WorkerKitAI workers for inbound sales
Set up AI workers to qualify inbound leads and prepare deal follow-ups, with draft review, scoped CRM access and a clear handoff between jobs.
Published by WorkerKit. Updated .
Give inbound sales two focused AI workers: one turns a new inquiry into a CRM record and a first-reply draft; the other prepares follow-ups for owned deals that have gone quiet. Your sales rep reviews the messages and decides when an inquiry becomes a deal.
This guide is for a team that already uses a connected email account and CRM. WorkerKit accounts open by invitation. You can browse the kits now; request access before connecting accounts or running workers.
The workers
Start with the intake job. Add the follow-up job when you have owned deals to review. Each worker kit creates a separate worker with its own permissions and run record.
| Worker | Its one job | Apps | When it runs |
|---|---|---|---|
| Inbound Lead Qualifier | Read an inquiry, qualify it against your rubric, create or update the contact and save a reply draft | Email and CRM | On demand, or from its configured inbound webhook |
| Deal Follow-up Assistant | Find owned deals with no activity for seven days and prepare a grounded follow-up draft | Email and CRM | On demand when you request a pipeline review |
These kits do not install as an automatically connected group. The shared CRM holds the business records, and the sales rep owns the handoff. Installing both does not make the qualifier start the follow-up worker.
Neither kit sends email. Their email permissions allow reads and drafts, with no send operation. The qualifier can write CRM records; the follow-up worker reads the CRM without changing it.
From a new inquiry to an owned deal
A prospect emails asking for a demonstration and names the problem they need to solve. The qualifier reads that inquiry and the earlier thread, looks for the matching CRM contact, and applies your ideal customer profile and scoring rubric. It records the grade with reasons tied to the email and CRM, then saves a first-reply draft.
The rep checks the match, qualification and draft. A possible duplicate or an ambiguous inquiry appears under Needs your call in the result. The rep resolves it, edits the reply if needed and sends from the email app. If the conversation merits a deal, the rep creates or updates that deal and assigns its owner in the CRM. Creating a contact is not the same as opening a deal.
Later, the rep requests a run of Deal Follow-up Assistant. The worker checks owned deals that have had no activity for seven days and reads the latest related email. A buyer who explicitly asked to reconnect next month should be skipped, with a reason. A deal that needs a nudge gets a draft tied to the actual last exchange. The rep decides whether to send it.
The workers remain available between those runs. That is the useful unit of an AI workforce: a job with a retained method and access, even when the next request is days away. Neither worker needs to be running continuously for this process to work.
How to set up AI workers for inbound sales
Before you start, choose the person responsible for the qualification rubric, the connected email account, the CRM account and the results. Have a small set of permitted inquiries and owned deals ready for the first review.
1. Define what a finished intake job looks like
Write an ideal customer profile that the sales team can apply consistently. State which company characteristics establish fit, what counts as buying intent, and which inquiries need a person to decide. Avoid treating a confident grade as evidence that the qualification is correct.
Use these acceptance checks for the first runs: the correct contact was matched without a duplicate; the grade cites the inquiry or CRM; unsupported facts were left blank; the reply answers the actual question; the email remains a draft. Keep a separate check for the follow-up job: a future commitment in the thread must not be mistaken for a stalled deal.
2. Install Inbound Lead Qualifier
Open Inbound Lead Qualifier, choose Get this kit and supply the required Ideal customer profile. Read the included Lead scoring rubric as well as the instructions. Its A/B/C grades and tie-breakers determine the assessment; supplying a company description alone does not review that policy for you.
Choose the connected email and CRM accounts the rep is authorized to use. Review the requested operations before finishing installation. Email needs reads and draft creation. CRM needs lookup, contact creation or update, and notes. The kit's CRM scope is owned records. Do not widen access simply because a record you expected is absent.
After installation, open the worker in Workers and confirm its instructions, accounts and permissions. Confirm that the ideal customer profile is the one you intended to use and that email send has not been granted.
3. Run and inspect one intake job
Set the worker's spending limits and choose the starting model before its first run. Start on demand with a permitted inquiry whose expected outcome you know. If using an agent to start it, approve the run and its budget first.
Read the result and run receipt, then check the actual CRM record and email draft. Verify the acceptance checks from step one, including every CRM create, update and note. A successful run status means the execution finished; it does not prove that the right contact was chosen or the message is ready to send.
The instructions use a CRM note to recognize inquiries already handled. Review that behavior before retrying a partly completed run. A failed run may already have written a contact or saved a draft, so inspect those results before starting again.
4. Choose the intake starting point and result delivery
Keep intake on demand until you have a supported event setup for your lead source and account. The qualifier's catalog lists inbound webhook and on-demand trigger modes; installing it does not itself make every incoming email start a run. Its published configuration includes no recurring schedule.
When your source is connected to a supported webhook trigger, use that setup's current authentication requirements and payload format. Send a permitted inquiry through the source and verify that it reaches the intended worker and produces the expected record and draft. If you do not have that event path, assign the rep to request an intake run when new inquiries need processing.
Assign someone to reconcile incoming inquiries with completed results so a missed event is found. Also choose how the responsible rep receives the result through the delivery options available to the account; check that the result actually arrives where the rep works.
5. Install Deal Follow-up Assistant
Open Deal Follow-up Assistant and choose Get this kit. Select the rep's email and CRM accounts, then confirm that CRM is read-only and email permits drafts without sending. This kit has no required custom input and no standing schedule in its published configuration.
Read its Follow-up script and instructions. The job selects owned deals with no activity in the last seven days, then checks the email context for a valid reason to wait. Run it on demand when the rep wants a pipeline review. Confirm the draft points to the actual last exchange and makes no new pricing, discount or timing promise.
6. Put the handoff and review on the team's calendar
Record who reviews new-lead drafts, who opens and owns deals, and who requests the follow-up sweep. Choose a team review cadence, such as a weekly pipeline meeting. That is a human operating routine; the follow-up worker still runs on demand when requested.
Keep the qualifier responsible for intake and the follow-up worker responsible for owned deals. When results are weak, identify whether the issue is source data, record matching, the rubric or the writing before changing instructions. The safe rollout walkthrough covers staging access and inspecting writes in more detail.
What each worker can touch
The difference between the two CRM grants matters more than the fact that both workers use the same app.
| Worker | Reads | Writes | Boundary to check |
|---|---|---|---|
| Inbound Lead Qualifier | Inquiry and earlier email thread; permitted CRM records and associations | Email drafts; CRM creates, updates and notes | Owned-record CRM scope; no email send; no CRM delete or merge operations |
| Deal Follow-up Assistant | Related email threads; owned CRM deals and their history | Email drafts | CRM read-only; no email send; results cover only the records it can see |
Review grants through the app firewall, and restrict fields where supported by the built-in app. A missing record can mean limited visibility rather than no existing relationship. The rep should resolve that ambiguity before authorizing broader access or creating another record.
Both kits keep their published working state in the connected apps rather than requiring a long-term worker-memory setup. The qualifier's CRM notes and the deal's activity history supply context for later runs. Do not assume installing the kits creates shared company knowledge or automatic learning.
What it costs
Measure the first accepted runs before estimating a month. The catalog does not give a fixed per-lead price. A new contact, an existing thread and an ambiguous match can require different amounts of work.
| Part of the process | Unit to record | How to budget it |
|---|---|---|
| Lead qualification | Actual tool calls and model cost per intake run | Add the observed usage for the number of inquiries processed, including retries |
| Deal follow-up | Actual tool calls and model cost per requested sweep | Multiply by the number of sweeps you plan, allowing for the number of deals inspected |
| Human review | Minutes checking matches, editing drafts and correcting records | Include the team's labour cost in the evaluation |
| Ongoing operation | Plan charges, applicable app charges and maintenance time | Allocate them to the jobs being evaluated without counting the same expense twice |
For a planning day with 20 separately triggered inquiries and one requested pipeline sweep, the starting run count is 21, before retries. It is not 21 tool calls: each run can call the apps several times. Multiply your observed intake calls per run by 20, then add the sweep's calls. These are planning inputs, not a measured workload or a throughput promise.
Compare total usage with the current daily and rate limits on /pricing. Model tokens are billed separately at provider list price with no markup; the wallet or your own model keys determine how they are paid. Keep spending limits per run, per day and for the account aligned with the volume you have authorized.
The evaluation measure is cost per acceptable AI result. For intake, an acceptable result includes the correct CRM handling and a usable draft. Fast generation that leaves the rep correcting every record can make the process uneconomical even when the final result is acceptable.
Where to keep a human in the loop
The sales rep owns sending messages, resolving ambiguous matches and deciding whether a lead becomes a deal. Review low-fit grades as well as high-fit ones: the rubric can otherwise hide inquiries the company should answer.
Drafting is a concrete boundary because these kits lack email-send operations. CRM writes by the qualifier are different: they happen during the authorized run. This setup does not place every CRM change behind a separate approval policy. If your team requires approval before any CRM write, narrow the grant and redesign that part of the job before enabling intake.
Pause the affected worker when its results need investigation. Check what it already changed and correct those records separately; stopping future work does not undo a completed write. At low lead volume, doing the intake manually may take less effort than maintaining the rubric and reviewing automated results.
FAQ
Do these workers send customer emails automatically?
No. Inbound Lead Qualifier and Deal Follow-up Assistant create drafts, and their published email permissions contain no send operation. The sales rep reviews and sends from the email app.
Does the lead qualifier automatically hand work to the follow-up worker?
No. The rep reviews the intake result and creates or updates the owned deal in the CRM. Deal Follow-up Assistant runs on demand when requested and reads that deal context. Installing the two kits does not configure automatic dispatch between them.
Can we use the qualifier without a webhook?
Yes. Inbound Lead Qualifier supports on-demand runs. Keep it on demand when your lead source does not deliver a supported event, and explicitly assign someone to initiate and review the work.
Can an AI agent set up the workers for us?
An authorized agent can use the WorkerKit MCP server to preview and install the kits after you have account access. For Claude, follow the Claude setup guide; use the corresponding assistant page for another client. The agent needs the installKits scope, calls kit_install_preview, shows you the proposed setup, and calls kit_install only after your approval. App connections, enabling automation and starting paid runs also need your authorization.
What should we add after these two jobs work?
Add a worker only when a specific handoff or backlog warrants it. The next job might be keeping meeting notes current in the CRM, but it needs its own owner, access and acceptance checks. Moving from AI tools to an AI workforce explains how to expand without losing those boundaries.