Dima Trubnikov

← All work

AI underwriting platform for institutional real estate · design exercise, client confidential

AI Underwriting — Five Decisions, Not a Thousand Actions

AI Underwriting — Five Decisions, Not a Thousand Actions

AI Underwriting — Five Decisions, Not a Thousand Actions

Ten agents read 47 documents. The analyst should only see where their work stopped.

Ten agents read 47 documents. The analyst should only see where their work stopped.

Client

AI underwriting platform for institutional real estate · design exercise, client confidential

My role

Product Design Lead (candidate exercise) — concept, UX, working prototype, design system

Timeline

September 2026 · one working day

Impact

47 documents · 10 agents · 1,240 checks → 5 decisions on screen · returns recalculate from every choice · working prototype in one day

The challenge

The challenge

The client builds a multi-agent system that underwrites real-estate acquisitions for institutional investors. The brief: design the core experience for an analyst working through an £80m UK logistics deal, and make the AI’s work trustworthy without making the analyst manage it. Behind the deal sit 47 documents, 10 specialist agents and 1,240 consistency checks. Every product in this space is tempted to show that machinery: agent timelines, activity feeds, a confidence score on everything. It demos well, and it hands the analyst a thousand things to supervise. But the analyst’s job is not to watch agents. It is to form a view they can defend at an investment committee. The question became: what is the smallest surface that keeps the evidence, the control and the consequence of every judgement?

The client builds a multi-agent system that underwrites real-estate acquisitions for institutional investors. The brief: design the core experience for an analyst working through an £80m UK logistics deal, and make the AI’s work trustworthy without making the analyst manage it. Behind the deal sit 47 documents, 10 specialist agents and 1,240 consistency checks. Every product in this space is tempted to show that machinery: agent timelines, activity feeds, a confidence score on everything. It demos well, and it hands the analyst a thousand things to supervise. But the analyst’s job is not to watch agents. It is to form a view they can defend at an investment committee. The question became: what is the smallest surface that keeps the evidence, the control and the consequence of every judgement?

Fig. 01 — Where the work stopsThe pipeline, collapsed
Deal pack
47
documents
Offering memorandum, leases, rent roll, title, surveys
Agents
10
specialist reads
1,240 consistency checks across the pack
Extracted
412
facts
407 uncontested, folded away in grey
Discrepancies
18
found
13 resolved where one source is clearly authoritative
On screen
5
judgements
What no agent can decide for the analyst
1
Reversionary rent assumed at £8.50/sq ft
12% above the three most recent lettings within 5 miles
Assumption
2
Iceland break option: rent roll contradicts the lease
Rent roll says mutual break; executed lease says tenant-only
Conflict
3
Brackley Components: covenant deteriorating
13% of income, break in 15 months, going-concern qualification filed
Tenant risk
4
DHL service charge capped below actual cost
Schedule 6 caps recovery at £0.85/sq ft; running cost is £1.12
Legal
5
Consented development next door shares the only lorry access
Not in any deal document; found on the council planning register
External

The verdict line and these five rows are the whole entry screen. The deal and its figures are synthetic but internally consistent.

What I did

What I did

1. Showed only where the work stopped. The system extracted 412 facts; 407 are uncontested and fold away in grey. Of 18 discrepancies it resolved 13 on its own, where one source is clearly authoritative. What reaches the analyst is a verdict line and five rows: the judgements no agent can make. The machinery is one collapsed line, not a dashboard.

1. Showed only where the work stopped. The system extracted 412 facts; 407 are uncontested and fold away in grey. Of 18 discrepancies it resolved 13 on its own, where one source is clearly authoritative. What reaches the analyst is a verdict line and five rows: the judgements no agent can make. The machinery is one collapsed line, not a dashboard.

2. Made provenance the structure of the table. Each row opens in place into three columns: what the document says, what the system made of it, and what you decide. A citation opens the actual page with the sentence highlighted. I left out confidence scores on purpose: “92% confident” hides the difference between an extracted fact and a multi-step inference, while a citation to page 34 promises nothing and shows the source.

2. Made provenance the structure of the table. Each row opens in place into three columns: what the document says, what the system made of it, and what you decide. A citation opens the actual page with the sentence highlighted. I left out confidence scores on purpose: “92% confident” hides the difference between an extracted fact and a multi-step inference, while a citation to page 34 promises nothing and shows the source.

Fig. 02 — The working prototypeInteractive · try it
1Open item 1, the reversionary rent2Choose “Use comparable evidence (£7.60)”, then accept3Watch the pinned strip: 14.2% becomes 12.8%, the mandate flips4Resolve all five to see the recommendation
underwriting workspace · Mercia Gateway
The system prepares five decisions; the analyst makes each one and sends the pack.

The prototype itself, running inside this page. The deal, tenants and figures are synthetic but internally consistent; returns are computed from your choices.

3. Kept the consequence in the same frame as the choice. A pinned strip carries the live return. Swap the seller’s £8.50 rent for the comparable £7.60 and levered IRR drops from 14.2% to 12.8% as you watch, while the mandate line flips from “clears by 120 bps” to “short by 20 bps”. There is no recalculate button anywhere in the product.

3. Kept the consequence in the same frame as the choice. A pinned strip carries the live return. Swap the seller’s £8.50 rent for the comparable £7.60 and levered IRR drops from 14.2% to 12.8% as you watch, while the mandate line flips from “clears by 120 bps” to “short by 20 bps”. There is no recalculate button anywhere in the product.

4. Let the product disagree with the deal. A tenth agent reads outside the pack — planning registers, infrastructure, tenant news — and finds a consented development that shares the site’s only lorry access. On the case path the five judgements take the return to 12.0% against a 13.0% hurdle, so the primary action is not Approve. It is “Send to VP — recommend counter at £75.4m”. Leave the assumptions alone and it recommends proceeding.

4. Let the product disagree with the deal. A tenth agent reads outside the pack — planning registers, infrastructure, tenant news — and finds a consented development that shares the site’s only lorry access. On the case path the five judgements take the return to 12.0% against a 13.0% hurdle, so the primary action is not Approve. It is “Send to VP — recommend counter at £75.4m”. Leave the assumptions alone and it recommends proceeding.

Fig. 03 — A tenth agent looks outwardHow the product can say no
Nine agents · inside the pack
Read what the seller provided
· Leases and rent roll· Offering memorandum· Title and surveys· Service-charge schedules
Agent 10 · outside the pack
Reads what nobody had to disclose
· Planning registers· Infrastructure schemes· Tenant news
Issues directions to check, never conclusions, and publishes where it looked.
Item 5 · source column
Nothing
A consented development next door shares the site's only lorry access.
Traced to the assumption it threatens, then handed to the analyst.
Five judgements, case path
Adjusted levered IRR 12.0% against a 13.0% hurdle
Send to VP — recommend counter at £75.4m
Assumptions left alone
Levered IRR 14.2%, clears by 120 bps
Send to VP — recommend proceeding

The recommendation is computed from the decisions the analyst actually made. A product that only ever confirms is a product that sells deals.

The solution

The solution

A working browser prototype of the underwriting flow: a verdict, five decisions, cited evidence, editable assumptions, returns that recalculate from the analyst’s own choices, and a handoff to the VP with the recommendation computed from them. Accepting a risk unchanged needs a written reason and an owner, and that reason travels into the investment pack. I deliberately left out a chat panel, an agent-activity view, modals and section navigation. Each would have put the complexity back. The deal, tenants and figures are invented but internally consistent, and the arithmetic ties together on every screen. The brief allowed one working day.

A working browser prototype of the underwriting flow: a verdict, five decisions, cited evidence, editable assumptions, returns that recalculate from the analyst’s own choices, and a handoff to the VP with the recommendation computed from them. Accepting a risk unchanged needs a written reason and an owner, and that reason travels into the investment pack. I deliberately left out a chat panel, an agent-activity view, modals and section navigation. Each would have put the complexity back. The deal, tenants and figures are invented but internally consistent, and the arithmetic ties together on every screen. The brief allowed one working day.

Fig. 04 — Where the shape comes fromProcurement, 2019 → underwriting, 2026
EIT · procurement · 2019
Designed value
Requested value
Agreed value
AI underwriting · 2026
What the document says
What the system made of it
What you decide

The three-column table from my procurement platform for offshore equipment transferred directly. What had to be added is the evidence layer: with a human counterparty you can ask why a number is 0.95. A model has nobody to ask, so the proof has to arrive before the question. Read the EIT case →

What I’d do differently

What I’d do differently

I designed the analyst’s side and stopped at the handoff. The decision only matters if a VP can challenge it, so the reviewer’s view — what changed, where the analyst overrode the system, what to probe — deserved a rough sketch in the same day. And the central bet is untested: I found no research showing that citations make people trust AI output more. Next I would remove evidence access for some sessions and watch whether analysts open sources less over time. If they don’t, the product isn’t earning trust, just being tolerated.

I designed the analyst’s side and stopped at the handoff. The decision only matters if a VP can challenge it, so the reviewer’s view — what changed, where the analyst overrode the system, what to probe — deserved a rough sketch in the same day. And the central bet is untested: I found no research showing that citations make people trust AI output more. Next I would remove evidence access for some sessions and watch whether analysts open sources less over time. If they don’t, the product isn’t earning trust, just being tolerated.

More work

All case studies →

All case studies →

All case studies →

d.trubnikov@me.com

© 2026 Dima Trubnikov · Vilnius, Lithuania (EU)