Predictions
A Prediction is an account-level score that estimates the probability of churn or renewal — calculated continuously from your actual customer data, not a manually configured formula. Where health scores ask you to decide in advance what matters, FunnelStory's prediction models learn the patterns from your own historical outcomes: which accounts renewed, which churned, and what their data looked like in the months before.
The result is a score grounded in the specific reality of your customer base, not a generic industry template.
The Health Score
Every account receives a health score from 0 to 100.
- 50 is neutral — no strong signal in either direction
- Below 50 — increasing churn risk
- Above 50 — healthy trajectory trending toward renewal
The score is a net result of two competing signals: the probability the account will stay, weighed against the probability it will churn. When both signals are strong, the score reflects genuine uncertainty — an account with high product usage but also high support escalations, for example, will land near the middle until the pattern resolves.
Predicted Outcome
Each account is assigned a predicted outcome alongside its health score:
| Outcome | What it means |
|---|---|
| Churn | The account matches patterns historically associated with churn |
| Retention | The account matches patterns historically associated with renewal |
| Neutral | No strong signal in either direction |
Confidence
Every prediction includes a confidence level reflecting how clearly the data matches the predicted outcome:
| Confidence | What it means |
|---|---|
| High | Strong, clear signal — the prediction is reliable and actionable |
| Medium | Moderate signal — worth investigating and acting on |
| Low | Weak signal — use alongside other context |
| Neutral | Insufficient data to form a reliable prediction |
Low confidence most commonly appears for newer accounts that haven't yet accumulated enough history to match patterns clearly.
Driving Factors
Each prediction surfaces the specific factors contributing to the score, split into two categories:
- Increase/Maintain these values — factors currently supporting retention. Protecting these is as important as addressing risk signals.
- Decrease/Maintain these values — factors that are pushing the score toward churn. These are your intervention priorities.
Each factor shows the account's current value on a min-max scale relative to the broader population. This makes it immediately clear whether an account is above or below average on any given signal — and by how much.
Driving factors pull from both structured data (product usage events, CRM attributes, support activity) and unstructured data (conversation sentiment, ticket themes, meeting transcripts). This combination is what allows the model to surface signals that pure usage-based health scores miss entirely.
What-If Analysis
The What-If Analysis lets you simulate how changing an account's data would affect its prediction.
Enter a hypothetical value for any driving factor — reduced usage, fewer active users, resolved support tickets — and see the projected impact on the health score. This is useful for prioritizing which gaps to close ahead of a renewal conversation: if increasing one metric would move the score substantially, that's where to focus.
How Predictions Learn Your Business
FunnelStory models are trained on your specific outcomes, not a generic baseline. The system learns what "churn" and "retention" look like in your customer base by analyzing historical accounts — which ones renewed, which ones churned, and what combinations of signals preceded each.
This is configured through Revenue Tags: you define what a churned account looks like and what a retained account looks like, using filters or specific account examples. The prediction model uses these labeled examples as its training set.
The more precise your Revenue Tags, the more accurately the model can learn the patterns that matter for your specific business. Your FunnelStory team works with you during setup to configure these correctly.
Needle Mover Weights
As part of model configuration, each Needle Mover type is assigned an impact weight — controlling how much influence conversation signals (competitor mentions, pricing concerns, personnel changes, etc.) have on the prediction score relative to structured activity data.
How Predictions Improve Over Time
Prediction models continuously improve as new outcomes are recorded.
When an account predicted to churn actually churns — or an account predicted to renew actually renews — that outcome is used to validate and refine the model in the next training cycle. Missed predictions are equally valuable: an account the model predicted as healthy that churned unexpectedly teaches the model to look for signals it may have underweighted.
This feedback loop means the model becomes more accurate over time, adapting to changes in your customer behavior, your product, and your market. You can trigger a manual retrain from the Revenue Tags configuration page when you've made significant changes to your tagging criteria.
Per-Product Predictions
For accounts with multiple products, FunnelStory generates per-product predictions — a separate health score and driving factors breakdown for each product line.
This is useful when:
- Different products have different renewal timelines or contract structures
- A single account has separate CSM ownership for different products
- You want to isolate which product relationship is at risk before a consolidated renewal conversation
Acting on Predictions
Predictions are designed to trigger action, not just inform awareness. From any account's prediction view, you can:
- Review driving factors — understand exactly what is moving the score before engaging the customer
- Run a What-If analysis — model which interventions would have the most impact
- Jump to Needle Movers — see the specific conversation signals and behavioral changes behind the prediction
- Launch a playbook — execute a structured response workflow directly from the prediction detail
- Create a CRM task — push the risk or opportunity to Salesforce or HubSpot for Account Executive follow-up
- Ask Renari — get an AI-synthesized action recommendation with full account context
Relationship to Needle Movers
Predictions and Needle Movers are complementary, not redundant.
A Prediction gives you the score — the probability that an account will churn or expand. A Needle Mover gives you the reason — the specific, sourced signal (a competitor mentioned in a QBR, a champion who has gone quiet, an unresolved pricing concern) that is moving that probability.
Together they provide both the "what" and the "why" needed to take confident action.
Related
- Needle Movers — the specific signals driving prediction scores
- Customer Intelligence Graph — how prediction scores are computed and stored as derived intelligence
- How FunnelStory Works — where predictions fit in the pre-computed intelligence layer
- AI Agents — automating responses when predictions cross risk thresholds