Commercial lending · Credit memos
WorkflowCommercial lendingDrafting commercial credit memos with a workflow the credit officer signs off
A commercial credit memo takes days, and most of that time goes on copying numbers and restating policy. This playbook builds one workflow that does the drafting and leaves the credit judgement with the officer.
Valnora & Elric team5 min read
- Format
- Playbook
- Steps
- 7
- Solution
- Workflow
- Parts used
- 6
A commercial credit memo is where a lending decision gets argued on paper. The officer has to describe the borrower, analyse several years of financials, test the request against credit policy, set out the risks and mitigants, and recommend terms. Supervisory expectations, such as the EBA’s guidelines on loan origination, ask for a documented creditworthiness assessment, and the memo is usually that document.
Much of the time it takes is not judgement. It is retyping figures from PDFs into a spreading template, recalculating ratios, and copying policy limits into the right paragraph. This playbook builds one workflow in Agent Serve that does those parts and hands the officer a draft to challenge, edit and sign. The recommendation stays the officer’s.
From borrower pack to a memo ready for review
Workflow · Credit memo drafting
- 01Borrower packFinancial statements, management accounts and the application, uploaded to Files.
- 02Extract and spreadOCR reads the statements and an Agent block maps line items into a spreading table.
- 03Calculate ratiosA Function block computes debt, coverage and liquidity ratios with your fixed formulas.
- 04Policy checksRatios and terms are compared with credit policy limits; breaches become listed exceptions.
- 05Draft sectionsAgent blocks write each memo section from the data, citing the policy passages they used.
- 06Credit officer reviewA Human block pauses until the officer edits and signs off the draft.
Agree what the workflow may write
Start with the memo template and go through it section by section with credit risk. Most teams land on three groups:
- Drafted by the workflow: borrower summary from the application, historical financial analysis, ratio tables, policy exceptions list.
- Drafted, then rewritten by the officer: industry and market context, risk factors.
- Written only by the officer: risk rating, mitigants for exceptions, recommendation and conditions.
Write this split down. It goes into the Agent blocks’ instructions, and it is the first thing a second-line reviewer will ask about.
Turn the borrower pack into structured numbers
The workflow starts when an analyst uploads the borrower pack to Files and runs it with the deal reference. A File block reads each document, and an OCR block, AWS Textract or Reducto, extracts the text and tables from scanned statements.
An Agent block then maps line items to your spreading template: revenue, cost of sales, EBITDA, interest, debt, current assets and so on. It writes one row per line item per year into a table, with the source page beside each figure. The analyst checks the spread in the table before anything else runs. Spreading errors are the most common cause of a wrong memo, so this check is worth more than any later one.
Calculate ratios with fixed formulas, not a model
Ratios must be the same every time and match how your credit team defines them. Put them in a Function block, which runs code you write, not a model’s arithmetic: net debt to EBITDA, debt service coverage, interest cover, current ratio, and any sector-specific measures. Pro forma versions include the proposed facility.
The results are written back to the spreading table, so the officer and any reviewer can see the inputs and outputs of every ratio.
Check the request against credit policy
Credit policy limits differ by sector, product and rating. Keep them in a table: one row per sector and product, with the limit for each ratio, maximum tenor and minimum security cover. A Condition block or a set of rules compares the deal’s ratios and terms with the matching row.
Every breach becomes an entry on an exceptions list, with the limit, the actual value and the policy reference. The workflow does not decide whether an exception is acceptable. It makes sure none is missed and leaves the mitigant and the decision to the officer.
Draft the sections from data and policy
With the numbers fixed and the exceptions known, Agent blocks draft the sections agreed in step one. Each block gets only what it needs: the spread, the ratios, the exceptions list and the relevant passages from the credit policy, found by a Knowledge block searching your policy documents in the Knowledge Base.
Instructions should require three things: state figures only from the spreading table, name the policy section for any limit mentioned, and mark any statement that is an inference rather than a fact. A Guardrails hallucination check on each section catches statements the inputs do not support.
What the solution produces
Memo section: financial analysis (illustrative)
MemoRevenue grew 8% to 42.1m in FY25, driven by two new distribution contracts. Gross margin held at 31%.
Net debt to EBITDA is 3.1x after the proposed facility, within the 3.5x policy limit for this sector.
Debt service coverage is 1.35x on a pro forma basis, above the 1.25x minimum in the credit policy.
Customer concentration is high: the largest customer is 38% of revenue. See exception 1.
Source data: audited FY24 and FY25 accounts, management accounts to June. Spreading reviewed by analyst.
Policy exceptions list (illustrative)
Checklist- Exception 1: single-customer concentration above the level the policy flags for review. Mitigant to be stated by the officer.
- Exception 2: management accounts are older than the policy allows at approval. Request updated figures.
- No exceptions on debt to EBITDA, coverage, tenor or security cover.
Put the officer in the loop and keep the record
A Human block pauses the run and sends the draft memo, the spread and the exceptions list to the credit officer. The officer edits the draft, writes the rating, mitigants and recommendation, and approves. Only then does the workflow save the final memo to Files under the deal reference, ready for the credit committee pack.
Each run is traced block by block in Logs: which documents were read, what was extracted, which ratios were calculated, which policy passages were used. If a committee member asks where a figure came from, the answer is a click away. Workflows run from a deployed version, so a change to a formula or a prompt is reviewed before it affects live deals.
If borrower data cannot leave your environment, run Agent Serve self-hosted and use your own model keys or a self-hosted model.
Pilot on closed deals before live ones
Run the workflow on ten to twenty deals that have already been approved, and compare its drafts with the memos the officers wrote. Track three things: spreading errors, missed or false exceptions, and how much of each drafted section the officer kept. Fix the spreading mappings and the policy table first; prompt changes come last.
Move to live deals with one team once spreading errors are rare and no exceptions are being missed. Lending teams that also price deals can pair this with a loan pricing scenario model, and insurers will find the same intake pattern in underwriting submission intake.
Controls and governance
Approvals before action
Relays never send email or book meetings without an approval, under any autonomy setting. An approval covers the exact payload the reviewer saw and expires if nobody decides.
Run logs
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Deployed versions
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Self-hosting
Run Agent Serve in your own cloud or data centre, next to the systems it reads.
Bring your own keys
Use your organisation’s own model provider keys at organisation or workspace level.
Zero data retention
Zero data retention with supported model providers, backed by your own provider keys and agreements or self-hosted models. Some provider features keep data by design, such as Gemini Deep Research, so they are outside zero data retention.
SOC 2 ready
The controls SOC 2 looks for are in place. Agent Serve does not claim a SOC 2 report or certification.
More on how Agent Serve handles access, data and audit on the security page.