Book a demo
Topics · GOVERNANCE-AGENTS

Asset Management AI approval gate Build vs Buy | Moring

For asset managers, the build-versus-buy question for approval gates matters less than the governance foundation underneath. Buying a platform accelerates runtime enforcement and evidence collection, while building offers control over proprietary workflows; either way the gate must encode agent authority, decision rights, and auditability that regulators and investors expect.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

For asset managers, the build-versus-buy question for approval gates matters less than the governance foundation underneath. Buying a platform accelerates runtime enforcement and evidence collection, while building offers control over proprietary workflows; either way the gate must encode agent authority, decision rights, and auditability that regulators and investors expect.

How it works

Asset managers evaluating approval gates usually start with a procurement question: should we build the control in-house or buy a governance platform? Elevate Consult argues the question misses the point, because AI initiatives stall for lack of foundational governance rather than model choice, with business challenges, regulatory pressures, and talent gaps cited as leading obstacles across large shares of enterprises. The gate is one enforcement expression of that foundation.

Buying accelerates the runtime layer. Commercial platforms arrive with policy engines, identity integration, logging, and evidence collection already built, which matters because agents call tools, use identities, and retrieve data at runtime in ways that are hard to intercept with homegrown scripts. For a mid-sized manager without a platform engineering team, this is frequently the difference between a control that operates and a control that is documented but inert.

Building preserves fit with proprietary workflows. A manager whose investment process depends on internally built research pipelines, order systems, and data platforms may find that a purchased gate integrates poorly with those systems, and the integration cost can exceed the license. Building also keeps control of sensitive workflow logic in-house, a consideration when the gate itself would observe portfolio-relevant activity.

A pragmatic hybrid is common: buy the gate and evidence layer, build the connectors and policy content. Databricks' governance guidance describes centralized standards with federated execution, and Liferay emphasizes embedding governance controls directly into existing platforms so execution is seamless. The policy definitions, risk tiers, and approval routing reflect the firm's own operating model regardless of who supplies the enforcement machinery.

Whichever route is chosen, the requirements do not change. Robeco's research frames the governance questions as what systems can do, what they can touch, and who remains accountable; Arcesium adds that agent authority, data access control, and behavior monitoring must be defined explicitly. A gate that cannot answer those questions is inadequate at any price, and one that can is valuable whether built or bought.

Before committing, run a scoped pilot against one consequential workflow, such as agent-initiated data purchases or report generation, and compare the evidence each option produces. Regulatory uncertainty is real, per Baker Tilly, so prefer the option that adapts fastest when expectations shift, because re-platforming a control is harder than tuning one.

Frequently asked questions

When does building an approval gate in-house make sense for an asset manager?

Building makes sense when the firm's investment workflows are highly proprietary and heavily integrated, when platform engineering capacity exists, and when the governance policies are already stable enough to encode. A manager with internally built research and order systems may find that purchased gates integrate poorly and that the integration effort rivals a build. The trade-off is maintenance: the firm owns the policy engine, logging, identity handling, and evidence storage indefinitely, so the build case weakens wherever those capabilities would duplicate commodity infrastructure.

When is buying the better route?

Buying is usually stronger when the firm lacks platform engineering capacity, needs runtime enforcement quickly, or wants mature evidence collection from day one. Commercial platforms ship with policy engines, identity integration, and audit logging that would take a small internal team quarters to reproduce, and agents act at runtime in ways that are difficult to intercept with custom scripts. Baker Tilly notes that regulatory uncertainty already slows AI adoption in asset management, so a vendor that adapts quickly to new supervisory expectations reduces a risk the firm would otherwise carry alone.

What requirements apply regardless of build or buy?

The gate must enumerate what agents may do, declare what systems and data they can touch, and record who authorized each consequential action, matching Robeco's accountability questions and Arcesium's emphasis on defining agent authority. It must log decisions with enough context to satisfy model risk documentation and investor diligence, support exception handling through a documented path, and allow permissions to widen as evidence accumulates. A gate missing any of these is inadequate at any price, and these requirements should be written into procurement criteria before vendors are evaluated.

How should we evaluate vendors against our own workflows?

Run a scoped pilot on one consequential workflow, such as agent-initiated data purchases or client report generation, and compare the evidence produced. Check integration effort against your existing platforms, since Liferay notes that controls embedded into existing digital platforms execute seamlessly rather than creating friction. Ask how policy changes propagate, how overrides are handled, and whether the vendor can adapt when supervisory expectations shift. Databricks describes centralized standards with federated execution as the scalable pattern, so favor vendors that support firm-specific policy content rather than one-size-fits-all rule sets.

Sources

  1. robeco.com
  2. liferay.com
  3. adaptivesecurity.com
  4. orangelogic.com
  5. databricks.com
  6. elevateconsult.com
  7. - federalreserve.gov
  8. - federalreserve.gov
  9. - artificialintelligenceact.eu