> ## Documentation Index
> Fetch the complete documentation index at: https://help.daylit.com/llms.txt
> Use this file to discover all available pages before exploring further.
# AI insights: signals, predictions, and strategy plans
> Daylit's AI engine generates insights on every customer and invoice — including risk signals, payment predictions, and recommended collection strategies.
**Not yet fully available:** Some insight types are live today — including risk signals, conversation summaries, and strategy plans. The other signals will be coming soon. This won't affect your normal usage; the most commonly used insights are already up.
Daylit's AI engine runs on a schedule and continuously generates insights for every customer, invoice, and your portfolio as a whole. You don't trigger these manually — they're produced automatically and stored so they're always ready when you open a customer detail page, an invoice record, or the AI Insights tab. Each insight is grounded in your actual data: payment history, communication threads, aging metrics, and behavioral patterns observed over time.
## The 7 insight types
Daylit generates seven insight types — three signals, two conversation summaries, one payment prediction, and one strategy plan. Use the tabs below to explore each one.
Signals detect risk conditions and behavioral patterns at the customer, invoice, and portfolio level. Each signal includes a severity rating, one to three tags from a fixed taxonomy, and metric-backed evidence explaining why the signal fired.
**Customer signal**
Analyzes payment behavior across a customer's full history to detect risk conditions such as:
* Payment behavior worsening or improving versus the customer's own baseline
* Chronic late payment patterns (mild, moderate, or severe)
* Accelerating deterioration month-over-month
* A historically reliable payer who has recently missed expected timing
* Ghosting — unresponsive to outreach despite an open overdue balance
* Churn risk — balance declined sharply, customer gone quiet, remaining balance deeply aged
* High portfolio impact — the customer's overdue balance is a material share of your total AR
The signal also flags positive conditions, such as a customer who is consistently on time or improving their payment behavior.
***
**Invoice signal**
Evaluates the status and risk of a single invoice. Example signals include:
* **Action required** — needs human intervention now
* **Promise to pay active** — customer committed to pay by a date; collection is paused
* **Auto-reminding** — the AI is running a drip campaign; no human action needed yet
* **Disputed** — flags pricing disputes, missing PO numbers, or delivery/quality issues
* **High risk of default** — the AI predicts this invoice is unlikely to be paid
* **Payment imminent** — customer has viewed the invoice, clicked a payment link, or stated it's being processed
***
**Portfolio signal**
Scans your entire AR book for trends that affect the portfolio as a whole — concentration risk, aggregate aging deterioration, shifts in average days-to-pay across all customers, and other portfolio-wide patterns that don't surface when you look at individual accounts in isolation.
***
Conversation summaries distill all communication history into a compact, structured snapshot. They surface the current state of a relationship or an invoice negotiation without requiring you to read every email and call transcript.
**Customer conversation summary**
Summarizes all communications with a customer across every invoice and thread. The summary includes:
* A concise clause (18 words or fewer) describing the current state of the relationship
* The most recent communication timestamp and direction (inbound or outbound)
* Any commitment date the customer has given for payment
* The conversation tone: amicable, cooperative, neutral, frustrated, avoidant, defiant, or unknown
* Which team currently owns the next action — AR, Sales, Operations, Executive, or Customer
* The concrete next step needed to move the conversation forward, split by who owns it (your team vs. the customer)
* One evidence line citing the specific communications that informed the summary
***
**Invoice conversation summary**
Summarizes all communications specifically about a single invoice. In addition to the fields in the customer summary, it includes:
* A **stress level** from 1 to 10, based on the Stress Level Scorecard (1 = proactive/routine, 10 = total silence with the balance deeply aged)
* The **active topic** — what the current to-do list or talking points are in the conversation
* The **desired outcome** — what resolution looks like for this invoice
* The **blocker** — why the desired outcome hasn't been reached yet (internal blocker, external blocker, or both)
***
**Invoice payment date prediction**
The AI forecasts when a specific invoice is likely to be paid, based on the customer's payment history, behavioral patterns, any commitments made in communications, and the current aging status of the invoice.
Each prediction includes:
* **Predicted date** — the AI's best estimate of the payment date (YYYY-MM-DD), or null if there's insufficient data
* **Confidence score** — a 0–1 score indicating how certain the prediction is
* **Reasoning** — a concise clause explaining the basis for the prediction
* **Evidence sources** — which data was used (for example, payment history, communication thread, or behavioral patterns)
* **Basis** — whether the prediction is grounded in an explicit customer promise, historical payment patterns, statistical behavioral modeling, or analyst inference
* **Alternative scenarios** — other possible payment dates with their probabilities, for cases where the outcome is less certain
Payment date predictions are more reliable for customers with longer payment histories. For new accounts with fewer than a handful of closed invoices, the AI will indicate low confidence or return no prediction.
***
**Customer strategy plan**
A recommended collection action for a specific customer, generated by the AI based on the customer's current risk signals, conversation history, open invoice balances, and behavioral patterns.
Each strategy plan includes:
* **Recommended action** — whether to send an email, make a call, or create an internal follow-up task
* **Reasoning** — why this action is recommended given the current account state
* **Urgency score and bucket** — a 0–100 score and one of five categories: CRITICAL, TODAY, THIS\_WEEK, UPCOMING, or INFO
* **Due date** — when the action should be performed
* **Action details** — depending on the recommended action type:
* *Email* — a drafted subject line, body, tone (friendly, urgent, firm, etc.), and template stage
* *Call* — a contact, call script with 3–7 talking points, anticipated objections, and success criteria
* *Internal follow-up* — an assignee, task description, priority level, and context explaining why the task is needed
***
## Where to view insights
| Location | What you'll find |
| ------------------------ | -------------------------------------------------------------------------- |
| **Customer detail page** | Customer conversation summary, customer signal, and customer strategy plan |
| **Invoice detail page** | Invoice conversation summary, invoice signal, and payment date prediction |
## Giving feedback on an insight
You can rate any insight with a thumbs up or thumbs down directly from the insight card. Your feedback is saved and helps the Daylit team understand where the AI is performing well and where it needs improvement. You can also add a short comment to explain your rating.
To remove your rating, click the same button again to toggle it off.
## How often insights are refreshed
Insights are generated automatically on a schedule — you don't need to request them manually. When new data arrives (a payment recorded, an email received, an invoice updated), the relevant insights are queued to refresh so the information you see reflects the current state of your AR book.
If you've just received a significant payment or email and want to see updated insights, check back after a few minutes. The scheduled pipeline processes new data continuously throughout the day.
## Insight confidence and evidence
Every insight includes a confidence indicator. Higher confidence means the AI had strong supporting data — a long payment history, recent communication, or clear behavioral pattern. Lower confidence means the AI is working with limited data, such as a new customer account or a customer with few closed invoices.
Where applicable, insights also include evidence — specific data points or communication excerpts that explain why a signal fired or why a prediction was made. You can use this evidence to verify the AI's reasoning before acting on it.