State of the API Report. That's exactly the documentation you're supposed to be reading before you sign, not after.
  • HubSpot's own API usage documentation shows daily call limits ranging from 250,000 requests on Free and Starter accounts up to 1,000,000 on Enterprise, a real example of how a subscription tier, not just a feature list, decides what your integration is allowed to do.

What software integration requirements are (and why "it has an API" is not an answer)

An integration requirement is a specific, testable statement about how two systems have to exchange data. "It syncs with our CRM" is not that. It doesn't say which direction data flows, what happens to a deleted record, or which system wins when both get edited at once. A vendor can say "it syncs with Salesforce" about a connector that only creates new leads and never touches them again, and every word of that sentence is technically true.

The fix is writing down, before a single demo, what "integrates with X" needs to mean for your actual workflow, then holding every vendor to that same written bar.

Marketing claim What it's actually saying The requirement you have to write instead
"Native integration with [tool]" A connector exists. Depth, direction, and field coverage are unspecified. Which fields sync, in which direction, on which trigger, verified against a written list
"Powerful API" An API exists. Rate limits and which tier it's gated behind are unspecified. The rate limit, the auth method, and the minimum plan that includes API access, confirmed in writing
"Real-time sync" Faster than a manual export. Could mean webhooks, could mean five-minute polling. The actual maximum delay between a change and it appearing elsewhere, tested, not quoted
"Two-way sync" Data moves both directions for at least one field. Doesn't confirm deletes or conflicts. A field-level map showing what syncs each direction, including what happens to deletes

Direction and depth is the single most consequential gap. A one-way push (new records only, no updates, no deletes) is a fundamentally different product from a bidirectional sync that keeps both systems current. Before you score anything else, get a straight answer on two questions: which direction does each type of change flow, and which system is the source of truth per field when both could plausibly hold it. If nobody on the vendor's side can answer that in one sentence, that's your answer.

The integration requirements checklist

Score each row 1-5 during a real trial, not from the sales deck. The "how to verify" column is the part most buyers skip and the part that actually catches problems.

Requirement What good looks like How to verify
Direction and depth Bidirectional sync of creates, updates, and deletes, with a documented source of truth per field Get the field-level sync matrix, then create, update, and delete a test record and watch both systems
Field-level mapping and custom fields Custom fields and picklists map by default or through a visible mapping UI, with defined behavior for one-sided fields Add a custom field on each system during the trial and confirm what happens to it on sync
Trigger model Real-time webhooks for anything time-sensitive, a documented polling interval or batch window otherwise Ask for the average and worst-case delay between a change and it appearing on the other side, then time it
API limits Published rate limits, a working sandbox, and a documented versioning and deprecation policy Read the vendor's developer documentation before the sales call, not after asking a rep to summarize it
Conflict handling and idempotency A documented rule for simultaneous edits, and a retried call that never creates a duplicate Update the same record in both systems within a minute of each other and see which value wins
Error handling and observability A visible sync log, an alert on failure, and a way to replay a failed sync without redoing manual work Force a failure (bad data, an expired token) and see whether anyone would notice without checking
Identity, auth, and permissions OAuth rather than a shared API key, SSO/SCIM support, and a connection that survives an employee leaving Ask who the integration authenticates as today, and what breaks if that account is deactivated tomorrow
Data residency, PII, and compliance A stated location for data in transit, compatible with your DPA if middleware is involved Ask directly where a record sits, even momentarily, if it passes through an iPaaS layer
Ownership and maintenance A clearly named owner (vendor, official partner, or community maintainer) with a stated support commitment Ask what happens, and how fast, when the connector itself breaks, not your configuration
Exit and portability A full export of your data, in a usable format, that doesn't depend on the integration staying active Run a real export during the trial, before you're relying on the integration for anything that matters

How to weight the requirements

Not every row matters equally for every purchase. Weight them against how the integration actually gets used, and freeze the weights before you score anyone.

Requirement Weight Vendor A score (1-5) Vendor A weighted Vendor B score (1-5) Vendor B weighted
Direction and depth 20%
Field-level mapping 10%
Trigger model 15%
API limits 15%
Conflict handling and idempotency 10%
Error handling and observability 15%
Identity, auth, and permissions 10%
Ownership and maintenance 5%
Total 100%

The right weighting shifts with what's actually on the line:

Buyer scenario Over-weight these Why
Replacing a system of record (CRM, ERP) Direction and depth, conflict handling A one-way sync quietly creates two versions of the truth
High-volume, time-sensitive process (support, fulfillment) Trigger model, error handling A nightly batch is too slow, and a silent failure costs a customer
Regulated data (health, finance, EU personal data) Data residency, identity and auth A compliance gap costs more than a missed sync ever will
Small team, few integrations API limits, ownership and maintenance No engineer on standby to build a workaround when a connector breaks
Platform team building custom connectors API limits, versioning policy Your own uptime depends on someone else's roadmap decisions

Native, iPaaS, or custom build: how to decide

Once the requirements are written down, the harder question is which path actually delivers them.

Path Best when Real cost Watch out for
Native integration The vendor already ships the exact connection you need, at the tier you're actually buying Usually bundled, sometimes gated to a higher tier "Native" covers everything from a deep bidirectional sync to a one-way trickle. Verify depth against the checklist, never the name
iPaaS (Zapier, Workato, Make, Pipedream) You need several integrations, none at extreme volume, and no engineer to maintain custom code Billing that scales with tasks or runs, not seats The middleware is now a system you depend on and secure. Zapier vs. Workato vs. Pipedream covers how billing models diverge
Custom build against the API The integration is core to the workflow, runs at real volume, or is genuinely unique Engineering time to build, then maintain indefinitely The bill doesn't stop at launch. Someone owns this connector for as long as it runs, through every vendor API change

If no native option or iPaaS recipe gets you there, how to choose workflow automation software and the best no-code automation tools are the next stop before committing engineering time to a custom build. The best Zapier alternatives and the best Workato alternatives both break down pricing structure against feature depth, and smaller teams weighing whether any of this is worth the setup time should start with how to choose automation software for a small business.

Questions to ask the vendor before you sign

Ask these after the demo, when the rep is answering directly instead of running the standard script.

  1. "Walk me through what happens when a record is deleted on our side, not just created." Most demos only show the happy path: a new record flowing one direction into a clean destination.

  2. "What's your API rate limit, and which subscription tier is API access actually gated behind?" The integration you priced out during the demo may not be the one your contract includes.

  3. "Show me your API versioning and deprecation policy." No documented policy here means the vendor can break your integration on their schedule, with no obligation to warn you.

  4. "If a sync fails at 2am, who finds out, and how?" "You'd notice when the data looks wrong" is a silent failure mode you're signing up for.

  5. "Does this integration run as a named employee's login, a service account, or an app-level OAuth connection?" A named-user connection breaks the day that person changes roles or leaves.

  6. "If we're using an iPaaS layer, where does our data sit while it's in transit?" Middleware is a third system holding your data, even briefly, and it needs its own answer on residency.

  7. "If we cancel, can we export our data without this integration being active?" A vendor who hesitates is betting you won't leave. See the CRM migration guide for what a clean data exit requires.

How to test integrations during a trial

A demo shows you the integration working once, under ideal conditions, with the vendor driving. A trial is where you break it on purpose.

Test What you're checking Pass looks like
Create a record on each side Whether creates sync both directions, not just one The record appears on the other system inside the stated trigger window
Update a field, including a custom one Whether updates sync, and custom fields actually map The change reflects correctly, custom and picklist fields included
Delete a record Whether deletes sync, or get silently ignored The other system reflects the deletion, or the vendor documents that it doesn't sync deletes
Edit the same record in both systems within a minute Conflict handling A documented, sane outcome, not a coin flip or a duplicate
Break it on purpose: bad data, expired token, mismatched field type Error visibility You get an alert or a log entry, not silence
Retry a failed sync Idempotency No duplicate record gets created on the retry
Export your data out Portability A complete, usable export that doesn't require the integration running

Run this list in the first few days of the trial, not the last. A failed delete test on day three still leaves time to raise it with the vendor or eliminate them; the same failure found on day thirteen just gets rationalized away under deadline pressure. How to run a software trial covers the full instrumentation approach for logging this alongside every other criterion you're scoring.

Common integration mistakes

Trusting the marketing page over the field-level sync matrix. "Integrates with Salesforce" and "syncs every custom field on your Salesforce opportunity object, both directions" are different claims. Only one is a requirement you can hold a vendor to.

Skipping the delete and conflict tests. Creates are the easy case every vendor demos by default. Deletes and simultaneous edits are where real integrations quietly diverge, and they rarely surface unless you force them.

Letting the integration run as a named employee's login. It works fine until that person changes teams or leaves, and a business-critical sync goes down with no warning tied to it.

Assuming "custom fields sync" without checking the specific field. Standard fields syncing correctly says nothing about whether your team's one custom picklist, the one the whole reporting process depends on, maps at all.

Signing before checking which tier the API sits behind. A functional-fit scorecard filled out against the wrong tier's feature set produces a confident, wrong answer.

What integration actually costs

The sticker price rarely covers the full cost of making two systems talk to each other. There's no reliable industry figure for "typical integration maintenance cost," so none appears below. What's consistent is where the cost shows up.

Where the cost hides What it looks like Who owns it
Pricing-tier jump Full API access, or the specific native connector you need, sits behind a higher subscription tier than the one priced during evaluation Procurement, discovered at or after signature
Middleware billing iPaaS tools price by task, operation, or run, which scales with actual usage rather than seat count Whoever holds the iPaaS subscription
Build cost Engineering time to build against the API in the first place The team assigned to build it
Maintenance cost Ongoing engineering time every time the vendor changes an API version, deprecates an endpoint, or alters a webhook payload The original build team, indefinitely, unless someone explicitly reassigns it

Don't accept a vendor's or a consultant's quoted "integration failure rate" or "average maintenance cost" percentage at face value. Ask for the source. If there isn't one, describe the mechanism instead, the way the table above does, and price your own build-and-maintain time from your team's actual engineering rates rather than a number from a deck.

Frequently asked questions

What is a software integration requirement, exactly?

A specific, testable statement about how two systems must exchange data: which direction, on which trigger, and what happens to a deleted or conflicting record. "It has an API" or "it integrates with X" describes a capability, not a requirement, because neither tells you what your workflow needs the connection to do.

Is a native integration always better than an iPaaS connection?

No. A shallow native integration that only pushes new records can be worse than a well-built iPaaS recipe that handles updates, deletes, and error alerts correctly. Judge the connection against the checklist, not the logo that built it.

What does "the API is Enterprise-only" actually mean for total cost?

The price you evaluated during the demo isn't the price of the product you actually need. Budget the higher tier from the start if API access is a requirement, or confirm in writing that your tier includes it before you sign.

How do I know if an integration actually syncs deletes, not just creates?

Test it during the trial. Delete a test record on one system and time how long it takes, if ever, to reflect on the other. Vendor documentation and sales answers are often vague on this exact point, which is why it belongs in the trial test plan, not just the requirements conversation.

Who should own an integration once it's live?

Someone specific, named in writing: an internal engineer, an official vendor partner, or a documented community maintainer. An integration with no named owner keeps working right up until it breaks, and then nobody knows whose job it is to fix.

Write the requirements down before you compare vendors

The checklist above isn't meant to slow down a purchase. It moves the hard questions earlier, when a vendor still has to earn the deal, instead of later, when the contract is signed and the leverage has changed hands. A vendor who answers every row in writing, with real numbers on rate limits and a real answer on deletes, has told you something a demo never will.

Category guides pick this up where the generic checklist stops: marketing automation evaluation criteria scores CRM sync depth as its own weighted line, and the marketing automation migration guide shows what happens when those connections have to be rebuilt on a deadline.

Write the requirements before the first call. Score them the same way for every vendor. Test the ones that matter most during the trial, not after signature. That's the difference between an integration that quietly does its job for three years and one that shows up as a support ticket in month four.

About the author

Calvin D.

Calvin D.

Head of Enterprise Solutions

Calvin D. is Head of Enterprise Solutions at Rework, with 5+ years and 40+ enterprise engagements spanning 20 to 500+ user deployments. Calvin helps Heads of Operations, IT Directors, and VPs connect CRM, workflow automation, and data into one stack that actually fits together. Readers get field-tested architecture decisions they can apply as their teams scale.