Beyond Copilot: sourcing to underwriting (EN)

25:37 · English · Acquisitions

Watch this video

Enter your details to access the library. We'll email you a code to verify your address.

See how we use your data in our Privacy Policy.

About this video

One deal, end to end: sourcing agent, 50-document screen, lease v T12 reconciliation, IC memo, audit trail, IRR model and Micro Modeller scenarios. From our APAC webinar.

Full description

The demo from Fifth Dimension's October 2026 APAC webinar "Beyond Copilot: the next step in real estate AI", re-cut as one flowing walkthrough in ten chapters. Robin Langdon follows a single hospitality opportunity from a morning sourcing dashboard into a deal workspace linked to SharePoint, where Ellie screens all fifty documents with every figure cited to source, extracts nine leases into structured records, and finds that the T12 adds up internally but does not tie to the leases: a 3% compounding step missed on eight suites. The analyst makes the call on the record, NOI is restated from $263,358 to $288,400, and the IC memo is drafted from the reconciled figure. Months later a regulator audit skill replays every human direction, AI action and calculation. From the restated numbers Ellie builds a five-year IRR model as a live Excel workbook, which Micro Modeller turns into a deterministic program of 263 bricks and runs five scenarios in seconds. The closing chapter shows the industry context behind it: 90+ mapped real-asset processes with Singapore, Australian and other market nuances set per workspace.

Key points

  • A sourcing agent scans hundreds of sources overnight and scores each opportunity against your gating criteria
  • Fifty documents screened in the workspace, every figure cited to its source passage
  • The T12 adds up internally but does not tie to the leases: NOI restated $263,358 to $288,400 with the working shown
  • A regulator audit skill replays every human decision, AI action and calculation months later
  • Micro Modeller turns the IRR workbook into a deterministic model of 263 bricks and runs five scenarios in seconds
  • 90+ real-asset processes mapped, with market nuances (Singapore, Australia, Dubai) set per workspace
Transcript

So hopefully now seeing the 5D platform. So in here we can see I'm now currently on my 5D, which is really a homepage for how everyone can work in the platform. In 5D we have the concept of workspaces. In here we can see I've got different workspaces, which are really collaboration spaces where you can work in a secure container linked to your data. Typically that might be a SharePoint folder, and being able to then provide insights on all the documents in that particular data.

So if we start with the life cycle and look at deal sourcing, how might that work in a workspace? We want to be able to go out and fetch opportunities, research the web, research various paid sources. You can see in my settings here I have a number of different connectors in the platform where I can link in different systems. The platform will link to your existing Microsoft structure, SharePoint, Outlook, plus lots of other partner systems. In my workspace here I've created a dashboard where I want to look at, in this case, hospitality deals. So I've set up an agent in the system, and that agent is based on a set of criteria, a set of sources: it's going to look at hundreds of different websites, different sources, social media. Every day I get a morning dashboard when I come in, which I can run and see here. In this case I'm looking at hospitality across the APAC region. I can see what sources I've scanned, what's new. I can see that overnight, this was yesterday's dashboard, there was a hotel at Sydney Airport that came onto the dashboard. I can look at that and start building up a dossier of information around that particular asset. The reason it pops up on my dashboard is that I've got some gating criteria around asset type, scale, geography and so on. All this is completely configurable in the system, and you start to score the various opportunities. When an opportunity gets to a certain level you could ask the system to give you more information, do a drill down, or transfer it into its own workspace. It may be: "Ellie, can you send the broker an email, request a detailed financial pack, and when you get the email back store it in the workspace so that we can start doing some screening on that opportunity." So this is the start of the life cycle, where we take a dashboard and manage lots of different opportunities. This is hospitality. It could very easily be development opportunities in a particular country, infrastructure, any criteria you want to set. It's a way of monitoring the marketplace.

If I come back to my workspaces, I'm going to go into this workspace called Folsom Corridor. Let's say we've taken an opportunity and we now want to do some basic screening, some underwriting and some basic analysis. I've got my workspace, I can look at the various files. I've linked this to my SharePoint folder. I've got an offering memorandum that's come in from a broker, a standard brochure. Already the platform is not just looking at that as a PDF document but classifying the document and extracting key insights about it. Before we've even started there is an understanding by the platform of what is in there. We've got different financial reports coming through, a lot of spreadsheets, financial statements, a T12 as well. The platform is analysing all these documents and recording whether it's a ledger, a valuation report, a T12, to give a better understanding of what those documents are.

The very first thing I might do is say: "Can you scan the entire data room and give me a quick deal screen? I'd like a portfolio overview. Look through all the leases and financial statements and cite everything back to source." Our main agent, Ellie, is the orchestrator agent: "I've screened all fifty documents in here." This is a key differentiator between a verticalised solution linked into your existing enterprise systems and attaching documents into an existing thread: those documents are already understood, and that intelligence layer is already built. We get a description of the analysis Ellie is doing walking through those documents, and there will always be citations back to source. Wherever you see an underline I can bring up the source document Ellie was looking at, with the actual passage highlighted. Here, for example, the Strand lease section 3.3, and you can see 3.3 highlighted in the source document. It's a way of building up trust in the platform: rather than an AI coming back and saying "this is the value", you're shown exactly how the information is built up. Ellie also gives me the calculated values: anything calculated is underlined in red, and I can show the working for all of those numbers. It's not a number with no background. We've heard from a lot of our customers that they don't just want the AI to give them an answer; they want to understand the answer so they can defend it later at the IC level. As we walk through the valuation, all the way through we can follow the reasoning and the logic. Ellie finds a lot of discrepancies, describing what is there and what is not there, and provides a spreadsheet summarising all of this information.

I can expand that spreadsheet and get the inputs. I can go into the lease extract side: all of the leases in my data folder, all of the values extracted into fields and put into a spreadsheet. You might want to use this as input into an ERP or one of your other enterprise systems. You're using the platform to clean up and understand scanned, unstructured data, pull it out, and send it into any other systems you have. In our extraction we've got the lease extract, the rent roll and the NOI, the debt, the valuations, everything pulled out into a structured format in this sheet.

One of the things Ellie said above was that there seemed to be an issue between the financial statement and the lease figures. So I've asked Ellie in this thread: "Can you go deeper on the financials? Look at the T12 and does it tie back to the lease documents?" Straight away: the T12 adds up internally, but doesn't tie to the leases. The financial bookings don't match some of the clauses in the leases. We've got a portfolio NOI of 263, which we can tie back to the spreadsheets, but there are various clauses in here: the lease steps rent by three percent compounding. There's a specific clause in the lease documents saying the leases should be stepping up by three percent, and that's been missed in the financial statements. We can work that all the way through to the impact and what a lease-compliant NOI would be. This is normal analysis that any analyst would do, but it would take them a lot of time. All I have to do as the analyst is say: "Great, use the lease-aligned figures and restate the in-place NOI. I'm making a human-in-the-loop decision here: accept that reconciled number as our basis." Ellie restates everything for the NOI: we jump to 288, up from the original 263, and get the restated graphics. The next step is: "Can we create the screening memo for the investment committee based on those figures?" Ellie pulls together an investment committee memo with all of the figures from those documents, structured, in a standard template. Our customers normally upload their templates into the system and Ellie learns from them, so these documents and spreadsheets would be branded with the template you use yourselves.

Let's say we've gone to the investment committee, approved everything, and a couple of months down the road someone says: "Why did you use that lease number and T12? I'm not sure why the AI came up with those numbers." One aspect of the system is that I can say: "Run the regulator audit skill." We audit everything that's gone on in this workspace and generate an audit report of all the work we've done. This could be taken to a regulatory authority, to an audit committee, or used as a defence at the investment committee. We go through all the evidence: the threads in the workspace, the activity, who's done what, where the AI has done something and where I've directed it. Here I'm directing the restatement of the NOI; I want to create the screening memo from the reconciled figures, and so on. All of those decisions, the human actions and the arithmetic get stored in this audit report. Everything is always defensible against any committee or anyone coming down the line later wanting to know how you got to a particular number. That is one of the key points of using a platform like this as opposed to putting documents through a general-purpose assistant and getting an answer back which might be fine, but can you defend it later in an enterprise situation?

The next level is heavy-duty modelling. I've said: "Build a five-year IRR model using all these restated figures, and create a spreadsheet I can download and work with." Ellie builds a complete five-year cash flow model with all of the returns, and I can download that spreadsheet. We've asked Ellie to create all of that modelling around unstructured data, and I now have a spreadsheet I can share and work with. It's got all the formulas I need: my cash flow, my returns, the sensitivity calculations. I can work out my levered and unlevered IRR.

That's great as a particular model. But say I want to start adding different variations and scenarios. This is where, using a normal LLM, things take a long time. LLMs are large language models, not large financial models; they're not deterministic by nature. Creating a model you want to be repeatable really means converting it into a mathematical program, where every time you run it you get the same answer. We have a module in our system called Micro Modeller. It takes a spreadsheet you can import. I've taken the five-year IRR model Ellie created and asked: can we model this visually? It's a complicated system: thousands of formulas, ten sheets. We have the concept of bricks: we've taken 263 bricks representing everything in that spreadsheet, all tied together, and as we step through we work towards the returns, the levered and unlevered IRR. Once stored, we have that spreadsheet represented as a deterministic mathematical program we can use to run different scenarios. I could say: "Try a five-year hold period, step through these scenarios," and build out all of those models. We're now generating brand-new cash flow forecasts with the different scenarios mapped through. I've picked five models here, generated in a few seconds, and I can see how my returns change based on the scenarios. Each scenario is a spreadsheet I can download and work with, or I can change the scenarios here. If you threw that back into a copilot-type scenario you could be taking five, ten, twenty minutes to work through, and that's if you can get the complexity of the model represented correctly, because LLMs are not designed for that kind of financial modelling. We have a whole library of bricks in the platform, which allow us to model lots of different spreadsheets. It's open and transparent: each brick has a contract with the other bricks and how spreadsheets are modelled, and these can be built out to model scenarios specific to your organisation; we help our customers build them. And once you have a model visualised, you can turn it into a simple prompt, which becomes a company asset: a library of your standard underwriting prompts. For any deal coming in you paste the prompt in and within a few seconds you have a completely new underwriting model in a spreadsheet, run variations and scenarios, find the one you like, download it and share it with the team. A deterministic approach to the modelling.

The next piece: when I was prompting Ellie and running through those analyses, I was using normal terms, normal language: go study the T12, look at this lease. If you try to do that with a horizontal LLM you end up with pages of prompt engineering to get the right context in place. With a platform like Ellie, she's already trained in the whole life cycle of commercial real assets, from investment management through asset management to financial operations. You can go to atlas.fifthdimensionai.com; we're very transparent about the information. You can register and see the context: here's the piece around investment management and the knowledge we build up. This is an individual process; we've mapped out over ninety processes to date across the industry. This one is deal screening, the initial deal screening and fit assessment: how the industry measures it, where it typically goes wrong, the cost of getting it wrong, how you automate it, and the automation ladder: start with something simple and scale up as you become more AI native. Very specific for APAC: each geographic region has different nuances. I'm looking at the Singapore version; I can just as easily switch to the Australian version and see specific changes and nuances to deal screening in Australia. For example in Singapore with lease renewals, you don't step up a lease during the term, you wait until the end of the term, which is very different to the UK. A lot of this geographic knowledge we encode into the platform as context, and you can set that up per workspace: a deal in Singapore in one workspace, an Australian deal in another, while your corporate reporting might be in Dubai and you want a roll-up using a different set of regulations. All of that is handled by the context and knowledge Ellie holds within the platform, set per workspace.

What we've just done is go through everything from deal sourcing, taking an opportunity and putting it into a collaboration workspace, analysing it into a deal screen, figuring out which numbers to use for the investment committee, auditing and defending those numbers, building a complete underwriting and cash flow model, taking that into the Micro Modeller module and running different scenarios, all while using the context and knowledge of the different countries we want to work in.

Fifth Dimension for deal teams

From data room to cited IC memo

Video details
Roles
Investor / LP, Owner, Asset Manager
Processes
Acquisitions, Due Diligence, Lease Abstraction, Valuation, Platform Features
Geographies
APAC
Tags
webinar, beyond-copilot, hospitality, micro-modeller, underwriting, audit-trail, beyond-copilot-walkthrough