Stop watching health scores drop and do something about it

This one is for customer success managers, account managers, and anyone who owns renewals.

What: This prompt takes the health score and engagement data already sitting in your Customer Success Workspace and asks Breeze Assistant to do the bit HubSpot's native score doesn't: explain why each account is slipping, rank who to call first, and draft a first-pass outreach message for each one.

Prompt of the week:

A health score tells you the number. It doesn't tell you the story. Health Score Builder is genuinely useful — it pulls engagement, support, and usage signals into one red/amber/green view, so you know an account has dropped from Green to Amber. What it won't do is tell your CSM why, or give them something to say when they pick up the phone.

That gap is where accounts quietly slip away. A score on a dashboard doesn't create urgency by itself. Someone still has to read it, decide it's worth an hour of their day, and then find the right words. On a bad week, that account sits in Amber for a fortnight before anyone acts.

This prompt closes the gap between "the data says something's wrong" and "here's what I'm sending them today." Export the relevant company/contact properties for your at-risk segment, paste them in, and Breeze will rank the accounts by urgency, name the likely driver for each drop, and draft an opening outreach line you can edit and send.

Frictionless doesn't mean invisible. The aim isn't to automate the relationship away from the customer — it's to remove the friction between spotting a risk and acting on it, so the human part of retention happens sooner and better-informed.

Prompt Structure

Role: You are a senior customer success strategist helping a CSM

prioritise a book of at-risk accounts.




Task: Review the exported account data below. For each account with

a Medium or Low health status, identify the most likely driver of

the score drop, rank the accounts by urgency (highest risk and

highest commercial value first), and draft a short, personalised

outreach opening line for each one.




Context: [PASTE EXPORTED DATA — company name, health score, health

status, key score inputs (e.g. logins in last 30 days, open tickets,

NPS, days to renewal, ARR/contract value)]




Constraints:

- Base the "likely driver" only on the data provided — don't invent

reasons not supported by the fields given.

- Keep each outreach line to 2-3 sentences, in a warm, non-scripted

tone — no "just checking in."

- Flag any account where the data is too thin to say anything

specific, rather than guessing.

- Don't recommend discounts, refunds, or commercial concessions —

flag those accounts for a human decision instead.




Output format: A table (Account | Health status | Likely driver |

Urgency rank | Suggested opening line), followed by a short list of

any accounts flagged as "needs more data."

Why this prompt works, and how to adapt it

It separates diagnosis from action, which is where most CS teams get stuck. Health scores are good at flagging that something's wrong; they're weak at explaining what. Asking Breeze to reason across the individual score inputs, rather than just report the final number, gives the CSM a starting hypothesis instead of a blank page.

The ranking does the prioritisation nobody has time for. With more than a handful of Amber or Red accounts, most teams work through them roughly in the order they appear on a list. Weighting urgency by both risk and commercial value means the highest-value slipping accounts don't sit behind smaller ones just because of list order.

The outreach draft is deliberately a first line, not a finished email. Assume it will be edited before sending — treat it as knocking down the blank-page problem, not as the final message.

Adapting it for your portal

Only using free-text engagement notes rather than a formal health score? Swap the Context section for exported notes and last-contact dates instead, and ask Breeze to infer a rough risk level from tone and recency rather than a numeric score.

Managing renewals across multiple products or tiers? Add a line to Constraints: "Group accounts by product line before ranking, since urgency drivers differ between [PRODUCT A] and [PRODUCT B]."

Want this to run monthly rather than as a one-off? Ask Breeze to also flag which accounts moved status since the last export, so the CSM can see direction of travel, not just a snapshot.

Beyond the prompt

Don't send the drafted line as-is — it's a starting point built from a handful of properties, not the full relationship history the CSM already holds in their head. Use it to skip the blank page, then edit in the specifics only a human on the account would know. Log the outreach against the record once sent, so the next health score read has that action reflected in the account's activity.

Once this becomes routine, the real win isn't the drafted email — it's that Amber accounts get a human response within days, not weeks.

FAQs

Does this work if we're on Service Hub Starter? (DRAFT) No. Health Score Builder and the Customer Success Workspace require Service Hub Professional or Enterprise, so this prompt assumes you already have a health score set up. If you're on Starter, you'd need to substitute manually tracked engagement data in the Context section instead.

Is this better than just looking at the health score dashboard? (DRAFT) It's not a replacement — the dashboard is still where the score lives. This prompt adds the explanation and the drafted first move that the dashboard itself doesn't generate, so the two are meant to be used together.

Will Breeze's "likely driver" always be accurate? (DRAFT) Only as accurate as the data you give it. The constraint against inventing unsupported reasons is there deliberately — if the export doesn't capture the real cause of a drop, the output will say the data is too thin rather than guess, so always sense-check the diagnosis against what you already know about the account.

Can I use this to automate outreach without a human sending it? (DRAFT) You could wire the output into a workflow, but we wouldn't recommend sending AI-drafted retention messages unedited — these are relationship-critical moments, and the prompt is built to produce a starting draft for a human to personalise, not a finished send.

What HubSpot data do I need to export for this to work well? (DRAFT) At minimum: health score, health status, and the individual inputs behind it (logins, open tickets, NPS, days to renewal). The more of the underlying signals you include rather than just the final score, the more specific the "likely driver" column will be