MES, WMS and planning portfolios: which projects create value in 2027, which destroy it, and a method to tell them apart

Every autumn, manufacturing and supply chain leaders arbitrate a list of projects that looks the same from one year to the next: an MES upgrade, a WMS replacement, an advanced planning tool, a rebuild of the ERP interfaces and, for the past two years, one or more projects labelled AI. The list is long, budgets are not, and the business cases that reach the steering committee are almost all written by the people who want the project.

After twenty years spent launching, repairing and sometimes stopping these systems, we at Strepia hold a conviction that vendors and integrators rarely enjoy hearing: most application rebuilds in production and logistics destroy value for their first two years, while most data and interface reliability programmes create value within the first quarter. Steering committees nevertheless fund the former and postpone the latter, because the former come with a vendor name and a demo, and the latter come with a flow diagram.

This is not just a field impression. In its 2026 ERP Report, based on 170 organisations surveyed between January 2025 and January 2026, Panorama Consulting finds that more than a quarter of projects exceed their budget and almost a quarter their schedule. The leading cause of budget overrun is not labour: it is an unexpected need for additional technology, discovered late, usually because the mismatch between the tool and the shop floor was only seen mid-flight. The leading cause of schedule overrun is organisational: governance, resistance to change, process redesign^[1]. Those are precisely the two blind spots of the business cases we see reaching committees.

This article offers a method to sort, in a single meeting, what deserves a budget in 2027. It rests on three technical questions that project dossiers almost never address, and on a grid you can apply tomorrow.

1. The question dossiers avoid: where does the value come from, mechanically?

An MES or WMS business case promises gains in productivity, quality and service levels. It almost never states the technical mechanism through which those gains will materialise. Yet in a plant or a warehouse, value comes from only four places:

1. A decision taken earlier: releasing an order, reassigning a line, triggering a replenishment before the stock-out.
2. An error avoided: a wrong consumption declared, a wrong batch shipped, a wrong quantity counted.
3. A gesture removed: double entry, manual reconciliation, recounting, re-keying after an interface rejection.
4. A stoppage shortened: an incident diagnosed in one hour instead of one day.

Every project in your portfolio should be traceable to one or more of these mechanisms, with a figure and a source. A project that cannot be traced is not a bad project: it is a project whose return nobody knows yet. In committee, that alone should push it behind the ones whose return is known.

Take the classic case of replacing a WMS "because the vendor ends support in three years". End of support is a real risk, but it is not a value mechanism. The project will cost a year of parallel running, a data migration, a rewrite of every interface and the temporary loss of the automations teams have built into the old tool. For eighteen months it will consume more than it returns. That is acceptable only once you have demonstrated that the existing system cannot be secured for five years at a fraction of the cost, and that demonstration is rarely made.

2. The second question: what is your integration debt?

The factor that decides whether an MES or WMS project succeeds is almost never the software chosen. It is the state of the interfaces around it and of the data flowing through them.

A typical production site or warehouse runs between fifteen and forty flows between its ERP, MES, WMS, planning tool, automation and quality systems. On the sites we have audited, a significant share of these flows shows at least one of the following defects:

• no functional acknowledgement: the sender knows the message left, not that it was processed;
• no idempotency: replaying a message after an incident creates a duplicate (a production order declared twice, a receipt counted twice);
• no ordering guarantee: an order closure can arrive before the last consumption declaration;
• divergent master data: units, item codes, bills of materials or locations that do not carry the same definition on both sides;
• no business monitoring: a flow can stop without anyone who understands its content being alerted.

This is what we call integration debt. Its cost is rarely measured, but its effects are documented: Panorama Consulting, in its 2026 analysis of MES failures, ranks ERP-to-MES data misalignment (items, routings, bills of materials, labour standards, quality specifications) among the leading causes of teams losing trust and reverting to spreadsheets^[2]. And when Deloitte asks 600 manufacturing executives about their investment priorities for the next two years, data foundations come first, ahead of AI itself^[3]. Above all, integration debt has a property committees underestimate: every application project launched on top of it inherits it in full. Changing the WMS does not fix a wrong bill of materials in the ERP. Deploying advanced planning on production declarations that arrive four hours late produces an optimised plan for a plant that does not exist. Connecting an AI agent to these flows teaches it the inconsistencies and has it restate them with confidence.

The practical consequence is counter-intuitive: the most profitable project in most 2027 portfolios has no vendor name. It is a six-to-twelve-week programme that maps the flows, measures their defects, puts acknowledgement, idempotency and monitoring on the ten critical flows, and aligns master data. It costs a fraction of a rebuild, pays back on completion, and conditions the success of everything that follows, AI included.

3. The third question: where is your human critical path?

Project schedules measure tasks; projects fail on people. On an MES or WMS programme, three roles sit on the critical path and can be neither parallelised nor replaced mid-course:

• the master data owner (items, bills of materials, routings, locations), who decides what is true when two systems disagree;
• the integration architect, who knows the flows end to end and knows what will break at cutover;
• the site operations manager, who will say whether the target process holds at three in the morning with the night shift.

If one of these three roles is not identified, available at least half-time and empowered to decide, the project will slip whatever the schedule says. And these profiles are the scarcest on the market: in the same Deloitte survey, 46% of manufacturers report moderate to significant difficulty filling planning and scheduling roles, and 48% for operations management^[3]. The integration architect does not even appear in the surveys, because few companies know they need one until cutover. Before funding a project, the committee should ask who holds these three roles and how long each has been with the company. A project whose integration architect is a contractor who arrived last month carries a risk the budget does not reflect.

4. The arbitration grid

These three questions combine into a simple grid. For each project in the portfolio, five scores from 0 to 3:

A project scoring below 7 out of 15 is not ready to be funded; it is ready to be investigated. A project above 11 deserves to be accelerated. In between, time to first gain decides.

Applied to the portfolios we see, this grid almost always produces the same reshuffle:

• Moving up: flow and master data reliability; business monitoring of interfaces; extending an existing system to a new site or flow; 30-day AI experiments on an instrumented perimeter.
• Moving down: tool replacements motivated by end of support alone; advanced planning deployed before declarations are reliable; "platform" AI projects without a measurable use case.
• Changing nature: the large rebuild becomes a sequence of three shorter programmes, each reversible, the first of which is almost always the integration debt.

5. What this means for AI

A word on AI projects, since they now occupy a line in every portfolio. A model, whatever it is, is worth only the data it is shown. On a site where production declarations arrive late, where logical stock diverges from physical stock and where flows lose messages without an alert, AI learns those defects and amplifies them with a confidence no experienced operator would ever have.

The 2025 figures leave little room for doubt: MIT's study of AI in business concludes that roughly 95% of organisations measure no return on their generative AI pilots, and that the 5% that succeed stand out through systems integrated into real workflows and able to learn from their context^[4]. Gartner, for its part, predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027, for reasons of cost, unclear business value and inadequate risk controls^[5]. This is not an algorithm problem; it is a flow, master data and integration problem, and it is solved upstream.

The right order is therefore always the same: instrument, make reliable, then and only then learn. In that order, an agent connected to an MES or a WMS detects anomalies no dashboard shows, and the return is fast and measurable. In the reverse order, you pay a licence to obtain elegant explanations of wrong data. We will devote a full article to this in February, with the report of a real experiment on test systems.

In practice

If you are preparing your 2027 arbitrations, here is what we would do in your place this week:

1. Run every project in the portfolio through the grid, in session, with the project owners present. The exercise takes an hour and almost always changes the order of the list.
2. Have the integration debt measured on the ten most critical flows before approving anything that depends on them. That is two to three weeks of work for someone who knows where to look.
3. Name the three critical-path roles for each retained project, and check they are genuinely available.

The download: the arbitration grid in spreadsheet format, with detailed criteria, the questions to ask each project owner and a worked example, is available below.

If you lack the time to do it yourself: we apply this grid to your portfolio in a thirty-minute conversation, based on your project list and a description of your main flows. You leave with a reasoned ranking you can present to your committee. Write to us with your list and we will propose a slot within 48 hours.

Strepia brings together senior experts who have spent twenty years designing, integrating and launching MES, WMS and integration systems for manufacturing and logistics sites, at software vendors, at integrators, then directly for large accounts. Strepia places these experts as reinforcements and delivers three-day flash audits.