Integration

Naming iFlows in SAP Cloud Integration: What Two Years of Production Taught Us

Vlado Balko

A revised, field-tested version of our original naming guide.

Two years ago I published A Practical Guide to Naming iFlows in SAP Cloud Integration (SCI). It described a clean scheme: a numeric ID, a direction, a partner name, the adapters.

Then we ran it. Across several customer landscapes, hundreds of artifacts, multiple teams, and a few years of change requests. The convention held — but not in the shape we wrote down. Real projects bent it, and the bends turned out to be improvements. Other parts sagged in ways worth warning you about.

This is version two: the same idea, reshaped by what actually survived.

What changed, in one paragraph

The original scheme said nothing about landscapes or standby twins, and gave a single <partner> slot that couldn't express a SAP-to-SAP flow. It defined IN / OUT without saying inbound to what. And its ID packed a message class and a partner lookup into eight digits — elegant, and quietly abandoned within months, because in practice people number by project, not by partner. Version two fixes all four.

The template

SCI[<env>]<project><sequence>_<direction>_<source>_<target>_<srcProto>_<tgtProto>_<message>[_<variant>]

A real one, on a multi-tenant landscape:

SCI1100010_OUT_PAYROLL-DE_BANKGW_SFTP_SFTP_PaymentOrder

And the same flow where all three stages share one tenant:

SCIP1100010_OUT_PAYROLL-DE_BANKGW_SFTP_SFTP_PaymentOrder

SCI — platform label

Constant. It looks redundant inside Cloud Integration, but names leak: into ticket subjects, monitoring exports, Excel handovers, and the inbox of someone who has never opened your tenant. Keep it.

<env> — one letter for the landscape, if you need one at all

D development · T test · P production

And here is the part we got wrong for a while, which is worth more than the letter itself.

The landscape letter is only useful on a single tenant. If Dev, Test and Prod are three separate SCI tenants — as they are in most serious setups — the tenant is the landscape. Encoding it again in every artifact name buys you nothing and costs you something real: the name of a flow now changes as it’s transported, so ticket references, runbooks, and endpoint URLs all drift between stages. Leave the letter out.

Keep it only when one tenant genuinely hosts several stages, which does happen — small landscapes, PoCs, cost-constrained tenants.

Then what about backup and internal? This was the actual bug in our scheme: one character was carrying three unrelated axes.

  • Landscape (D / T / P) is a property of the tenant. Drop it when the tenant already answers the question.

  • Standby, backup, blue/green is a property of one artifact inside a tenant — the twin lives right next to the original, in the same tenant, whatever your landscape model. So it is never a landscape letter. It belongs in the _<variant> suffix: _blue, _green, _standby.

  • Internal is a direction, not a landscape and not a variant. It’s INT, in the next token. Our early draft had I as an env letter; that's the same word doing two jobs, and it will confuse someone within the month.

One consequence to accept deliberately: with the letter optional, prefix length differs between landscape models — SCI1100010 versus SCIP1100010. That's fine, as long as it's fixed within one landscape and your validation regex knows which model you're on.

<project><sequence> — seven digits, zero-padded

Two digits for the project or business area (11 payroll, 13 banking, 07 partner logistics), five for the flow itself.

Sequences step by 10 — 00010, 00020, 00030 — leaving room to insert. Helper flows attach to their parent with +1: the request flow is 00110, its response handler 00111, its post-processor 00112. A support engineer reading SCIP0700111 in an alert knows instantly it belongs to 00110.

Zero-pad without exception. Seven digits, always. This single rule carries the entire value of a numeric prefix, and it is the first one your team will break.

<direction> — from the customer's perspective

IN into the customer's landscape · OUT out of it · INT between the customer's own systems · EXT between two systems that are both outside it

The anchor clause matters more than the list. “Inbound” is meaningless until you name the thing it is inbound to. Write the perspective down, once, in the convention itself — otherwise every developer picks their own, and half your names invert.

<source> and <target> — systems, not partners

Short names from a closed list. Where an endpoint is country-specific, append the ISO 3166-1 alpha-2 code: PAYROLL-DE, WMS-PT.

Splitting v1’s single <partner> into two slots is what finally let the scheme describe internal integrations, hub-and-spoke routing, and three-legged flows where the partner isn't either endpoint.

<srcProto> and <tgtProto> — protocols, from a closed list

HTTP SFTP AS2 IDOC SOAP ODATA AMQP JMS MAIL FTPS PD (ProcessDirect) JOB

Closed list, one spelling each. Pick PD or ProcessDirect, ODATA or OData, and never the other. Mixed spellings break every search and sort you built the convention for.

<message> — the business content

PaymentOrder, InboundDelivery, MasterData. This is what a business analyst reads. v1 had it; our production practice let it dissolve into free text. Keep it as a real, required slot.

<variant> — optional, and from a closed list

manual · migration · retry · resp · blue / green · standby

The hard-won rule: a free-text tail becomes a landfill. Ours accumulated _tmp, _bkp, _test, _manual, _migration, _maninput2, and one _manualTestFinal before anyone noticed. If a variant isn't on the list, it doesn't go in the name — it goes in the description field.

Two worked examples

SCIP0700110_OUT_ESB_PARTNERWMS_HTTP_AS2_DeliveryConfirm

Production. Outbound from the customer’s landscape. From the internal ESB to a partner warehouse system, HTTP in, AS2 out, carrying delivery confirmations. Project 07, flow 00110.

SCIP0700111_OUT_ESB_PARTNERWMS_HTTP_AS2_DeliveryConfirm_resp

The response handler for exactly that flow — +1 says so before you read another character.

How naming conventions actually rot

Every failure below is one we made, in a live tenant, having written the rules ourselves.

Padding drifts. 130050 shows up next to 1300010 and 10000010. Six, seven, eight digits, same registry. Sorting is gone, and sorting was the point. Validate the length; don't trust discipline.

One slot ends up carrying several ideas. Our landscape letter accumulated P and T (genuinely landscapes), B (a standby twin, which is not a landscape), and I (internal — which is a direction). Three unrelated axes, one character. Backups then also picked up a _bkp suffix, and _bkp_green on top, so no single query finds all of them. One axis per slot, one slot per axis.

Optional tokens quietly become absent tokens. A handful of utility flows shipped with no direction at all. Once a slot is skippable, position-based parsing dies. If it’s in the template, it’s mandatory.

Metadata contradicts the name. We have a P-prefixed artifact sitting in a test package, and country codes that disagree with their own registry column — BGR against BG, RS against SR, and one payroll flow labelled SD for what the registry calls France. The name and the registry must be checkable against each other, or one will lie.

The registry becomes a place things are copied into. Six byte-identical rows for the same artifact. Uniqueness needs a constraint, not a convention.

Notice that none of these are design flaws. The scheme was fine. What was missing was enforcement — and a naming convention with no validation step is a style guide, not a standard.

The three rules that keep it alive

  1. One registry, and it is the authority. Project codes, system short names, protocol abbreviations, and variant keywords all come from it. Notion, Confluence, a repo — the tool is irrelevant; single-source is not.

  2. Uniqueness is enforced, not assumed. Project codes unique globally; sequences unique within a project; the full artifact name unique per landscape.

  3. Names are validated before deploy. A regex in a pipeline, or a checklist item in review. Fifteen minutes of work, and it is the difference between the v1 article and this one.

Conclusion

The original guide wasn’t wrong. It was untested. What two years added was a landscape letter, an honest perspective anchor, two endpoints instead of one partner, project-based numbering, and closed lists everywhere a human might improvise.

And the real lesson isn’t in the template at all: a convention nobody validates degrades into a suggestion. Write the regex.

The original version stays up for anyone who wants to see where this started.

Addendum — extension topics

Optional reading. The core convention above is complete without it.

Name vs. ID: which one does this govern?

Both, and this trips people up. An SCI integration flow has a Name (display, freely editable) and an ID (technical, embedded in the artifact, painful to change once deployed and referenced).

Apply the convention to both, identically. A flow whose Name and ID disagree is worse than one with no convention at all, because tooling reads the ID while humans read the Name, and your monitoring and your support engineer will describe different worlds.

Practical consequence: get the ID right at creation time. Renaming an ID after deployment means undeploy, edit, redeploy, plus updating every ProcessDirect address, every partner-facing endpoint, and every reference in a value mapping that points at it.

Length and character limits

Keep names under roughly 120 characters. Names propagate into HTTP endpoint addresses, ProcessDirect URIs, monitoring filters, and CSV exports — and something in that chain will truncate. Stick to A–Z a–z 0–9 _ -. No spaces, no dots, no diacritics.

SCIT0700114_OUT_BIZTALK_PARTNERWMS_SFTP_AS2_OutboundDeliveryReplication is already 70 characters and comfortable. ..._SHP_OBDLV_SAVE_REPLICA_bkp_green is where it stops being comfortable.

Adopting this in a tenant full of legacy flows

Do not mass-rename. The cost is real (undeploy/redeploy, endpoint churn, partner coordination) and the benefit is cosmetic on flows nobody touches.

Instead:

  • All new artifacts follow the convention from day one. No exceptions, including throwaways — especially throwaways, since those are the ones that survive.

  • Rename on substantial change. If a flow is being reworked and redeployed anyway, the marginal cost is near zero.

  • Leave stable legacy flows alone, but record them in the registry with their would-be name. You get searchability without the risk.

  • Never rename a partner-facing endpoint without a coordinated change window.

Expect a mixed tenant for a year or two. That’s not failure; that’s the honest cost of naming conventions arriving after the integrations do.

Sibling artifacts

The same numeric prefix, reused, is what ties a solution together:

  • Packages — <project code> <business area>, e.g. 13 Bank Integrations. One package per project code per landscape.

  • Value mappings — SCI<project>_VM_<purpose>

  • Script collections — SCI<project>_SC_<purpose>

  • Message mappings — SCI<project><sequence>_MM_<message>, sharing the flow's sequence

  • Credentials / OAuth — <SYSTEM>_<usage>, no flow number; they're shared and outlive individual flows. Add a stage marker only if one tenant really holds credentials for several stages — the same rule as the <env> letter, for the same reason.

Registry columns worth having

Artifact Name · ID · Project code · Landscape · Direction · Source · Target · Source protocol · Target protocol · Message type · Country · Package · Owner · Status · Deployed-on

The last three are what turn a naming table into something operations will actually open.

Blue/green and standby twins

Always the variant suffix, never a prefix letter — and the reason is structural, not stylistic.

A twin exists alongside its original, in the same tenant. A landscape marker answers “which stage is this?”, and both twins give the same answer. So encoding the twin as a landscape letter (SCIB0700110_...) mislabels it, and worse, it makes the twin sort away from the flow it shadows. SCIP0700110_..._blue and _green sit next to each other in the artifact list, exactly where you want them during a cutover.

What we did before was use the letter and a _bkp suffix and combinations of both — which is how you end up with SCIP08000100_IN_..._bkp_green and a support engineer at 2 a.m. who cannot tell you which artifact is live.

If you also keep a cold standby distinct from an active blue/green pair, that’s _standby. Another suffix, same slot.

Single tenant vs. multi-tenant, restated

The one decision to make before you adopt any of this:

  • Separate tenants per stage (the common case) — omit the <env> letter. The tenant is the landscape. Names stay identical as an artifact is transported, so every reference to it survives the trip.

  • One tenant, several stages — keep the letter, first position, D / T / P only. Never overload it with twins or directions.

Either way, twins are suffixes and directions are the direction token. The letter, when present, means exactly one thing.

Why numeric prefixes at all?

The objection is fair: names are searchable, so why encode meaning in digits? Three reasons that only show up at scale.

Stable identity. Partners get renamed, acquired, and merged. SCIP0700110 doesn't change when PARTNERWMS becomes something else, so tickets, runbooks, and monitoring dashboards from three years ago still resolve.

Sortability. An artifact list ordered by name groups every flow of a project, in flow order, with helpers directly beneath their parents. Alphabetical sorting by system name scatters exactly the things you need together.

Conversational compression. “Check 0700111” is a sentence a support engineer can say on a call. SCIP0700111_OUT_ESB_PARTNERWMS_HTTP_AS2_DeliveryConfirm_resp is not.

The digits are for machines and for tired humans at 2 a.m. The words are for everyone else. That’s why both halves are in the name.

SMARTER TECH, SMARTER BUSINESS

Cloud, AI a parťáci. Integrace, vývoj, automatizace a AI na SAP Business Technology Platform a Microsoft.

Napište nám →

© 2026 CLAIMATE · CLOUD · AI · MATES

SAP PartnerEdge · Microsoft Partner

SMARTER TECH, SMARTER BUSINESS

Cloud, AI a parťáci. Integrace, vývoj, automatizace a AI na SAP Business Technology Platform a Microsoft.

Napište nám →

© 2026 CLAIMATE · CLOUD · AI · MATES

SAP PartnerEdge · Microsoft Partner

SMARTER TECH, SMARTER BUSINESS

Cloud, AI a parťáci. Integrace, vývoj, automatizace a AI na SAP Business Technology Platform a Microsoft.

Napište nám →

© 2026 CLAIMATE · CLOUD · AI · MATES

SAP PartnerEdge · Microsoft Partner

Integration

Naming iFlows in SAP Cloud Integration: What Two Years of Production Taught Us