Salesloft Salesforce Integration: Flag Stalled Opportunities
Connect Salesloft activity to Salesforce pipeline reviews, then catch open opportunities whose cadence has gone quiet before the next forecast meeting.

TL;DR: what you'll build. A Sales Manager owns pipeline hygiene, but a quiet Salesloft cadence on an open Salesforce Opportunity is easy to miss until the deal is already cold. This guide builds a Major app and scheduled workflow that cross-references Salesloft Cadences, People and activity against open Salesforce Opportunities, flags any deal with no logged email or call touch for seven or more days, and routes an agent's recommendation to the rep and manager for approval before anything changes in the CRM.
Why a connected cadence still needs pipeline monitoring
Salesloft and Salesforce already talk to each other. The Salesloft Connect for Salesforce extension lets reps import Leads, Contacts and Accounts, log calls and emails, and work a cadence side by side with the CRM record.
That connection answers a different question than the one a Sales Manager actually asks every Monday: which open deals have gone quiet. A synced activity log tells you what happened. It does not tell you what stopped happening.
This is the gap this build closes. It belongs to the sales function the same way the AI agents for sales hub frames it: agents reserved for judgment, with the repeatable detection work running as deterministic code. The ZoomInfo-Outreach integration article covers the front end of this same pipeline, enrollment and first-touch personalization. This article picks up further downstream, once a deal is already open and a rep is supposed to be working it.
A quiet cadence is not proof a deal is dead. It is a visibility gap. The Salesloft Salesforce FAQ covers account linking, deletion and org migration edge cases, but it does not describe a durable way to watch for cadence silence over time. That is the part Major builds.
If you are weighing how much of your Salesforce side should run through agents versus plain automation, the Salesforce AI agents article covers that governance question directly.
The systems and records behind the check
This workflow reads from two systems and writes almost nothing back automatically. Here is what each one holds and how the Major side talks to it.
Salesloft
Salesloft's platform API is a RESTful API authorized through OAuth 2.0.
Customer integrations typically use a Bearer API key created in the Salesloft account, while partner-built integrations go through a registered OAuth app.
The objects this build reads are:
- People: the Salesloft record that links a contact to Salesforce, used to confirm the matching key between systems.
- Cadences: the active sequence a Person is enrolled in, which tells you whether outreach is supposed to be happening at all.
- Emails and Calls: the logged activity records with timestamps, which are the actual evidence of a touch.
Salesloft also exposes a Webhook Subscriptions API so you can get event-driven notifications instead of polling for every new activity, with delivery headers documented separately and a defined event types list.
One gotcha worth knowing before you build: rate limits are shared across the whole team, not per integration. Salesloft's documented model allocates 600 cost points per minute across the account, so a heavy daily scan needs to be mindful of other integrations drawing from the same budget.
Salesforce
Salesforce's REST API is accessed through OAuth 2.0, using either a Connected App or, as of Spring '26, the newer external client apps model (Salesforce now recommends external client apps for new integrations; existing Connected Apps keep working).
The object this build reads is the Opportunity, specifically open Opportunities with their stage, owner and last activity fields.
Salesforce reports API usage through the Sforce-Limit-Info response header, and the daily allocation varies by org edition. Full REST, SOAP and Bulk API access requires an Enterprise, Unlimited, Performance or Developer edition with API access enabled on the org, so confirm that before you plan a daily poll against a smaller edition.
Salesforce does not have a generic webhook system the way Salesloft does. Event-driven options are Platform Events, PushTopics or Change Data Capture, all of which require more setup than a standard webhook. For this build, a scheduled poll of open Opportunities is simpler and sufficient, since the check only needs to run once a day.
One honest caveat from the vendor docs: Salesloft activity can sync into Salesforce as Task records with a configurable Task Subtype, but the setup guides do not establish a built-in feature for monitoring cadence-touch health over time. That monitoring is the custom workflow this article describes, not a native toggle in either product.

Cadence Stall Watch: how the workflow, app and agent split the work
Split the work across app, agent and workflow
The instinct with a problem like this is to hand it all to an agent: have it check Salesforce every morning, reason about every deal, and decide what to do. That is slow, expensive, and the model re-derives the same filter logic every single run.
A webhook is a doorbell: the tool rings you when something changes, so you don't keep checking the door. The same principle applies here even without a native webhook. Instead of an agent polling and reasoning from scratch each day, a scheduled workflow does the deterministic filtering, and the agent is called only once there is something ambiguous to judge.
Here is how the three pieces split:
- The workflow runs on a schedule, pulls open Opportunities from Salesforce, pulls matching Salesloft People and Cadence activity, computes the days-since-last-touch for each linked record, and upserts one row per stalled deal into the app's table. This is pure deterministic code: same inputs, same output, every time.
- The app is the manager's control surface and the durable state. It holds the stalled-deal table, the touch history, the review status, and the audit log of every past flag and outcome. Nothing here depends on the model running again to remember what happened last week.
- The agent is called only for the records the workflow flags. It reads the deal's stage, owner, and touch history, and drafts a rationale for one of three recommendations: rep follow-up, manager escalation, or marking the opportunity as cooling. It never writes to Salesforce directly and never sends a message to a prospect.
This is the predictable-agent pattern in practice: reason once per flagged deal, not once per record scanned. The model is reserved for the judgment call a human would otherwise have to make cold, and the repeatable scan runs as software the organization can inspect.
How to build this on Major
Here is the build sequence, in order.
- Connect Salesforce through its official OAuth 2.0 flow, using an external client app (or an existing Connected App) scoped to read Opportunity records: stage, owner, close date, and last activity date. You should see a successful token exchange and a test query returning open Opportunities in the Major connector test panel.
- Connect Salesloft as a custom API connector, since Major does not list an official Salesloft connector today. Store the Bearer API key in Major's credential proxy, never in code, and scope the key's use to the People, Cadences, Emails and Calls endpoints. You should see a successful test call returning a People record from your Salesloft instance.
- Build the app's data model with a table called stalledopportunities, holding: the Salesforce Opportunity ID, the matched Salesloft Person ID, deal stage, owner, last touch type and timestamp, days since last touch, the agent's recommendation and rationale, review status, and reviewer. Add a second table, reviewlog, that records every approval or dismissal with a timestamp and reviewer name for audit purposes.
- Add the agent and give it a scoped prompt that only drafts a recommendation, never acts on it. A prompt like this works well as a starting point:
"Given this Opportunity's stage, owner, days since last Salesloft touch, and the last three logged activities, recommend one of: rep follow-up, manager escalation, or mark as cooling. Write a two-sentence rationale citing the specific gap. Do not draft outreach copy and do not change any field. Flag for manager escalation only when the gap exceeds 14 days or the deal is in a late stage."
- Build the app's queue view showing one row per stalled deal, sorted by days since last touch, with the agent's recommendation and rationale visible and a one-click approve or dismiss action for the rep or manager. Example prompt for building this view:
"Build a table view of stalledopportunities filtered to reviewstatus = pending, sorted by dayssincetouch descending, with inline approve and dismiss buttons that write to review_log."
- Set the workflow trigger to run on a weekday morning schedule, for example 6am in the sales org's primary time zone, using the opportunity ID plus the scan date as the idempotency key so a retry or a second run the same day never creates duplicate queue rows. You should see one row created or updated per stalled deal, not a growing pile of duplicates.
- Test the workflow in a sandbox using a handful of real but non-production Opportunities, including one Opportunity with no matched Salesloft Person at all, to confirm the workflow correctly skips unlinked records rather than treating a missing link as a false stall. You should see the unlinked record logged as skipped, not flagged.
- Roll out with role-based permissions: reps see and act only on their own flagged deals, managers see the full queue and the escalation tier, and only a manager role can approve a CRM field update if the org later decides to add a write-back step. You should see a rep's view correctly scoped to their own Opportunities only.

No CRM write or outreach happens without a human approval
Here is one illustrative record this queue would show, labeled as mock data: Opportunity "Northbridge Logistics - Platform Renewal," stage Negotiation, owner Dana Reyes, last touch an Email logged 11 days ago, agent recommendation "escalate to manager," rationale "late-stage deal with an 11-day gap exceeding the 7-day threshold, no call logged since the last email," review status pending.
Governance and edge cases
Treat this like any other system that reads CRM and sales-engagement data, not like a side script a rep built on their own laptop.
- Least privilege. The Salesforce connection reads Opportunity fields only, with no object create or delete permission, and the Salesloft key is scoped to the four read endpoints this workflow uses, nothing broader.
- SSO and role-based access. Reps and managers sign into the app through the org's existing SSO, and the queue view is filtered by role so a rep never sees another rep's full pipeline review.
- Audit log. Every flag, every agent recommendation, and every approve or dismiss action is written to the review_log table with a timestamp and the acting user, so a manager can reconstruct exactly why a deal was or was not escalated.
- Scoped secrets. Both the Salesforce OAuth token and the Salesloft API key live in Major's credential proxy, never embedded in workflow code or agent prompts.
- Human approval gates. No message goes to a prospect and no Opportunity field changes without an explicit rep or manager approval in the app. The agent's output is a draft recommendation, not an action.
Failure modes to plan for:
- API errors and rate limits. If Salesforce returns a limit warning in
Sforce-Limit-Info, or Salesloft's shared 600-points-per-minute budget is exhausted, the workflow should back off and retry on the next scheduled run rather than hammering the API mid-scan. - Partial syncs. If the Salesforce pull succeeds but the Salesloft pull fails partway through, the workflow should skip updating affected rows rather than flag a deal as stalled based on incomplete data.
- Duplicate events. The opportunity-ID-plus-date idempotency key prevents a retried run from creating a second queue row for the same deal on the same day.
- Stale or missing links. An Opportunity with no matched Salesloft Person is not a stalled cadence, it is an unlinked record, and the workflow should log it separately rather than flag it the same way.
- What the agent must never do. It must never update the Opportunity stage, never send an email or call note on a rep's behalf, and never mark a deal "cooling" as a final status without a human confirming it. Any ambiguity routes back to a person.
What to measure
Track a handful of numbers the sales org already watches, rather than inventing new ones:
- Count of Opportunities flagged as stalled per week, by stage.
- Median days from flag to rep action (approve, re-engage, or dismiss).
- Share of flagged deals where the rep's follow-up produced a new logged touch within 48 hours.
- Count of deals escalated to a manager versus resolved at the rep level.
- False-flag rate: deals flagged as stalled that turn out to have an unlinked or mismatched Salesloft record.
None of these numbers exist yet in your org until you run this for a few weeks. Don't promise a specific lift before you have your own baseline.
Before you go live
- [ ] Salesforce OAuth connection scoped to Opportunity read only, tested against a sandbox org
- [ ] Salesloft API key scoped to People, Cadences, Emails and Calls, stored in the credential proxy
- [ ] Idempotency key (Opportunity ID plus scan date) confirmed to prevent duplicate queue rows
- [ ] Unlinked-record handling tested with at least one Opportunity with no matched Salesloft Person
- [ ] Role-based views confirmed: reps see only their own deals, managers see the full queue
- [ ] Agent prompt confirmed to draft recommendations only, with no CRM write or outreach permission
- [ ] Audit log (review_log) capturing every flag, recommendation and human decision
- [ ] Owner named for weekly review of the false-flag rate and threshold tuning
More sales AI automation use cases
- AI agents for sales: every use case by tool
- Automate lead-to-sequence handoff between ZoomInfo enrichment and Outreach cadences
Start building on Major
Major is the enterprise platform where agents build the software they run on, so the daily Salesloft-to-Salesforce stall check runs as a governed app instead of a fresh reasoning pass every morning, with the escalation agent reserved for the judgment call on each flagged deal.
Related articles
- Automate Shopify-ShipStation Delay Alerts with AI
- Datadog PagerDuty Integration: Automate Incident Triage
- Automate Ramp-to-NetSuite Expense Coding with AI Review
About the author
Ritvij Gautam is Head of Growth at Major.
Last reviewed November 8, 2026 by Ritvij Gautam.
Frequently asked questions
- How does the Salesloft Salesforce integration sync activity?
- The official Salesloft Connect for Salesforce extension lets reps log Emails and Calls that can sync into Salesforce as Task records, with a configurable Task Subtype and sync logs for troubleshooting. That native sync tells you what happened, but it does not include a built-in check for cadences that have gone quiet. The stalled-opportunity scan described here is a separate, purpose-built workflow layered on top of that sync, not a default feature of either product.
- How can a sales manager find Salesforce opportunities with no recent Salesloft activity?
- Match each open Salesforce Opportunity to its linked Salesloft Person and active Cadence, then compute the days since the last logged Email or Call for that Person. A scheduled workflow can filter for any gap past a chosen threshold, such as seven days, and write the result to a manager-facing queue. Always validate the Person-to-Opportunity link first, since an unlinked record is a data gap, not a stalled deal.
- What permissions are needed to connect Salesloft and Salesforce?
- Salesloft customer integrations typically use a Bearer API key created in the account's developer settings, scoped to the specific endpoints the integration needs. Salesforce requires an OAuth 2.0 connection through a Connected App or the newer external client app model, with API access enabled on the org and a Salesforce edition that supports REST API access. An admin should grant least-privilege, read-only scopes for this use case rather than full CRM access.
- Can an AI agent automatically update a stalled Salesforce Opportunity?
- The recommended approach keeps the agent out of the write path entirely. The agent reads the deal's history and drafts a rationale and recommendation, such as rep follow-up or manager escalation, and a deterministic app logs that draft to a review queue. A rep or manager must approve any resulting action before the Opportunity record changes or any message goes to a prospect.