Buy the Fundamentals, Build the Rest: Rethinking Customization With AI

Read Time: 4 minutes
Authored by: Bernie Cabada
Operations & Growth
All Segments

Summary

Customization used to mean a development cycle, so buyers penalized platforms for functional gaps. Agentic AI now lets business users build most of what they need above the platform. That shifts the evaluation toward the fundamentals a firm cannot retrofit, including bitemporality, lineage, a data catalog, and permissioning.

Reviewing platforms for functional gaps against requirements has always played a major part in vendor review and procurement discussions. Buyers would either penalize a platform for its functional gaps or concede the point by planning for vendor or internal customizations. Product vendors often came up short because closing a gap meant a development cycle.

To make a platform fit your operation, you spent your own people’s time on the specification and paid exorbitant fees to the servicer’s development teams, or you built it in-house. Bespoke work meant testing and sign-off before deploying into a production environment. It also tied companies' hands in deciding whether to deploy platform upgrades and enhancements that might conflict with such customizations.

Standard vendor-solution software comes with a standard set of processes along with automated workflows that are already encoded within the platform... Customization from the business side, however, can compromise the platforms’ value-creation and operational capabilities by making them more complex and costly. — McKinseyi

With AI, the ability for buyers or individual business users to layer agents on top of a core platform has changed the equation. Your own team now closes many of the gaps without multi-month implementation efforts. This potential should also change the calculus and scoring methodology of product evaluation.

How the AI layer refactors customization

Layering took place long before agentic AI. Instead of paying for changes within the product, it was common to have your own people build on top of it, incorporate point solutions, or hire consultants for custom integrations. People who mastered Excel learned to develop advanced macros that could pull data from multiple systems and work with them locally. In other cases, superusers or tech resources were able to layer on SQL, JavaScript, or XML to make a product seem customized without touching the core code.

This workaround often provided the answer for dealing with legacy systems. In shops that have been in business for decades, these may be bespoke systems built before anyone on the team was hired. These firms sometimes concede that code written a decade or two ago is good enough, even though a modern software-as-a-service platform could deliver the same capabilities. The workaround still makes more sense than a full-on migration.

Agentic AI has a similar impact, but it's more accessible to an ordinary business user. A business user who describes what they want in plain language gets a working feature back. For example:

  • An operations analyst can write their own data quality rules, stand up pipelines, debug broken ones, and set up a new API request without an in-depth understanding of the underlying technology.
  • An accountant fielding an auditor's question asks an agent for the exact inputs behind the accrual, the interest calculation, or the fee, and has them while the auditor is still on the call. In the old model, that question opened a multi-day investigation between the firm and its servicer.
  • A separately managed account or a limited partner needs their own specialized report format. Your own analyst builds it in an hour or two rather than filing it with a development team, and changes its layouts and datasets over the life of the relationship.

In each of these scenarios, turnarounds that once took months or years now take days or weeks. The level of effort for implementation on the vendor or internal tech team side changes. Ownership shifts as well, with technologists focusing on the architecture and the underlying controls.

The customization stack

We can also zoom out to a much larger scenario. Imagine a large pension fund running legacy SQL Server databases alongside in-house tools. If they want to use a new feed of pricing or risk data, ETL cycles can take days while the people who need it wait. The business team might start pushing for changes to the data model, whether it can bring its own, and requests to customize tables. These business requirements can also show up in RFPs and vendor reviews.

Buyers often think of all these kinds of things as a choice between native support and customization. The thinking goes, “Either the product does it, or someone has to build it.” But there's a more nuanced and ultimately more constructive way to think about it.

In an investment management environment, the core platform defines how portfolios, public and private assets, and lifecycle events relate to one another and should come from the platform provider. On top of that, you configure things like a user-defined field on a trade, a security, or reference data capturing a strategy.

The third category covers what used to require customization. Agentic capabilities enable your team to build genuine one-offs for various desks and lines of business. For example, a front-office desk can monitor risk against its own model based on agents your team supplies. The platform holds the foundational data and runs the calculations. The output might return as a dynamic dashboard or a PDF, with Model Context Protocol (MCP) providing a mechanism for your agent to retrieve the data.

That build sits in your tenancy alone. When a platform deploys as single-tenant, a custom feature affects only the client that requested it, so it never enters the all-client testing and sign-off cycle. The upgrade question that used to tie your hands no longer applies because nothing you built is riding on a release everyone else shares.

The responsibility line

Even with empowered end users, your IT team remains the end architect, deciding how you connect to internal systems, how you model data across old and new systems, where the permissioning layers sit, and what controls run where. The difference lies in who takes on responsibility for wrangling data to produce a deliverable.

Above the architecture, your business users own their datasets, data products, and reports. They have more options to iterate on those without opening a ticket. You might have an analyst looking at sector-level IRR who can tell the agent to add look-through capability, get it back, and keep working on the results. They are also in the best position to validate what they built, because they sit closest to the numbers. The technical side ensures the data arrives clean and the pipelines run.

When an ingested file changes and a pipeline starts failing, exception monitors catch it, so your business user watches the failures stack up. A power user could even run a diagnostic with another agent, comparing the new file against the original mapping and redeploying the pipeline.

This changes the roles of the product provider and the IT team. Some things still need fundamentals built into the platform rather than on top of it. Those situations highlight the value of a single-tenant model. Your product vendor can build that feature into your instance alone, so it never enters a testing and sign-off cycle across every client.

Bringing procurement back to the fundamentals

Cost of ownership is a fundamental question for any procurement analysis. Customization carries additional costs because it paints you into a corner, leaving you inside a world of legacy technology that you may no longer want. It helps explain why 92% of asset servicers report their ability to innovate is at least partially constrained by legacy systems, according to Deloitte.ii

Easy building carries two costs of its own. First, there is a risk that users build agents for a need that an existing agent already covers, so you pay for the same capability twice. Second, if a user builds a report using datasets assembled by an agent but can't say where the data came from, nobody can stand behind the numbers.

You can mitigate these procurement risks via two mechanisms. A catalog lets a user search what already exists before building anything new. And with lineage running from raw data through cleaned and normalized data to the finished data product, users can walk any number back to its source table. These become new things to inquire about during a vendor review.

Fundamentals matter more than the pre-existing availability of every conceivable feature and report. These fundamentals should be central to the checklist you use for vendor selection. Each of them is nonnegotiable when considering tools and platforms that support your investment operations.

  • Bitemporality: Do records hold both when something happened and when the system learned of it?
  • Data catalog: Can a user find what already exists, and does it allow user-defined fields and direct data querying?
  • Integration: Can users connect via API, data share, or SFTP with external systems?
  • Lineage: Can you trace the path from source to golden master through each step in between?
  • Governance: Does access control at the dataset, column, and record level clear your firm’s bar?

Much of the time, clients who seek us out when considering a new platform provider previously focused too much on feature-level requirements and not enough on fundamentals. When that happens, the need for the feature decays over time as investor needs and strategies change. New needs or new asset classes then reveal that the fundamentals were shaky. They bring unexpected procurement risks back to the table.

Buy the data foundation, build the interface yourself — why asset managers are going headless.

Bernie Cabada's profile photo
Authored By

Bernie Cabada

As Vice President of Presales Consultants at Arcesium, Bernardo leads in showcasing the company's cutting-edge capabilities through executive-level engagements, delivering compelling presentations, technical demonstrations, and proof-of-concepts. His proficiency extends across crucial areas of middle and back-office investment operations, with a particular emphasis on data governance and enterprise data management for investment managers and asset owners.

View Author Profile

Share This Post

Subscribe Today

No spam. Just the latest releases and tips, interesting articles, and exclusive interviews in your inbox every week.