EDI Integration

Moving EDI Off BizTalk Without Breaking Trading Partner Relationships

Moving EDI Off BizTalk Without Breaking Trading Partner Relationships

Ask an integration team what worries them about a BizTalk migration and the EDI answer comes back fast: “we have 140 trading partners.”

The number is not the problem. The problem is that each of those partners is an independent organisation with its own change-control process, its own testing appetite, and no interest whatsoever in your Azure roadmap. You cannot ask 140 counterparties to coordinate a cutover with you. So the design constraint is simple to state and hard to honour: the migration must be invisible from the outside.

You cannot ask 140 counterparties to coordinate a cutover with you. So the migration must be invisible from the outside.

Here is what that requires.

The only five things a trading partner can observe Everything behind the boundary can change freely: BizTalk or Azure, the mapper, the hosting. A trading partner observes only five things, and holding all five stable is what makes an EDI migration invisible from the outside. The endpoint hostname and IP, because partners keep firewall allowlists. The certificate, because a change is a change-control event at every partner. Acknowledgement type and timing, because partners branch on them and build timeouts around them. Control numbers, which must increment monotonically and never repeat or regress. And document content, down to segment-level quirks. Hold five things stable and the migration is invisible Your partners did not agree to your Azure roadmap. Fortunately, there are only five things they can actually see. YOUR SIDE Change freely BizTalk → Azure The mapper Hosting and network Orchestration engine Monitoring and alerting Dead-letter handling None of this is visible from the other side. THE BOUNDARY 1 Endpoint hostname and IP Partners keep firewall allowlists — audit these first 2 The certificate A change-control event at every partner, all at once 3 Acknowledgements — and their timing 997, CONTRL, MDN — partners branch on the type 4 Control numbers — the trap Must increment, never repeat, never regress 5 Document content Down to the quirks BizTalk quietly tolerated for a decade THEIR SIDE 140 partners Own change control Own testing appetite Own retry logic No interest in your Azure roadmap You cannot ask them to coordinate a cutover with you.
Everything behind the boundary can change freely. A trading partner sees only five things — hold those stable and the migration is invisible from the outside.

What partners actually observe

Strip away the internals and a trading partner sees exactly five things:

  1. The endpoint they send to — URL, IP, SFTP host.
  2. The certificate used to sign and encrypt.
  3. The acknowledgements they get back — 997, CONTRL, MDN — and their timing.
  4. The control numbers in what you send, and their continuity.
  5. The content of your outbound documents, down to segment-level quirks.

Nothing else. Not your platform, not your mapper, not your hosting. Which means migration invisibility reduces to holding those five stable while everything behind them changes.

Four of the five are straightforwardly achievable. One is a genuine trap.

Keep the endpoint, move what is behind it

An AS2 URL or SFTP host is a DNS name and a certificate. Both can point at new infrastructure without the partner knowing, provided you preserve:

  • The hostname, so partner firewall allowlists stay valid.
  • The certificate, or a properly cross-signed successor — a certificate change is a partner-side change, and therefore a change-control event at 140 organisations.
  • The IP address, or your partner’s IP allowlist becomes your problem. This one is easy to miss and it is the most common cause of “we cut over and one partner went dark.”

Audit which partners restrict by IP before you plan the network topology, not after. The answer is usually a small minority, and it constrains your architecture.

Control numbers are the trap

Interchange, group, and transaction control numbers must increment monotonically and must not repeat. Your partner’s system will reject a duplicate, and — worse — many will silently reject a regression to a lower number and simply stop acknowledging.

In BizTalk those counters live in the party configuration in a SQL database. If the new platform starts its own sequence at 1, every outbound interchange to that partner is invalid.

So: extract current control numbers per partner per document type before cutover, seed the new platform above the current value with a deliberate gap, and confirm the sequence is persisted somewhere durable rather than held in memory. Then verify on the first live interchange rather than assuming.

This is a fifteen-minute task that, skipped, produces a week of partner escalations. It is the single highest-value item on the EDI migration checklist.

15 min

Extracting current control numbers, seeding the new platform above them with a deliberate gap, and confirming the sequence persists durably is roughly a fifteen-minute task. Skipped, it produces a week of partner escalations — and it is the highest-value item on the whole checklist.

Acknowledgements are a contract, including their timing

A 997 functional acknowledgement is not a courtesy — for many partners it drives their own retry logic and exception queues. Three properties matter and all three are easy to change accidentally:

  • Whether you send one at all, per partner per document type. BizTalk party configuration encodes this granularly and it does not survive a careless rebuild.
  • What it says. Accept, accept-with-errors, and reject are distinct, and partners branch on them.
  • When it arrives. A partner whose process assumed a 997 within fifteen minutes may have built a timeout at twenty. A new platform that acknowledges in four hours because it batches is technically compliant and operationally broken.

Export the acknowledgement configuration from BizTalk as a table — partner, document type, ack type, expected latency — and treat it as an acceptance criterion for each wave.

Sequencing EDI trading-partner waves by consequence, not by count Waves ordered by consequence rather than volume. Wave one moves two or three low-volume, low-consequence partners with responsive technical contacts; its purpose is to discover what the runbook is missing. Wave two moves the well-behaved majority with standard document types; it should feel repetitive, and repetitive is the goal. Wave three moves the awkward partners with non-standard segment usage, unusual acknowledgement expectations, high volume or contractual service levels, by which point the team has real judgement and a rehearsed rollback. A partner in the middle of a commercial renegotiation never moves. Each wave runs both platforms in parallel with daily reconciliation, and the parallel window must include a month-end. Order the waves by consequence, not by count The instinct is to move partners in bulk to show progress. Resist it — wave one exists to find what your runbook is missing. WAVE ONE 2–3 partners Low volume, low consequence, responsive technical contact. Purpose: discover what the runbook is missing. WAVE TWO The well-behaved majority Standard document types, no quirks, moderate volume. Should feel repetitive. Repetitive is the goal. WAVE THREE The awkward ones Non-standard segments, unusual ack expectations, high volume, SLAs with teeth. By now the team has judgement and a rehearsed rollback. NEVER A partner mid-renegotiation Not into an open commercial conversation. EVERY WAVE Parallel run + daily reconciliation Received, acknowledged, rejected — and include a month-end. A divergence where the new platform accepts what the old one rejected is also a finding — that is how bad data reaches downstream systems.
Waves ordered by consequence rather than volume. Wave one exists to find what your runbook is missing, not to show progress.

Sequence the waves by consequence, not by count

The instinct is to move partners in bulk to show progress. Resist it. Order the waves this way:

Wave one — two or three low-volume, low-consequence partners. Ideally ones with a responsive technical contact who will tell you if something looks odd. The purpose of wave one is not throughput; it is to discover what your runbook is missing.

Wave two — the well-behaved majority. Standard document types, no custom quirks, moderate volume. This is where most of the estate moves, and it should feel repetitive. Repetitive is the goal.

Wave three — the awkward ones. Non-standard segment usage, unusual acknowledgement expectations, high volume, or a partner whose contract has an SLA with teeth. By now your team has real judgment about the new platform and a rehearsed rollback.

Never — the partner who is mid-renegotiation. If a commercial conversation is open with a partner, do not introduce a technical change into it. Wait.

Run acknowledgement reconciliation in parallel

For each partner in the current wave, run both platforms for a defined period and reconcile daily on three counts: interchanges received, interchanges acknowledged, and documents rejected. Any divergence is a finding, including — especially — a divergence where the new platform accepts something the old one rejected. Silent new permissiveness is how bad data enters downstream systems.

Two weeks of clean reconciliation, including at least one month-end, is a reasonable bar before declaring a wave complete. Month-end matters because volume patterns and document mixes change, and because the unusual messages tend to appear then.

What this looks like on Azure

For the mechanics: X12 and EDIFACT processing, AS2 transport, and partner agreements are available through Logic Apps with an Integration Account, and for many estates that is a clean destination. Where it gets more involved is high partner counts with heavily customised agreements, very large or heavily-conditional maps, and volume that makes per-action billing material — the same three gaps we cover in our Logic Apps assessment, and part of why Art2link ESB exists.

The platform choice matters less than the sequencing discipline. We have seen well-architected Azure EDI platforms damage partner relationships through careless cutover, and unremarkable ones migrate 140 partners with nobody outside the company noticing. The second outcome is the goal, and it is a project-management achievement as much as a technical one.

The checklist

Before any partner moves

Nine checks. The control-number line is the one that, skipped, costs a week.

  • Endpoint hostname preserved
  • Certificate preserved or cross-signed
  • Outbound IP addresses confirmed against partner allowlists
  • Control numbers extracted, seeded above current, persisted durably
  • Acknowledgement configuration exported per partner and document type
  • Acknowledgement latency expectations documented
  • Rollback rehearsed for this wave
  • Reconciliation running on both platforms
  • Month-end included in the parallel window

Cerebrum City has run EDI in production for two decades across manufacturing, healthcare, logistics, and financial services — X12, EDIFACT, AS2 and AS4 — as a Microsoft V-TSP–recognized BizTalk specialist. Our EDI integration practice is where that work lives, and an assessment will tell you how many of your partners are in wave three.


Jorge Pastorini · Cerebrum City

Cerebrum City is a Microsoft V‑TSP–recognized BizTalk specialist in Atlanta, Georgia, with two decades running BizTalk Server in enterprise production for clients including Humana, Deloitte, JetBlue, and U‑Haul.

Courtesy assessment

Want this mapped against your BizTalk estate?

We inventory your ports, orchestrations, maps, and adapters, flag the interfaces most likely to break, and hand back a staged Azure roadmap. No obligation.

Keep reading

All insights →

Ready to modernize your stack?

From integration pioneers to cloud-native experts — start with a courtesy assessment and we’ll map the fastest path to a modern architecture.

Talk to an Integration Architect