Company of the Week:

Most CRE investment teams already have a huge amount of potentially useful information sitting inside the business.

These might include past underwritings, OMs, rent rolls, comps, market reports, and previous assumptions, along with deals that were pursued, rejected, or simply buried in old folders, never to see the light of day again. 

The challenge is making that accumulated information both available and useful when the next deal comes in. That’s exactly the problem Starboard is trying to solve.

We recently sat down with co-founders Rehan Ackrim and Jasper Parish for a live demo of the platform. Rehan previously worked in acquisitions and development at Hines, while Jasper comes from a computer science background at Cambridge.

Their basic objective is straightforward: help investment teams reach conviction on a new deal faster. And the key to that is Starboard’s context engine.

Starboard’s core product is what the team calls its context engine.

Connect the platform to a firm’s internal files in places like SharePoint, Box, or Dropbox, and Starboard begins organizing information from previous deals alongside whatever new opportunity is being analyzed.

When a fresh deal room arrives, the platform can ingest the OM, T12, leases, rent roll, marketing materials, and other documents, extract the underlying information, and connect it with the rest of the firm’s history.

That means the next deal isn’t being considered in isolation.

Previous transactions can become comps, earlier underwriting assumptions can provide a reference point, and existing market knowledge can be brought back into the analysis. Information buried inside an old OM can potentially become relevant again, even years later.

Jasper described the challenge as finding a middle ground between two existing types of software.

On one side you’ve got traditional systems, which tend to be highly structured. That makes them reliable, but relatively rigid. On the other you’ve got general-purpose tools like Claude and ChatGPT. These are extremely flexible, but they don’t automatically carry around a properly organized understanding of an investment firm’s history.

Starboard is trying to find the middle ground here, combining flexibility with enough structure to make the output genuinely useful for serious investment work.

Our live demo started with a new deal being added to Starboard.

That can happen by uploading the deal room directly, or forwarding the materials from a broker. Starboard then extracts the information and begins building out the analysis.

Inside each deal, the work is organized into customizable ‘chapters.’ A chapter might cover sales comps, supply, demographics, leases, the surrounding area, or another part of the firm’s underwriting process. A team can also create its own chapters around work it regularly performs, and have them populate whenever a new deal comes in.

There’s a screening layer too. Starboard calls this thesis scoring, where a firm can define the criteria it cares about, and have incoming deals evaluated against them.

Some criteria are simple: building age, number of stories, location. Others can be more nuanced, like whether the market has achieved sufficient rent growth over a particular period.

For an acquisitions team looking through a high volume of opportunities, the overall idea is to move more quickly from ‘another deal just landed in the inbox’ to ‘this one actually deserves our attention.’

The context engine becomes easier to understand with a practical example. In this case, let’s take comps.

Starboard can use comps contained in the broker’s OM and external information from MSCI Real Capital Analytics. But it can also look across deals the firm itself has previously reviewed. A property from six months ago may now be a useful comp for the opportunity lying before you today.

The same principle carries through to other parts of the analysis.

During the demo, Starboard surfaced a connection between the property manager on the new deal and another asset the firm had previously considered. Lease information could be extracted into the relevant section, while individual numbers remained linked back to their source.

Over time, the firm’s own history becomes a much more active part of how new opportunities are evaluated. That feels especially relevant in CRE, where plenty of institutional knowledge - even now - still lives in folders, spreadsheets, inboxes, or simply the memories of individual employees.

This is one of the most practical parts of the product.

Starboard isn’t asking investment teams to suddenly abandon their existing underwriting models. Its Excel plugin is designed instead to work with the templates a firm already uses. The platform can learn from previous underwritings, pull information from the context engine, and populate the model inside Excel.

In the demo, the instruction was essentially ‘underwrite this deal’. Starboard then used the underlying deal information to begin filling the existing model. 

Importantly, it can also bring competing assumptions to the analyst rather than just choosing one itself. 

Rent growth, for example, might come from the broker’s materials, a recent market report, or the assumption the firm itself used on another deal in the same market. That leaves the analyst with a much more focused decision to make: which assumption is appropriate here?

The same context can carry into PowerPoint and Word. An underwriting produced in Excel can feed the numbers used in the investment committee presentation, rather than requiring someone to manually recreate the same work across several files.

There is an obvious problem with bringing generative AI anywhere near institutional underwriting: nobody wants an unexplained number appearing in the model. Starboard has therefore put a lot of emphasis on auditability.

Every data point displayed in the platform can be traced back to its source, whether that’s a cell in a rent roll, a line in an OM, another internal document, or an external dataset.

The system is also deliberately cautious about assumptions.

Jasper gave the example of a firm repeatedly using a 5% cap rate. Starboard may recognize that pattern, but it still isn’t supposed to blindly insert 5% into every future deal simply just because that’s what happened before.

If the available evidence conflicts, the system can surface the different possibilities and ask the user to decide.

This is where the conversation became particularly interesting.

Today, much of the portfolio analysis is backward-looking. Starboard can tell you what changed, compare current assumptions with previous ones, combine that information with external datasets, and help run different scenarios.

Rehan was explicit that they aren’t yet at the point of reliably predicting what happens next. But that is where they want to go.

The long-term idea is to combine a firm’s historical data with much larger external datasets, then use AI to identify patterns and make increasingly accurate forecasts. One example raised in the conversation was an industrial investor wanting to know where the top of a warehouse market might be 18 months from now.

Everyone would like the answer to that question. Producing one reliably is - of course - much harder.

Regardless, it still points toward why Starboard is investing so heavily in the context underneath the analysis. If predictive systems do become genuinely useful in CRE, a firm with years of well-structured proprietary deal data may be in a much stronger position than one starting from the same public information as everybody else.

We’ve seen plenty of tools that can extract an OM or help fill an underwriting model. Starboard’s broader pitch is that those individual tasks become more valuable when the AI understands the investment process surrounding them.

Which deals has the firm looked at before? Which assumptions did it use? What did the team reject? What counts as a relevant comp internally? Which source supports this number?

The platform is trying to put that context behind every new analysis, while still leaving the investment team in control of the decisions that follow.

For acquisitions teams, that could mean less time gathering and re-entering information, and more time examining whether the underwriting actually makes sense.

And over time, the real value may come from something harder to replicate: an AI system that becomes increasingly familiar with the way your specific firm invests.