CreditIntel · the flagship
The Credit Portfolio Diagnostic.
A diagnostic is one pass over one customer portfolio, ending in a document. You define which accounts are in scope, hand over a single extract, and get back a review written to be read in a meeting rather than clicked through in a browser.
Reconciliation is the gate on the whole engagement. Every figure printed in the review is recomputed from your extract and checked against it, and a review that fails that check is not delivered until it passes.
Who it is for
Teams that own a receivables number they have to defend.
Credit managers, collections leads, controllers and the finance chief above them, at mid-sized businesses selling on terms.
The constraint this addresses is not analytical skill. It is that a portfolio view has to be built before it can be judged, and building it consumes the hours that were meant for judging it. Balances live in the ERP. Aging arrives as its own export. Limits sit in a spreadsheet one person maintains. Collector context exists as notes, or as memory. Putting those four things beside each other reliably, in the same shape as last month, is most of the job.
Two signals say the fit is wrong. If the portfolio view is already assembled, already checked, and already comparable period to period, the bottleneck is somewhere else and this will not move it. If nobody can say which accounts belong in scope, the engagement has no boundary and should not start yet.
The signal that says the fit is right is narrower: an analyst who could reach the right conclusion given the view, and who does not currently get the view in time.
What it reviews
Four measurements, each answering a question the last one cannot.
They are computed separately on purpose. Blended into a single score, each one stops being able to contradict the others, which is exactly when it stops being useful.
Distribution, not the total
A portfolio total is a summary statistic, and summary statistics are where deterioration goes to hide. What gets measured instead is the shape underneath: which accounts hold the balance, how far past terms each slice has traveled, and how much value crossed from one bucket into an older one since the last period. Migration between periods is the reading that matters, because a steady headline can sit on top of a segment moving quickly in one direction.
- balance held per account, and the shift since the prior period
- value that moved into an older bucket rather than the standing bucket totals
- how much of the portfolio depends on a small number of names
- accounts worsening separated from accounts recovering
Whether reality still matches the approval
A limit records an authorization given by a named person on a named date, against the facts available then. It does not expire on its own, so the risk is silent drift: the customer changes, the approval does not, and nobody is prompted because nothing broke. Measuring current draw against the approval, and both against the age of the approval, turns that drift into something with a date attached.
- draw against the approved ceiling, per account
- accounts sitting outside their approval, and for how long
- breaches carrying no recorded hold or release either way
- age of the last recorded review, set beside current draw
Timing drift, which nobody reports
A disputed invoice announces itself and gets worked, because someone is on the other end of it. The costlier pattern is silent: settlement dates sliding a few days per cycle, with no dispute raised, no complaint logged, and no threshold crossed that would trigger a call. Because each individual slip is unremarkable, only the trend across cycles reveals it, so trend is what gets computed.
- days beyond terms, plain and weighted by value
- how long since the account last settled anything
- direction of travel across cycles rather than a single reading
- late settlement arriving with no dispute reason recorded
The queue that has to be worked
Accounts that broke a rule the team already agreed on, each carrying the rule that surfaced it. An exception without its reason attached is just another row to argue about.
- over the approved limit
- on hold or orders blocked
- rating worsened since the last review
- no review recorded in twelve months
What it needs from you
One export you already trust.
No system access is required beyond the extracts you choose to share. There is nothing to install and no login to maintain.
A customer and receivables extract
Whatever your system already produces, from NetSuite, Snowflake, or another source. The fields that matter are the ones a review runs on.
- customer identifier and name
- open AR by aging bucket
- approved credit limit and current balance
- invoice and payment dates, or days beyond terms
Your review rules, if you have them written down
Thresholds, hold conditions, risk scale, and what counts as an exception. If they are not written down yet, the first pass writes down what the review actually assumed, so the second pass is comparable to the first.
One named person who can confirm what a field means
Field mapping is where reviews go wrong. Fifteen minutes with someone who knows the export prevents a report built on the wrong column.
What you receive
Four artifacts, sized for the meeting they are used in.
The deliverable is prose and tables, not access to a tool. Nothing here needs an account, a license, or a person to keep it running.
The written review
An executive summary, the movement since the last period, and the names that changed enough to be worth a decision. Written to be read, not clicked through.
Account level briefs
A page for each customer that matters: exposure, aging, utilization, rating history, and payment behavior together, so a reviewer can understand the change before deciding.
The watchlist
Accounts flagged for consecutive weeks, so a slow slide is visible as a pattern rather than as one bad week that gets explained away.
Caveats and confidence
What the review is unsure about and why, plus the analyst follow up it recommends. Every judgment is labeled as analyst inference.
The review can arrive as a document or as a working screening desk in the browser. Both are built from the same reconciled figures. Both put the decision with your credit professionals.
How it runs
Three stages, and a gate between the second and the third.
Fixed scope agreed in advance. The only thing that crosses from your side to mine is the extract you choose to send.
Establish what the extract actually says
Before anything is computed, each column is given a written definition and that definition is confirmed with whoever owns the source. This stage is unglamorous and it is where most automated reporting fails, because a column named plausibly is not the same as a column meaning what its name suggests. The output of this stage is a mapping document, and it belongs to you afterwards.
Compute the four measurements, then prove them
Distribution, authority, timing and exceptions are calculated independently, then reconciled back against the extract they came from. This is the gate: totals either agree with the source or the run stops and the disagreement gets investigated. A number that cannot be traced back to a row does not get to appear in a document that someone will act on.
Write it up, including what is unresolved
The findings are written out with the uncertainty attached rather than smoothed away, so a reader can tell measurement apart from interpretation. Recommendations are phrased as what to examine next, never as an instruction. Authority over every credit decision stays exactly where it started, with your people.
The mechanics of the weekly version of this review are written up in full, with a worked example: how to prepare a weekly customer credit review.
Boundaries
Four limits, and what each one means in practice.
These are constraints on the engagement, not disclaimers at the bottom of it. Each one changes what you should expect to receive.
Underwriting authority does not move
Nothing in the deliverable approves, declines, holds or releases anything. Where the review reaches a conclusion the extract cannot fully support, that conclusion is marked as inference so a reader can weigh it accordingly. In practice: you will find opinions in the document, and they will be visibly labeled as opinions.
It removes assembly, not judgment
The hours this recovers are the ones spent gathering and formatting, and those hours come back to the same analysts. In practice: a credit function does not get smaller because of this work, it gets its review preparation time back and spends it on the accounts that need a conversation.
It never becomes infrastructure
The engagement reads a file you send and returns a document. It does not connect to a system, hold credentials, run on a schedule, or leave anything behind that needs maintaining. In practice: when the engagement ends, nothing has to be decommissioned and no vendor sits between your team and your data.
Findings are stated, results are not forecast
There is no recovery figure, no return calculation, and no assertion that a given account would otherwise have gone bad. Those depend on decisions and events after the review. In practice: the document tells you what the portfolio currently shows, and stops there.
The same reconciliation rule governs the demonstrations and paid work equally. Figures in the demonstration book are generated, then checked back against the generated source by script, and a build that disagrees with its source does not complete. Applying the identical gate to invented data is the only way to show the gate exists at all, since a client portfolio cannot be published to prove it.
no client is named anywhere on this site, because no client engagement is being claimedDemonstrations
Two ways to look at the format before you send anything.
Both run on a demonstration book of 48 customers and twelve months of history. Every company, balance, rating and history in them is fictional.
Both demonstration pages carry a noindex directive and are excluded from search results on purpose. They are working demonstrations, not published marketing pages, and they contain synthetic data only. Nothing in them is client data or a record of any real company.
Start here
Pick the portfolio nobody can summarize in a meeting.
The quickest way to judge this is to look at the output format against a book you already know. That takes a quarter of an hour on a shared screen, using the invented portfolio, and the useful outcome is often deciding it is not what you need.