A highways engineer running the resurfacing of a major A road has done her job properly. Water, gas, electricity, telecoms, every utility coordinated before the resurfacing crew arrived. The new surface goes down on Friday. The road reopens on Monday. The KPIs on closures will land where they need to.

Then an email arrives. A new fibre rollout has been mandated. Twenty-four months to compliance.

The irony is not lost on her as she sits at temporary lights on her commute home, staring at the cones and the open trench, as she runs the numbers. Lay it now, while the trench is still open and the contractors are still on site, and the cost is marginal. Lay it later, after the surface has bedded in, and the bill is an order of magnitude higher. The complaints are guaranteed. The KPIs she has just earned will be undone by the next team to come through.

The right call is obvious in hindsight. The harder question is who raises it before Friday.

This is the pattern playing out inside S/4HANA programmes at scale right now, with a different name on the door.

S/4 programmes are complex, expensive, and run by professionals under intense pressure on time, cost, and scope. Things get missed, not through negligence, but because programme governance optimises for go-live. It has to. There is no other way to land a transformation of this complexity. Scope is controlled. Dependencies are managed. Risk is contained. The integration workstream is given a clear brief: migrate the existing partner connectivity, retest the connections, keep the lights on, go live clean.

That brief was reasonable when it was written. The world around it has moved.

At SAP Sapphire in May 2026, the company announced the Autonomous Enterprise as its platform direction. More than 200 domain-specific Joule agents. Autonomous Spend covering procurement, sourcing, supplier management, and invoicing. Autonomous Supply Chain embedding AI into planning, manufacturing, logistics, and business networks. This is not a research agenda or a three-year roadmap. It is product, with delivery dates measured in quarters.

That changes what the integration workstream needs to deliver. The brief was written before the platform vendor announced autonomous agents that need to reason across the supply base in real time. The scope was set before the Autonomous Enterprise existed. Now it does. And the brief has not caught up.

The result is that a connectivity decision with a twenty-year horizon gets made inside a six-week workstream, by people who would be the first to tell you they were not asked to make a twenty-year decision. They were asked to keep the lights on.

Here is the uncomfortable truth from the field: nobody actively chooses EDI over the more evolved SAP Business Network. It persists by default. When the question “are we losing agility to the default?” is posed internally, the answer tends to be the same: nothing is visibly broken. The architecture works. And for the largest trading partners, the case is even easier to dismiss. High-frequency document exchange, structured, reliable, largely invisible to the operational business. It just works. That is agility of a kind: speed of execution along a fixed path. But it is not agility of direction. It is not the ability to onboard a new supplier in hours, ask a question the format was never designed to carry, or support a process that did not exist when the connection was built. When the business needs to change direction, the architecture that “just works” is the architecture that only works one way.

The connectivity decision is not made. It is assumed. The question is never asked because the people in the room were not asked to answer it. They were asked to keep the lights on. And keeping the lights on means rebuilding what was already there. The on-premises middleware gets migrated to a managed cloud equivalent. The pipe gets rebuilt. The programme goes live. Everyone moves on.

The managed cloud equivalent is better than what it replaced. Resilience improves. Partner onboarding gets faster. That alone justifies the migration. None of it changes what the architecture is. It is still bilateral. It is still a document moving between two known parties, accurate at the moment of transmission and silent the moment anything changes.

And it still costs more to run than most organisations realise. The headline figure from EDI providers is 4-6 cents per document. The effective blended cost, once platform fees, minimums, overage charges, middleware maintenance, staffing, and error recovery are included, is closer to €70,000-€300,000 per year for a mid-market enterprise (Orderful; IOFM; PeerSpot; independently validated by IDC). Not setup cost. Running cost.

EDI has always been a top-tier capability. The largest trading partners by volume or spend are connected electronically. The rest, often the significant majority, are still sending purchase orders by email, confirming shipments by PDF, reconciling invoices from paper. Billentis estimates that only 29% of B2B invoices globally are exchanged in a structured electronic format (E-Invoicing / E-Reporting market report, 2026). In most enterprises, EDI reaches 20-50% of suppliers by count; the remaining 50-80% operate through manual channels. A modern ERP transformation that leaves that untouched is building a digital core with an analogue perimeter. When autonomous agents need structured data from across the supply base, not just the largest partners, that gap becomes the constraint.

If you have read the earlier posts in this series (blog 1, blog 2) the architecture argument will be familiar: an autonomous agent can only act on what reaches its system. When that data arrives through EDI, it is limited to predefined document types defined decades ago, as current as the last scheduled transmission, from only the fraction of suppliers with bilateral connections. SAP Business Network changes all three: richer data types, continuously refreshed, from a broader supply base. The question now is what to do about it inside a live programme, before the integration workstream closes and the decision becomes permanent.

The temptation is to call it a Phase 2 problem. Get to go-live first, deal with how the organisation collaborates with its trading partners later. That instinct is understandable. It is also one of the most expensive decisions the programme can make. Suppliers confirm the reality bluntly: “As everyone knows, phase two never happens.” The migration window is the only window. If the connectivity architecture is not addressed during the programme, it will not be addressed at all.

Enterprise benchmarks from integration consultancies (Remedi, Ecosio, and compiled SAP data) put the cost of onboarding a single supplier through traditional EDI at between €4,000 and €12,000, and 4 to 12 weeks of elapsed time, once the core platform is operational. For a wave of 300 suppliers, that is a programme budget of €1.2 to €2.4 million and 9 to 24 months to fully deploy. Every wave added later incurs the same cost again. The long tail remains untouched.

A supplier already on SAP Business Network requires no bilateral setup. The connection exists. The data is already structured. Autonomous agents can already reach it. When a buyer begins enablement, typically 50% to 70% of their supplier base by count is already on the network, transacting with other buyers (SAP Business Network Q1 2026 enablement data). Those suppliers are auto-matched and auto-connected without bilateral testing.

For the remaining suppliers, the economics have shifted. AI-assisted OnBoarding, powered by Joule, auto-extracts supplier data from S/4HANA, auto-matches against the existing network membership, auto-connects those already present, and auto-invites the rest. SAP's early estimates indicate a 75% reduction in manual supplier identification, extraction and follow-up activity, and a 31% improvement in onboarding cycle time. Work that previously took 70+ hours per customer reduces to minutes of agent execution. Suppliers not ready for technical integration can start transacting by clicking a link and using the web interface. No integration project required. This is AI applied to a real process: automating extraction, matching, and outreach, and presenting back to the human only what needs attention. The person still decides. The machine does the work that used to consume the week.

Network coverage varies by industry and geography, and no network reaches every supplier on day one. But every supplier onboarded to the network is one that never needs to be onboarded again, and the economics of activation versus bilateral setup compound with every wave. Independent research confirms the pattern: organisations using network-based supplier connectivity report 70% faster onboarding and 60% reduction in invoice-related FTEs (IDC, Business Value of SAP Business Network, March 2025).

Two points the integration workstream gets wrong when it's framed as “EDI or network.”

First: it is not a choice between the two. EDI is a format. The network is a platform. Suppliers already running EDI can route those documents through the network rather than through isolated point-to-point pipes. They do not change format. They change platform: from one that carries only documents to one that adds visibility, business rules, compliance, and multi-process capability. The existing EDI investment is not discarded. It is redirected.

Second: on a shared-participation network, every document flow follows one process regardless of how the supplier connects. EDI, cXML, web interface, invoice conversion: same workflow, same business rules, processed identically. One process, regardless of the supplier's technical maturity or the country's regulatory model. Take invoicing: when you are facing mandatory e-invoicing in Belgium, Poland, France, Germany, and the EU's ViDA reforms by 2030, that single-process architecture is what makes compliance at scale manageable rather than unmanageable.

This is the economic shape of the road analogy. Lay the fibre while the trench is open and the marginal cost is small. Come back later, after go-live, and the cost compounds with every wave. The programme that thought it had protected scope has not protected scope. It has deferred cost, at a premium, to a successor team that will pay the bill without ever having been in the room when the choice was made.

The intervention is small, and it has to happen while the trench is still open.

SAP Business Network is included with both RISE and GROW with SAP subscriptions. For customers on these programmes, the entitlement is already in the contract. This is not additional spend. It is activation of a capability the organisation has already paid for. A cross-functional design sprint, two days, procurement, IT, supply chain, and architecture in the same room, is enough to surface the decision and frame it correctly. One question on the agenda: can the connectivity plan we have today support the autonomous products our platform vendor is already delivering. Not an architectural redesign. Not additional scope. A conversation that either confirms the current plan or surfaces a decision that is better made now than discovered later.

When that conversation happens, the programme does not slow down. It gets sharper. The integration workstream delivers against a brief that accounts for what the platform is actually becoming, not just what it replaced. The connectivity layer is designed once, for the architecture the business will need, not retrofitted later at a multiple of the cost. The programme sponsor can look at the governance board and say the decision was made by design, not by default.

The surface goes down on Friday. The fibre mandate has already been issued.

Unlike the road, this surface does not come up again on a five-year cycle. The next time the integration workstream is genuinely open for redesign is a decade away, if it happens at all. Whatever gets decided in this programme, by default or by design, is the answer for the working life of most people currently in the room.

The trench is open today. Someone in your integration workstream is writing the plan right now. The only question is whether anyone has told them about the fibre.