Automate Shopper Onboarding with Zapier AI and Make

Title each step between a signed contract and the second a brand new consumer can truly do one thing helpful, and do it with out opening your onboarding guidelines. Strive it out loud. The listing runs longer than you anticipate, and each merchandise on it’s anyone’s afternoon: the shared folder, the kickoff transient, the CRM document, the welcome e-mail, the bill that no person needs to be the one to ship.

You’re going to construct that listing as soon as as software program, then construct it a second time on a special platform. The Zapier model catches an consumption webhook and converts the messy a part of the shape into validated JSON. From there it creates the consumer’s folder and kickoff transient, writes the CRM document, and stops at a human approval step earlier than something contractual leaves the constructing. You then rebuild the identical run as a Make situation with an actual error route, and value each in opposition to month-to-month consumer quantity.

What One Onboarding Run Accommodates

Begin with the unit every platform payments you for, as a result of the 2 are usually not comparable by title. In Zapier a workflow is a Zap, and also you pay per job. Zapier counts a job for each profitable motion step, whereas triggers, Filter steps and Paths steps price nothing. In Make a workflow is a situation on a visible canvas, and also you pay per credit score, the unit Make beforehand referred to as an operation. For the non-AI modules on this construct, one credit score covers one module run in opposition to one bundle of information.

One onboarding run right here means one consumption submission goes in, and one consumer comes out with a folder, a kickoff transient, a CRM document, and a welcome e-mail that references all three. Something contractual waits for an individual to say sure.

Stipulations

To construct alongside, you want:

• A Zapier Skilled plan or increased, as a result of webhooks, Paths, and the Human within the Loop approval step are all gated above Free and this construct makes use of all three.

• A Make account on any tier. The Free plan’s 1,000 credit per thirty days cowl testing, and the 15-minute scheduling ground by no means applies as a result of the set off right here is an immediate webhook.

• An OpenAI API key, wanted just for the Make construct. Zapier’s built-in AI step runs with out one.

• A Google account with Drive and Docs linked on each platforms, plus a kickoff-brief doc that makes use of {{merge_field}} placeholders.

• A CRM you possibly can write to. HubSpot, Pipedrive, or Airtable all work, so long as the CRM exposes a create-record motion on each platforms.

• An consumption kind that may POST JSON. Skip this in case your kind app has a local set off on each platforms; the uncooked webhook is right here so you possibly can learn the payload with your individual eyes earlier than you map a single discipline.

The Consumption Payload That Begins All the pieces

Each consumer onboarding automation begins with a consumer consumption kind, and the form of that submission decides how a lot work the remainder of the workflow has to do. Typeform and Google Varieties each put up JSON, and so does a kind you construct inside Zapier Varieties. Right here is the payload this construct expects. Put it aside as intake-sample.json within the listing you run curl from; the webhook take a look at beneath and the staging guidelines on the finish each put up this precise file:

{
  "submitted_at": "2026-03-04T14:22:08Z",
  "form_id": "client-intake-v3",
  "contact": { "title": "Dana Ruiz", "e-mail": "[email protected]", "position": "COO" },
  "firm": { "title": "Northgate Logistics", "area": "northgate.io", "dimension": "11-50" },
  "engagement": {
    "sort": "retainer",
    "start_date": "2026-03-17",
    "monthly_value_usd": 6500,
    "scope_notes": "want the weekly service report automated, plus one thing for the driving force check-in texts, ideally earlier than our Q2 board assembly"
  }
}

Solely scope_notes is unstructured. All the pieces else is already typed by the shape itself. The AI step will get the free-text discipline and nothing extra, so it by no means has the chance to reformat an e-mail handle or spherical a greenback quantity. That boundary holds via each builds.

Construct the Onboarding Zap

The Zapier construct is linear by design, which is its benefit right here. Eight steps run in a hard and fast order from webhook to welcome e-mail, with a single filter that stops unhealthy information earlier than it reaches something the consumer can see.

Catch the Consumption Webhook

Add Webhooks by Zapier because the set off and select the Catch Hook occasion. Zapier points a URL; paste it into your kind’s webhook setting, then push an actual pattern via it out of your terminal:

curl -X POST https://hooks.zapier.com/hooks/catch/000000/abcdef/ 
  -H "Content material-Sort: utility/json" 
  -d @intake-sample.json

Zapier flattens nested JSON on the way in which in, so engagement.scope_notes turns into a mappable discipline named engagement__scope_notes. Use Catch Hook quite than Catch Uncooked Hook except you propose to parse the physique your self. The set off prices no duties, so you possibly can re-test it as usually as you want.

Flip the Scope Notes Into Structured JSON

Add an motion step, seek for AI by Zapier, and choose it. There is no such thing as a motion occasion to decide on right here; the step opens straight into its Configure panel. Map engagement__scope_notes as the one enter, then write a immediate that tells the mannequin to drag the onboarding particulars out of that discipline and invent nothing.

The immediate alone is not going to provide you with something to map. With no output fields outlined, AI by Zapier returns one mixed end result, which the Filter within the subsequent step can’t take a look at and the welcome e-mail can’t reference. Open Settings, broaden Output Fields, and click on + Add discipline as soon as for every of the 5 beneath, giving each one a reputation and a discipline sort. That is the form these fields describe, not textual content you paste into the immediate:

{
  "project_title": "string, 60 characters or fewer",
  "deliverables": "one string, deliverables separated by commas",
  "first_milestone_date": "YYYY-MM-DD",
  "risk_flags": "one string, flags separated by commas",
  "welcome_paragraph": "two sentences, no pricing, no dates"
}

Examine Is that this output discipline required? on project_title and first_milestone_date. These are the 2 the Filter gates on within the subsequent step, and marking them required is what stops the mannequin from quietly dropping a key. Put the constraints in every discipline’s description: welcome_paragraph forbids pricing and dates, so an invented quantity can by no means attain a consumer inbox.

Two of these keys maintain a couple of worth, and that’s the place builders get caught. Zapier paperwork the Subject sort management however publishes nothing in regards to the choices inside it, so don’t design the step round a sort you haven’t seen in your individual dropdown. Describe deliverables and risk_flags of their discipline descriptions as a single string with the values separated by commas, and pin the chance vocabulary there too: scope_creep, tight_deadline, unclear_owner, nothing else. Something downstream reads that string with the (Textual content) Accommodates rule, which Zapier paperwork for each Filter and Paths steps and which isn’t case-sensitive. A rule studying risk_flags (Textual content) Accommodates tight_deadline fires whether or not the mannequin returned that one flag or all three.

Warning: The scope notes your consumer typed are consumer information. Earlier than you ship that discipline to any mannequin, verify which supplier processes it and what your engagement letter guarantees about third-party processors.

Mannequin tier modifications the invoice, not simply the standard. AI by Zapier costs by tier: Customary prices 1x duties, Superior 3x, Premium 5x, with the method (1 × mannequin fee) + (variety of software calls × mannequin fee). This step wants no instruments, so Superior prices three duties per run.

Gate the Run on the Schema

Add a Filter step instantly after the AI motion. Proceed solely when project_title exists and first_milestone_date just isn’t empty. Filters by no means devour duties, so this gate is free insurance coverage.

With out it, a mannequin that returns a partial object nonetheless lets the run proceed, and the failure surfaces as a kickoff transient with a clean undertaking title sitting within the consumer’s folder. You discover out after they reply asking what it’s.

Fan Out to Folder, Temporary, and CRM

Google Drive → Create Folder makes the consumer’s listing and returns a shareable hyperlink. Google Docs → Create Doc from Template copies your transient and fills the moustache placeholders from the AI output. Your CRM’s create-record motion writes the corporate, contact, and engagement worth. Every of these three actions prices one job.

The Create Doc from Template dropdown lists solely Google Docs you created your self, not the inventory templates Google ships, and trying to find a file that by no means exhibits up there prices individuals an hour. In case your transient is lacking from the listing, it’s as a result of Zapier solely exhibits paperwork you created. Make your individual copy of the file in your Drive and choose that.

Pause Earlier than Something Contractual

Add Human within the Loop → Request Approval. The Zap run stops right here and notifies a reviewer by e-mail or Slack with the AI-generated transient hooked up for evaluate.

Set Timeout to one thing shorter than your precise response time, then select intentionally between Skip and proceed and Finish run. Skip and proceed means an unanswered request nonetheless sends the welcome e-mail; Finish run means a consumer who signed on Friday hears nothing till Monday. Accredited fields arrive downstream labeled Submitted Content material {discipline title}, and something the reviewer modified arrives as a separate Edited Content material {discipline title} discipline. On a Skilled plan you possibly can solely route requests to your self, which is sufficient for testing however not for a staff.

Ship the Welcome E-mail

The ultimate Gmail step maps three issues: the AI welcome_paragraph, the Drive folder hyperlink, and the kickoff transient URL. That’s welcome e-mail personalization with no template variables left dangling, as a result of the filter already proved they exist.

Counting duties for the entire run: three for the AI step, one every for Drive, Docs, CRM, approval, and Gmail. Eight duties per consumer, because the step listing beneath exhibits.

Rebuild the Identical Workflow in Make

The Make model does the identical work with two variations that matter: you management the AI request physique instantly, and a failing module fingers its bundle to a devoted error route whereas the remainder of the run finishes. Construct it as a brand new situation on a clean canvas.

Begin With a Customized Webhook

Add Webhooks → Customized webhook, click on Add, title it, and replica the URL. Ship the identical pattern payload with curl whereas the module is listening, and Make derives the information construction from what arrives.

When your kind later positive factors a discipline, that construction doesn’t replace itself. Click on Re-determine information construction and re-send a pattern, or the brand new discipline won’t ever seem within the mapping panel. Make’s webhook documentation covers the excellence between customized webhooks and app-specific immediate triggers.

Pressure the AI Step to Return Your Schema

Make’s OpenAI modules do carry an output format selector, so this isn’t a lacking function you might be routing round. Use HTTP → Make a request anyway, as a result of it fingers you the entire response_format object quite than a dropdown, together with strict: true and additionalProperties: false, and it retains the precise payload readable. Level it on the Chat Completions endpoint with this physique:

{
  "mannequin": "gpt-4.1",
  "messages": [
    { "role": "system", "content": "Extract onboarding fields from the notes. Never invent dates or amounts." },
    { "role": "user", "content": "{{1.engagement.scope_notes}}" }
  ],
  "response_format": {
    "sort": "json_schema",
    "json_schema": {
      "title": "onboarding_brief",
      "strict": true,
      "schema": {
        "sort": "object",
        "properties": {
          "project_title": { "sort": "string" },
          "deliverables": { "sort": "array", "gadgets": { "sort": "string" } },
          "first_milestone_date": { "sort": "string" },
          "welcome_paragraph": { "sort": "string" }
        },
        "required": ["project_title", "deliverables", "first_milestone_date", "welcome_paragraph"],
        "additionalProperties": false
      }
    }
  }
}

strict: true binds the mannequin to the schema at decode time. required should listing each property, and additionalProperties: false blocks additional keys; OpenAI’s structured outputs information requires each. The mannequin worth needs to be one which helps strict schemas, so verify that information earlier than you paste a mannequin title. And a schema-valid response can nonetheless be a refusal, so learn the refusal discipline earlier than you belief content material.

The 2 builds diverge on goal at deliverables: it’s a actual array right here as a result of OpenAI’s information lists Array among the many supported structured-output sorts, whereas the Zapier model carries the identical values as comma-separated textual content.

Route the Legitimate Runs and Park the Relaxation

Add a Router after the JSON parse. The primary route filters on project_title current and carries on to Drive, Docs, CRM, and Gmail. The second route catches all the things else and writes the uncooked bundle to a Information Retailer with the parse error hooked up.

Rely modules, not visible steps, whenever you value this: each module that touches a bundle prices a credit score. The parked route prices one credit score and turns a malformed submission right into a row you possibly can open and browse.

Hand the Approval to a Second State of affairs

Make has no approval pause on any plan you’d truly purchase for this construct. It does publish a Human within the loop app, however that one is Enterprise-only and nonetheless in closed beta, so break up the workflow quite than faking a pause with a sleep. State of affairs A ends by sending the reviewer a message containing an approve hyperlink that factors at State of affairs B’s customized webhook, with the Information Retailer document ID within the question string. State of affairs B fetches that document and sends the contract and first bill.

The break up earns its preserve. The cash and authorized path will get its personal execution historical past and its personal error route, so a failed bill by no means exhibits up as a failed onboarding, as you possibly can see within the canvas beneath.

Zapier Duties Versus Make Credit

Now value the identical run on each platforms. The Zapier construct consumes eight duties per consumer. The Make construct consumes roughly twelve credit per consumer: eight modules in State of affairs A together with the approval notification, and 4 in State of affairs B.

The Invoice at 10, 50, and 200 Shoppers

Multiply these per-run figures by month-to-month quantity and the 2 pricing fashions separate instantly.

Illustrative: job and credit score counts derived from this construct’s step listing, not from vendor benchmarks.

Actuality Examine: The AI step is 37.5 p.c of the Zapier run’s job price on the Superior tier. Drop it to Customary and the identical workflow prices six duties.

The place the Price Curve Bends

Quantity strikes you up Zapier’s value ladder, as a result of each paid Zapier tier sells you a particular job depend. Quantity barely strikes you on Make, as a result of each paid Make plan begins at 10,000 credit; there you climb tiers to unlock options. Customized variables and full-text execution log search start at Professional.

Record costs on the entry tiers, learn from every vendor’s personal web page: Zapier Skilled begins at $29.99 per thirty days billed month-to-month, or $19.99 billed yearly, at 750 duties, per Zapier’s printed plan pricing. Make Core begins at $16 per thirty days billed month-to-month, or $12 billed yearly, at 10,000 credit; Professional is $28 billed month-to-month, or $21 billed yearly. Each distributors promote quantity in fastened steps, so value your precise tier earlier than committing.

Deal with the Failures That Really Occur

Whole outages announce themselves. The run that creates the folder, writes the CRM document, after which dies on the CRM’s fee restrict leaves a consumer half-onboarded with no alert anyplace.

Make: Retry, Then Repair the Bundle

Proper-click the module probably to fail and select Add error handler, then connect the Retry error handler. Retry pulls the failing bundle out of the circulate, shops it as an incomplete execution, and lets the remaining bundles end.

Retry wants Retailer incomplete executions enabled in situation settings, and that may be a scenario-level toggle, not a module-level one. Set the variety of makes an attempt and the interval between them; Make’s personal instance is three extra tries at fifteen-minute intervals. Make already auto-retries RateLimitError, ConnectionError, and ModuleTimeoutError as soon as incomplete executions are on. Don’t attain for Rollback right here: it solely reverses database-style actions, and it can’t unsend the welcome e-mail. On the canvas the handler seems as a second route off the module, drawn as a clear, dotted connection, because the screenshot beneath exhibits.

Zapier: Autoreplay Prices Duties and Hides the Failure

Autoreplay is a setting you activate account-wide or per Zap, on each plan together with Free, and it replays a failed step as much as 5 instances, with the final try touchdown roughly ten and a half hours after the primary error. Whereas it’s retrying, Zapier sends no error e-mail, so a damaged CRM connection stays quiet most of a working day.

The 2 replay modes behave in another way, and mixing them up will price you. Autoreplay and guide replay from Zap historical past by no means contact Filter and Paths steps in any respect. Replaying a whole run from the Zap editor does the other: it re-charges the actions that already succeeded and re-runs your Filter and Paths steps in opposition to the presently printed Zap. Order your steps round that break up.

Professional Tip: Put the welcome e-mail final, after the approval step. An entire-run replay re-executes earlier actions. You may delete a reproduction Drive folder quietly; a reproduction welcome e-mail prices you an apology.

Check It Earlier than a Actual Shopper Indicators

Run this sequence in opposition to a staging copy earlier than you level the manufacturing kind at something:

1. POST intake-sample.json to the staging webhook with curl, then run the Make situation as soon as and ensure each downstream discipline maps.

2. Use an organization area you management so the welcome e-mail lands in your individual inbox.

3. Ship scope_notes as an empty string. The Zapier Filter or the Make router ought to cease the run; a clean kickoff transient in Drive means the gate is incorrect.

4. Approve one request, then decline one other. Verify the declined path leaves no orphan CRM document behind.

5. Open the Zap run in Zap historical past or learn Make’s execution log, take the precise job and credit score depend for the run, and multiply by your actual month-to-month quantity.

Roll again by deleting the take a look at folder, the generated doc, the CRM document, and the Information Retailer row. Depart the Zap run itself in historical past; its job depend is the one actual measurement you will have earlier than a paying consumer arrives.

Zapier or Make, and How To Transfer Between Them

Your app listing and your department depend settle this, and the 2 builds you simply walked via present why.

• Decide Zapier when your stack features a area of interest app that solely Zapier connects to. A local approval step and a readable linear step listing matter too, particularly if the one that inherits this from you’d quite not be taught a canvas.

• Decide Make if the run branches greater than twice, or for those who want the uncooked request physique for API calls just like the structured-output one above. Excessive month-to-month quantity pushes the identical means, as a result of per-task billing stings as soon as you might be onboarding just a few hundred shoppers a month.

You don’t have to select as soon as and stay with it. Bridge the 2 with webhooks: a Webhooks by Zapier POST motion can hand a payload to a Make customized webhook mid-run, and a Make HTTP module can hand it again. Migrate the AI and branching logic to Make first, depart the area of interest app actions in Zapier, and lower over one part at a time. Should you would quite have somebody map your current course of earlier than you construct it twice, Adam works on precisely this sort of enterprise course of automation.

The place To Take This Subsequent

You now have a consumer onboarding automation that validates its personal AI output, stops for a human earlier than cash or contracts transfer, and recovers from the mid-run failures that used to strand a consumer between the folder and the welcome e-mail. The reusable form beneath is a typed consumption payload that feeds a schema-bound AI step, with approval damaged out onto its personal path. Swap the shape and the templates and the identical form covers vendor consumption or worker onboarding.

The tight_deadline flag is already in your schema and presently goes nowhere. Repair that first:

1. Add a risk_flags department that pings you in Slack when the mannequin tags an engagement as tight_deadline. In Zapier that may be a Path carrying one rule, risk_flags (Textual content) Accommodates tight_deadline; in Make it’s a second router route testing the identical string. Zapier’s information to conditional logic covers the branching aspect.

2. Set a calendar reminder to re-read your precise job and credit score consumption after thirty actual shoppers. The estimate within the desk above is arithmetic; your bill is the measurement.