EKSuper (JUL25)(AI)
Updated as of 13-AUG-25
EKS AI Summary (Test 1)
EKSuper AI Summary (TB) (Test 2)
EKSuper AI Summary (Test3)
// Next Versions Notes //
Since Test 3 works, we can start working on an AI Assessment Report based on uploaded details.
Transactional details, qty, and pricing valuations will stay within Excel for transparency purposes.
EKSuper AI Results (Test3)
AI changed our Fund name to Investment Type. Excel we showed by Fund (with collapsed rows), but after creating subtotals, AI defined these as Types. We’ll keep this for now.
AI recommended using Scenario Overlays and provided 3 versions.
EKSuper Scenario Overlays (Test3)
We won’t comment whether or not these scenarios work for us, but it looks AI is looking at these scenario overlays as target, partial, or growth models. We haven’t reflected any assumptions or cash flow in these overlays, so may decide to incorporate cash flow modeling and create an overlay that works for us in the next versions.
// Test 3 Notes //
AI extract 2 structured tables after several revisions of the Excel report layout.
We initially created 2 columns (Act%) in EKS_Act on the right side table, and EKS% row after Total EKS. Initially, AI extracted the Act%, but showed EKS% as a separate table, even though it correctly represented 2 distinct owners. Second version, we removed Act%, but AI still showed EKS% as a separate table. Somehow, AI doesn’t like showing metrics after the Total calculation and doesn’t assume it’s within the same table. Another version was removing EKS% row, but kept Act%. This upload, however, showed Act% column results correctly. Since we already had Act% in the Cat table, there wasn’t a point in including Act% in this table, so we removed it. Somehow, AI doesn’t prefer % metrics as rows, but prefers it shown as columns, and therefore split % allocations shouldn’t be included in the initial upload, but can be created as separate prompt.
Second Cat table all extracts are accurate. No table header, and no bolded rows, but details are correct.
Another revision that we did was removing the individual funds, and instead created as separate totals within Excel, while collapsing the rows. Using this method allows for us to subtotal by Category when comparing Act$ and Tar$.
Although the EKS_Act shows KS and ES separately, we’ve created column EKS_Act as a total, and linking this total to the EKS_Cat table. Doesn’t make sense to show EKS separately for targets, so consolidated totals work out better.
Act$ and Tar$ are linked from our EKSuper Excel report. Only Act% and Tar% are calculated.
// Testing 1-2 Notes //
Based on the first two tests, AI can extract all 7 structured tables with accuracy (excluding accurate headers).
Another test we did was to change the quantity for one of our funds. AI somehow assumed it was a Buy order, and automatically created a Buy Transaction entry even though we never uploaded our initial transaction log. AI also assumed that existing tables aren’t associated or correlated, so they’re considered independent tables even though our Excel tables are linked. This is most likely caused by non-prompting within AI and AI assumed all of the tables and amounts are independent and static as opposed to linked and updated instataneously.
Based on these test results, we can use AI for consolidated reporting metrics assessment, as opposed to transactional entries. What this means is, AI can make assessments for Actuals vs Targets, %’s allocations, and by Categories for EKS. However, the challenge would be by individual funds due to its nature of its non-static environment. In Excel, would be much easier to update for individual funds and link the tables instead of within AI. Once we create the Excel summary report, we can then upload to AI and perform our metrics assessment from there.
We’ll focus on doing this for Test 3-4.
// Test 2 Notes //
AI passed extracting all structured tables
AI failed to use our table headers and created its own
// Test 1 Notes //
Flow: EKS_Input (Set Tar$) -> EKS_Tar% (Set Tar%)
Actuals: EKS_Qty / EKS_CP_History
Pending to include for Act: Cash / Cash Inv / Balanced
AI failed to extract structured tables in mid and right sections of the sheet
// Test Notes //
AI does an exceptional job when calculating using % metrics. Prompting is easy once we create the structured table and set the identifiers. Regarding the initial upload, it can hallucinate when extracting the data details (PP, Qty) after uploading the PDF snipping on the Excel template, so would definitely need to verify all uploaded information, and can prompt AI to make any necessary changes.
We’ll be including a few % metrics on the AI, which includes T% and EKS%. T% is the Fund % allocation, and the EKS% will be ES/KS % allocation. Definitely nice to have when making our assessments.
// Next Phase //
[UR G/(L)] We can calculate UR G/(L), but would require CP for that Fund. CP is 1:1, so CP for the Fund.
[APP] APP is the Average Purchase Price, so it’s the weighted average of the total purchases divided by the total quantity. APP is not 1:1, as it’s specific to the ES/KS. We also cannot assume that every time we make a purchase for KS, we do the same for ES, which is not the case due to cash liquidity and allocation methods.
[TPP] TPP is the Target Purchase Price, which we calculate using different strategies. We’ve now separated EKS purchases from EFI, so now EKS is now distinct and is 1:1. KS uses calculations in Excel, and ES links to KS TPP. We don’t think TPP should be calculated by AI as we risk hallucinations, so for now, we’ll keep our Excel TPP calculations.
[TAPP] TAPP is the Target Average Purchase Price, which we calculate in Excel. It’s definitely a risk with AI on hallucinations and not worth the extra effort, so we’ll skip.
[VAPP] Exclude from AI.
[LPP] LPP is Last Purchase Price. Don’t think AI would know how to automate this as it includes 3+ identifiers.
[52WH/52WL] Excel feed.
