Consulting

01 · Narrative

AI Planning

Creating a future-proof roadmap

AI is fast moving and disruptive. The idea of horizontal and vertical products gets skewed as AI agents possess deeper domain expertise and move from deliberative roles to taking actions. This is a fundamental challenge for any business. From now on, deciding what is worth using AI for is dynamic, not a one-time discussion. Every organization must become a data-first organization. Work must become legible to AI, automated by agents, and turned into an accretive flywheel that separates one firm from another.

I work with firms to identify their core differentiators and build a plan to capitalize on them as AI capability grows. The analysis below came out of my work at HighRoad, where we signed eight funds in commercial real estate finance, worked with investors across equity and debt deploying $500M to $10B, and spoke with dozens more. It is the market read I brought to firms designing a multi-year AI plan, and it led to firm-specific rollouts.

State of CRE AI

Across every fund, we saw similar challenges, cultural and technical, as they tried to adopt AI, deploy it for measurable ROI, and capture advantage in a competitive market. A fund has never had to run its data as an operation, and none we worked with had yet built the foundation it requires. These are the core themes any multi-year AI plan has to address.

The core challenges:

  1. No firm-wide standard or enforcement mechanism for where data lives and how it is represented.
    • Most funds run on a hub-and-spoke model, consolidating summary data as it flows into a central pipeline table, but otherwise giving autonomy to origination teams to run their books as they see fit. Inputs may exist in hosted servers, local files, or on SharePoint. Only a select number of output metrics are tracked, and forcing conformity for entries is challenging, leading to bloated picklists and unresolved entities.
    • Excel is the central data store, so most firms do not get the automatic enforcement and standardized columns that databases, from simple (MS Access) to complex (Snowflake, Databricks), provide by default.
    • All of this creates immense challenges for AI to navigate the firm’s accumulated knowledge and produce useful outputs on top of it.
  2. Vertical platforms each solve one slice of the workflow and trap the data behind their walled gardens.
    • At best, funds assemble a stack of point solutions: a CRM, a pipeline tool for deals, a servicing portal for the book, each owning one workflow. The data sits in each vendor’s schema, built for a person to read rather than an agent to query, and predates AI entirely.
    • Even fully adopted, these tools offer partial value to management but do little to reduce the administrative burden on the analysts entering the data, so they hold the minimum necessary information.
  3. Horizontal tools leave assembly and customization to the fund.
    • Data warehouses, integration layers, and search (Snowflake, Zapier, Hebbia) are powerful but unopinionated building blocks, and funds must pay an upfront fixed cost in implementation before any of it returns value.
    • That wiring is real data and software engineering, a competency funds do not have and should not want to build.
    • Success hinges on employees adopting a new way of working, a cultural change that typically meets stiff resistance.

Additional challenges:

  1. Funds see their custom memos, pipelines, and IC processes as core differentiators, and uniform products struggle to support them.
    • Off-the-shelf products, my own former product included, impose a single schema and workflow, but each fund has its own way of doing this and treats it as a source of edge it will not standardize away.
    • A tool that cannot bend to the firm’s own fields, definitions, and logic gets worked around or abandoned, so customization is the factor that decides adoption.
  2. The fund exists to do good deals at volume, so AI must add leverage to that mandate.
    • AI can obtain deal information faster, but the fund itself must understand and clearly define where it has levers for volume, cost of capital, pricing model, network, and the rest. Any good plan measures implementation effort against those core drivers of deals.
  3. Funds want to own their data and build a compounding flywheel, but the prerequisites are not yet in place.
    • A flywheel only compounds on a foundation that already exists, consolidated data, consistent capture, and someone accountable for both, and laying that groundwork requires process and technical change.
  4. Adoption requires an internal champion, but without technical planning, solutions become brittle and impossible to maintain.
    • Getting a firm to actually use AI takes someone inside who owns it, drives adoption, designs the new workflows, and holds the metrics.
    • As scripts, skills, and connections proliferate, adoption and maintenance become an ongoing challenge.

Future Trajectory

Every fund we worked with faced some version of these challenges. They reflect the state of the industry, not a failing of any one firm, and none has yet turned them into an advantage. Firms that treat this as a decision to make now will ensure their opportunity set grows with the technology. There are three clear ways to think about the future.

  1. AI capability is compounding while the cost of a given task collapses. The price of a fixed level of AI performance has been halving roughly every two months, even as frontier capability expands quickly. Buying vertical tools now risks handing over control of these trends. Stay model-agnostic and ride the cost and capability curve instead of locking to whatever tool is best this quarter.
  2. CRE trails other verticals in AI adoption, and the gap widens. CRE has been slower to adopt AI than even other corners of finance, and adoption within it is uneven, so the field is early with no durable advantage yet built. As capability compounds, differences in AI implementation grow more impactful. Build the data and workflow foundations now, and determine the core competencies AI should master while creating an architecture that can broaden scope as the cost curves change.
  3. CRE vendors are converging, creating an unstable market. AI lets incumbents and startups build the same feature sets. Product commoditization drives consolidation, with clear winners and losers, and betting a firm’s core on a single vendor can mean starting from scratch in a few years. The durable layer is the firm’s own data and the standard it enforces. Own that outright, at the very least as redundancy to adopted tools.

What a firm is left with

After years in this space building a cross-fund perspective, I know what holds up and what does not. What I bring a firm is that experience, applied to its tools and its way of working, to design a system its own team owns and runs. The work is to set near-term AI priorities with clear ROI, design and implement a maintainable system that runs on the existing technology stack, and craft a plan that grows with the firm and with frontier capability. What the firm is left with is durable, a system and a standard that keep improving as models and costs change, run by its own people and tied to no single vendor.