
Vertical software decisions get framed as a binary. Buy a platform built for your industry, or build your own. Both options are usually wrong, and the third option — assembly — rarely gets scoped because nobody sells it.
I want to make the case for it using property development as the worked example, though the reasoning transfers to any industry where a small number of universal systems sit beneath genuinely idiosyncratic coordination logic.
Why buying disappoints
A vertical platform encodes an assumed process. If your process matches, it is excellent value — you inherit years of accumulated domain handling for a licence fee.
The trouble starts when your process differs in ways that matter. In property development, that is the normal case rather than the exception: a portfolio mixing asset classes, capital structures and jurisdictions has combinatorial complexity that a platform necessarily flattens, because flattening is how it serves many customers at once.
You then configure. Configuration works until it does not, and the failure signal is consistent and measurable. Count the spreadsheets maintained alongside the platform. Three or more, with at least one regarded as more authoritative than the system, means the platform cannot express how the business actually works. More configuration will not fix that — the constraint is structural.
Why building everything is also wrong
The reflexive response to a bad platform fit is to build the whole thing. This substitutes one error for another.
A full build means rebuilding capabilities the market commoditised long ago. Document storage with versioning and access control. Accounting with tax handling and audit trails. Field tooling with offline sync. Each of these is a serious product in its own right, none of them differentiate you, and all of them will absorb engineering attention for years after launch.
Worse, they carry ongoing regulatory maintenance. A ledger has to keep up with tax rule changes forever. That is a permanent liability accepted in exchange for nothing.
The assembly position
Keep the commodity layers bought. Build only the layer that encodes how your business is different.
In this domain that means buying accounting, document storage, the leasing or sales CRM and field tooling, then building the pipeline logic, stage-gate model and reporting layer that ties them together.
The ratio is usually favourable: a relatively small custom surface sitting on top of substantial bought capability. You own the part that differentiates and rent the parts that do not.
Three conditions have to hold for this to work.
The commodity systems must be integrable. This is the hard gate. Can you extract the full general ledger on a schedule without a professional services engagement? Does the document store expose version history through an API, or only the current file? Can the field tool emit events as they occur rather than batching overnight? Each negative answer converts into permanent manual reconciliation and erodes the case for assembly.
You must own the canonical model. The custom layer holds the entities that matter — asset, stage gate, commitment — and the bought systems are subordinate participants. If the accounting system's chart of accounts becomes your de facto data model, you have inverted the relationship and the flexibility disappears.
Someone must own the integration surface long term. Assembly trades licence cost for integration maintenance. That is usually a good trade, but only if it is resourced. Unowned integrations decay quietly.
What AI changes about this
Two things, and they both favour assembly.
The custom layer got cheaper to build. Pipeline logic, stage-gate rules and reporting are high-volume, well-specified, testable code driven by domain knowledge the business already holds — close to a best case for AI-assisted development. The part you build in an assembly strategy is precisely the part whose cost fell most.
And integration got cheaper to maintain. A meaningful share of integration work is mapping between representations and handling shape variations, which is now considerably less laborious than it was. That directly reduces the main ongoing cost of the assembly approach.
The caveat: AI-assisted development did not make it cheaper to own a ledger. The commodity layers stayed commodity for the same reason as always — their cost is regulatory, not developmental.
Deciding
-
Portfolio homogeneous, process standard, no appetite to own software → buy
-
Portfolio mixed, coordination logic is your advantage, commodity systems integrate cleanly → assemble
-
Commodity systems cannot integrate at all, or the domain has no usable products → build, reluctantly
-
Three or more spreadsheets alongside the current platform → stop configuring; the fit is structural
Frequently Asked Questions
What does assembly mean in practice?
Buying the commodity layers — accounting, document storage, CRM, field tooling — and building only the coordination logic, domain model and reporting layer that encodes how your business differs, then integrating the rest.
When does buying a vertical platform work well?
When your process genuinely matches the one the platform encodes. You then inherit years of domain handling for a licence fee, which is excellent value.
Why is building everything usually a mistake?
Because it means rebuilding commoditised capabilities like document versioning, accounting and offline field sync — each a serious product, none differentiating, and several carrying permanent regulatory maintenance.
What has to be true for assembly to work?
The commodity systems must be genuinely integrable, your custom layer must own the canonical data model rather than inheriting one, and someone must own integration maintenance long term.
How does AI affect this decision?
It favours assembly on both sides: the custom coordination layer is exactly the kind of code whose cost fell most, and integration mapping work became considerably less laborious. It did not change the cost of owning a ledger.
What is the clearest sign the current platform is wrong?
Three or more spreadsheets maintained alongside it, with at least one treated as more authoritative than the system. That indicates a structural fit problem that configuration cannot resolve.
Full guide with the data model, field capture and cost analysis: Real Estate Project Management Software: The 2026 Build Guide
TechCirkle: custom software development | SaaS development