# How we turn the pipeline into a forecast every week Each week our agent reads HubSpot's open deals, weights them into a forecast against quota by stage, rep and segment, names the deals slipping out of the quarter, and posts it to Slack. It only reads. Canonical page: https://themachine.myapps.ai/use-cases/sales-forecast Each quarter your sales team is chasing a target, and most teams check progress the same way. Somebody exports the open deals from HubSpot, weights them by instinct or by a spreadsheet formula whose author left long ago, and calls the result a forecast. It holds up for roughly a day, the day it was made. It will not tell you about the deal whose close date drifted past last week, or the six-figure opportunity with no logged activity for a month. Our answer is an agent on Machine that reads HubSpot once a week and posts a weighted forecast in our sales Slack channel. It splits the number by stage, by rep and by segment, measures it against quota, and names the deals that put it at risk. It only reads the pipeline, and the Slack post is the one thing it produces. We forecast our own quarter this way. - **Team:** Machine - **CRM:** HubSpot - **Runs on:** A weekly schedule - **Mode:** Reads only ยท one Slack post a week ## The problem Seeing your pipeline stage by stage is not the same as having a forecast. The pipeline will tell you a deal is open and parked in "Proposal Sent". It will not tell you whether it is likely to close this quarter, whether the rep's commit squares with what that stage usually delivers, or whether a big deal has gone quiet while still padding the total. Getting from open deals to a forecast means weighting each deal by how often its stage really converts, adding those up three ways (by stage, by rep, by segment) and setting the result against quota. The usual approach is a spreadsheet rebuilt the week before the forecast meeting. Someone exports the deals, applies a weighting rule by hand, and checks it against what each rep says they will close. It is out of date as soon as it is finished, it only exists if someone remembers to make it, and it seldom spots a deal past its close date, or one stuck late in the pipeline, until the quarter is nearly gone. ## What we built With Machine, a weekly cron starts an agent. Each run gets a fresh **session**, a clean cloud computer (a sandbox) for that week only, with read-only access to HubSpot. The agent collects every open deal with its stage, amount, close date, owner and segment. It weights each deal using the probability already set on its HubSpot pipeline stage, then totals the weighted amounts by stage, by rep and by segment. It sets the total against this quarter's quota. It flags deals whose close date has already passed, or that have too little time left in their current stage to plausibly close this quarter. It also flags large deals that look shaky for other reasons, such as no recent activity or being stuck in a late stage. Then it posts the full forecast to Slack, and writes nothing back to HubSpot. ## How it works ### Run it once a week A **cron trigger** starts the agent every week. Each run opens a new **session** in a sandbox of its own: one week, one run, no leftovers. Because the forecast is rebuilt from what HubSpot holds right now, it can never wander away from the CRM. ### Teach it how to add things up The rules live in a **skill**, the playbook the agent brings to every run: how deals are weighted, which ones belong to this quarter, and what "slipping" and "at risk" mean. Every deal is weighted by the probability HubSpot already has on its stage, never by a fixed guess. A deal belongs to the quarter according to its close date. A deal counts as slipping once its close date is behind us, or when its stage leaves too little time to close before the quarter ends. ### Give it a look-only link to HubSpot With a narrow **connector**, whose keys stay on our servers so the model never sees a raw token, the agent reads: - **Open deals.** Amount, stage, close date, owner, and the segment field that drives the breakdown. - **How the stages are set up.** It uses the win probability HubSpot already assigns to each stage as the weighting, rather than assuming one. - **Slack, for posting.** The weekly forecast goes there, split by stage, rep and segment, with slipping and at-risk deals named. ### Set the limits Across HubSpot, the agent is **read-only**. It cannot write to a deal's amount, close date, stage or owner, so whatever the numbers show, it has no way to change the figure attached to any deal. A Slack post is all it produces. Credentials are stored encrypted in the **Secrets Manager** and only handed to the agents you grant them to. ### Share the forecast Every week there is a single Slack post. It shows the total weighted forecast next to quota, the split by stage, by rep and by segment, the deals that have slipped or are running short of time this quarter, and the large deals with other warning signs. The sales team reads it and chooses what to do next. HubSpot is never updated automatically. > **The pattern** > A weekly **cron** opens a session with a look-only **connector** to HubSpot. > The weighting and totalling rules sit in a **skill**. The agent reads the > entire pipeline and produces one thing: the weekly Slack post. ## Guardrails Reading the whole pipeline every week only works if people trust it, so the agent's access is narrow and runs in one direction: - **Isolation.** Each run happens in a sandbox of its own. The session can reach only HubSpot and Slack, and the forecast post is the one thing it sends out. - **Scoped secrets.** The HubSpot key is stored encrypted in the Secrets Manager and only handed to the agents you grant it to. - **Read-only, no exceptions.** The HubSpot connector can only read. The agent never alters a deal's amount, close date, stage or owner. All it can do is report what it sees. - **Everything is code.** The weighting rule, the quarter window and the definitions of slipping and at risk are files in the repo. They keep their history, and any change goes through a reviewed **change request** instead of a dashboard toggle. ## The outcome - **Every week:** Forecast rebuilt from what HubSpot holds right now - **Read-only:** No deal amount, date, stage or owner is ever changed - **3 cuts:** Stage, rep and segment totals in a single weekly post The forecast someone used to rebuild by hand before each call now turns up every Monday in the channel the team already lives in. It is weighted by HubSpot's own stage probabilities and names the exact deals that threaten the number. The agent reads and reports, nothing more. Every deal still belongs to the sales team.