AI for Supply Chain Build or Buy
The Build or Buy decision in supply chain planning has changed. Generative AI sharply reduced the cost of building the interface layer. It did not reduce the cost of building and sustaining the calculation engine. If you do not separate the two, you will make the wrong decision.
For years, the question was simple to ask, even if difficult to answer: buy a ready-made platform or develop one internally. Both paths required significant investment, time, a team and an architectural decision that was difficult to reverse.
That is no longer true across the entire stack. One specific part of the work became inexpensive, while another remains just as costly as before. The mistake today is treating a planning platform as a single block and applying to the whole a conclusion that is only true for half of it.
The thesis behind this article is the same one behind the program: AI is an interface, interpretation and orchestration layer over deterministic engines. It is not a replacement for calculation. Put calculation inside a language model and you lose reproducibility and auditability—usually discovering it too late.
What became cheaper—and what did not.
This separation is the first filter in any Build or Buy assessment today. Without it, the business case mixes two kinds of work with completely different cost and risk curves.
- Unstructured context capture
- Exception triage and prioritization
- Explaining why the plan looks the way it does
- Drafting communications and scenarios
- KPI and variance narratives
- MRP with correct netting and pegging
- Multi-echelon safety stock
- Finite-capacity scheduling
- Statistical forecasting engines
The right-hand column requires determinism, reproducibility and auditability. The same inputs must produce the same plan, and that plan must remain explainable line by line months later. Those properties still require specialized engineering, validation and support. Generative AI did not eliminate that cost.
The practical consequence is direct. If your operational pain sits in the left-hand column, building is now a real alternative, at a cost and speed that did not exist before. If the pain sits on the right, the economics have not changed: building a planning engine remains a multi-year project with strong dependence on scarce talent.
Most companies have pain in both columns, in different proportions. That is why the honest answer is rarely pure Build or pure Buy.
Four lenses for designing the boundary.
No single criterion can draw the line between Build and Buy. The decision becomes more robust through four independent lenses: rate of change, criticality and differentiation, market maturity and vendor dependency.
01
Rate of change
Separate applications by how quickly they change, not by function. The core of record and calculation changes over decades and holds master data and critical transactions. The rules of your operation change over years and carry company-specific capabilities. The experimentation layer changes over months, exists to test hypotheses and should not carry project-level governance.
In planning, the split is direct. Deterministic engines are the core of record and calculation: buy and protect. AI lives in the experimentation layer: build and replace freely. Governing both with the same rules either freezes one or destabilizes the other.
The three layers are not ordered by importance but by how much sense it makes to build them: from what the market solved better than you would to what no vendor can deliver ready-made.
Layer Examples Recommendation Rationale Core of record and calculationsystem of record · changes in decadesWhere the official data lives and the numbers are calculated. An error here is an error on the shop floor. MRP, finite capacity, master data, transactions Buy and protect Requires determinism and auditability. The market has solved it; customization adds risk without differentiation. Rules of your operationsystem of differentiation · changes in yearsHow your company decides priority, dates and inventory, in a way your competitors do not. Proprietary business rules, inventory policy, prioritization logic Extend what you bought This distinguishes your operation. No vendor delivers it ready-made, and locking it into a third-party product sacrifices flexibility. Experimentation layersystem of innovation · changes in monthsWhat is still under test and can be discarded without leaving a trace. Conversational interface, exception triage, plan narrative Build and discard Changes too quickly to become a project. Build it, version it lightly and accept that it will be replaced. Core of record and calculation
system of record · changes in decades
Where the official data lives and the numbers are calculated. An error here is an error on the shop floor.
Buy and protect
- Examples
- MRP, finite capacity, master data, transactions
- Rationale
- Requires determinism and auditability. The market has solved it; customization adds risk without differentiation.
Rules of your operation
system of differentiation · changes in years
How your company decides priority, dates and inventory, in a way your competitors do not.
Extend what you bought
- Examples
- Proprietary business rules, inventory policy, prioritization logic
- Rationale
- This distinguishes your operation. No vendor delivers it ready-made, and locking it into a third-party product sacrifices flexibility.
Experimentation layer
system of innovation · changes in months
What is still under test and can be discarded without leaving a trace.
Build and discard
- Examples
- Conversational interface, exception triage, plan narrative
- Rationale
- Changes too quickly to become a project. Build it, version it lightly and accept that it will be replaced.
The difference is not only what to build. Each layer requires a different management regime.
The management regime loosens in the same order the decision moves from buy to build.Hover over or select a layer. - Process changeStrict controlExperimentation
- ArchitectureTraditionalAlternative platform
- InvestmentCorporate budgetDepartmental
- Project typeWaterfallAgile
- GovernanceCorporateSpecialist team
From left to right, the regime loosens across all five dimensions. Dark marks strict management, gray the transition and light the flexible regime.
02
Criticality and differentiation
Cross criticality with differentiation. Critical capabilities must work; differentiating capabilities deserve your best people. Planning engines often land in the awkward combination: too critical to fail, too generic to justify proprietary development.
Business rules and decision experience sit at the other extreme: they carry knowledge specific to your operation. That is where building starts to make sense.
What differentiates and what is merely necessary.Select a quadrant.Focus your best people on differentiation. Buy, automate or minimize the rest.
Horizontal: differentiation. Vertical: criticality. Position determines strategy.
03
Market consolidation stage
Position each component from emerging to commodity and derive sourcing from that position: build what is new, buy what has become a product and consume commodities as services.
Solvers, optimizers and MRP are mature markets. Conversational interfaces and agent orchestration are still forming. Building became more attractive precisely where the market has not consolidated a product.
The more consolidated a component market is, the less sense it makes to build it.Select a stage. BuyMature productBuyA market consolidated over decades. Building means paying more to remain behind it.
04
Where the vendor boundary sits
Buying is not binary. It means deciding where the vendor ends and the capability that remains yours begins.
With an expansive boundary, integration, rules, engine, interface, analytics and AI sit inside the platform. With a narrow boundary, the vendor supplies engine and interface while data, rules, analytics and AI remain under your control.
The difference appears when something changes. In an expansive boundary, replacing the vendor changes much of the operation. In a narrow one, the purchased component is replaceable.
Where the vendor boundary sitsHover over a component to compare.Buying is not binary. The decision is how much of your operation sits inside the vendor.
Expansive boundary
VendorCapability lock-in
The vendor supplies more than technology; it carries a material part of your operation.
Changing vendors means changing part of the operation.
Narrow boundary
VendorIn-houseIn-houseReplaceable component
The vendor supplies engine and interface. Data, rules, analytics and AI remain under your control.
Changing vendors means replacing one component.
The four lenses are not our invention; their sources appear below. What follows is ours: translating them into supply chain planning, defining the criteria, weights and scoring method.
NEO Build or Buy Framework
Six criteria that turn an architecture debate into an objective decision.
NEO developed the criteria, weights and scoring method from real planning projects. The model is proprietary, but it is not a black box: the criteria are explicit, the weights can be debated and every answer requires evidence from the operation itself.
A framework that does not expose its own criteria cannot be audited.
Filter by criterion
Select a criterion to read it on its own.
- 01
Cost
Five-year total cost
What to investigate: every cash outflow over five years, not only the license or first-year team cost. Include licensing, development, infrastructure, integration, version upgrades, data rework, internal IT hours, training and support.
Support is the most common blind spot. Do not budget only for the build: accumulated support in later years often exceeds the initial investment, yet rarely appears in the financial model behind the decision.
The test is simple: ask for the isolated year-three cost in both scenarios. If nobody can answer without rebuilding the calculation, the business case is not ready to decide anything.
- 02
Flexibility
Flexibility to change business rules without a project
What to investigate: how long it actually takes to change an allocation rule, prioritization criterion or inventory policy. Use the real elapsed time from the last change, including prioritization queue, validation and deployment window.
Supply chain rules change faster than the IT project cycle can support. A rigid platform turns every adjustment into a formal request, leaving operations with the wrong rule for months. A poorly architected internal system creates the same problem with an internal queue.
A common mistake is confusing screen configurability with rule flexibility. Renaming a field is not the same as changing how the plan is calculated.
- 03
Risk
Execution and key-person risk
What to investigate: how many people understand enough to run and evolve the solution alone. If the answer is one or two, the dependency is already material.
Large IT projects carry schedule, budget and scope risk, whether implementing a platform or building internally.
The best predictor is not the plan’s optimism but the company’s own project history. What happened in the last comparable implementation? What shipped, what slipped and what was abandoned?
- 04
Time
Time to first production result
What to investigate: how soon the operation needs to see the first result in production. Not the signing date or the end of the whole project: the first delivery someone in the operation actually uses.
Initiatives that remain invisible for too long lose sponsorship before completion, and the decision to continue stops being technical. A short deadline pulls toward buying a base that delivers early; a long one tolerates building.
A common mistake is measuring the time to go-live of the whole project. What sustains the initiative is the first partial result, and it has to fit the attention span of whoever sponsors it.
- 05
Maturity
Data and planning-process maturity
What to investigate: master-data reliability—lead times, BOMs, production calendars, resource capacity and inventory policies. Do not ask whether the field exists; ask whether its value reflects operational reality.
Also determine whether the planning process is documented and stable, or depends on one person’s spreadsheet and unwritten adjustments.
Poor master data is not fixed by a new system. The new system inherits it and produces wrong answers faster and with greater confidence.
- 06
Ownership
Who sustains the solution after go-live
What to investigate: the name of the accountable person and function. Not a generic structure, “IT” or “vendor support.” Who specifically owns this in year two?
The lists differ by path. Buying requires product ownership, configuration, releases, integrations and planning knowledge. Building adds code, architecture, infrastructure and evolution.
Teams often assess the ability to go live, not the ability to sustain. If this question has no named answer before signing, it will have a poor answer afterward.
Where your case lands between Build and Buy
The six criteria create a position on the axis. The result is more than a recommendation: it shows what pulled the decision each way, the principal risks and what needs to happen next.
First
Are the foundations strong enough to proceed?
Each failed minimum condition removes a path. The result is NO-GO only when Build and Buy are both removed.
If Build and Buy are both removed
NO-GONeither Build nor Buy. Not yet.
The assessment stops here and identifies what must be in place before the decision makes sense. Establish the conditions before investing.
If at least one path remains
The case is positioned on the axis.
The six criteria are scored using operational evidence, placing the result in one of four positions between building and buying, never in a zone that has been removed.
What each condition removes
- Master data has no ownerBuild and Buy (NO-GO)
- There is no planning processBuild and Buy (NO-GO)
- Return covers neither path within five yearsBuild and Buy (NO-GO)
- Return covers only one pathThe path that does not pay back
- The process runs on spreadsheets and peopleFull Build
- No named owner for supportBuild
- Nobody masters the calculationBuilding the engine
A framework that cannot produce this outcome is a sales funnel, not a framework.
Select a position.
Find where your case lands between Build and Buy
Facilitated by NEO
Answer the six criteria with evidence from your operation. The assessment shows where your case lands, what pulled the decision each way and whether the foundations are strong enough to decide now.
The complete assessment, including weights and a criterion-by-criterion reading, is facilitated by NEO. Contact us directly to evaluate your case.
Falar com a gente
If the result is NO-GO
Where to begin.
When the basis for a decision is missing, the gap is usually readiness or a quantified return. We offer both assessments.
- ReadinessSupply Chain Diagnostic
Assesses readiness and maps opportunities: planning maturity, data quality and governance, sources of loss and prerequisites before investment.
Learn more- ReturnROI Study
Quantifies expected returns against five-year total cost, so the investment hypothesis can be supported—or rejected—with numbers.
Learn more
Disclosure
NEO works on both sides of this decision: implementing platforms and enabling internal teams to build. That is why the criteria and weights are published openly. The method can recommend Build, Buy, an intermediate position or conclude that neither path has adequate foundations yet.
Sources and further reading behind this section
Listed in the order in which the ideas appear. Public sources include where they can be found.
- 01
Gartner. Pace-Layered Application Strategy, 2010.
gartner.com
Used in: Rate of change
- 02
Moore, Geoffrey A. Dealing with Darwin: How Great Companies Innovate at Every Phase of Their Evolution, 2005.
Used in: Criticality and differentiation
- 03
Wardley, Simon. Wardley Mapping, 2005.
docs.onlinewardleymaps.com
Used in: Market consolidation stage
- 04
Haberlah, David. Build vs Buy in 2026: Using Wardley Mapping to Navigate the Agentic AI Shift. Medium.
Used in: Market consolidation stage
- 05
Cochran, Tim; Gandhi, Prashant; Nygard, Carl. Build versus buy. ThoughtWorks, 2022.
thoughtworks.com
Used in: Vendor boundary
The diagrams are NEO’s interpretation of these sources as applied to supply chain planning. Any error of interpretation is ours, not the source’s.