Question 1

n8n: self-host or cloud?

Accepted Answer

1. Estimate executions per month
2. Self-host on small VPS if Docker is comfortable
3. Pick cloud if no sysadmin in-house
4. Count backups, updates and TLS in self-host cost

When to bring us in: 
Get quotes for both if executions reach tens of thousands.

Question 2

Zapier or Make: which fits better?

Accepted Answer

1. List required connectors
2. Count steps per run (Zapier task vs Make operation)
3. Pick Make for branching and data mapping
4. Pick Zapier for quick one-off integrations

When to bring us in: 
Build a proof-of-concept in both if this is a 5-year choice.

Question 3

Power Automate Cloud or Desktop: which to use?

Accepted Answer

1. Cloud flow if trigger comes from M365, SharePoint, Dataverse
2. Desktop flow if you must drive a GUI without an API
3. Combine: cloud starts desktop flow via gateway machine
4. Plan a dedicated machine for unattended desktop flows

When to bring us in: 
Ask a Microsoft partner for licensing advice if attended/unattended split is unclear.

Question 4

Someone has critical integrations in IFTTT

Accepted Answer

1. Inventory all active applets and owners
2. Flag which are business-critical (billing, customer contact)
3. Rebuild critical applets in Make or Zapier
4. Set IFTTT account read-only and plan retirement

When to bring us in: 
If applet history is lost, vendor has no audit trail to provide.

Question 5

Can n8n go in a commercial product?

Accepted Answer

1. Read current license on n8n site, not just GitHub
2. Internal workflows: free to use
3. Offering n8n as white-label: request commercial license
4. Document scope internally for later audit

When to bring us in: 
If unsure, send use-case to n8n licensing team.

Question 6

Trigger-based or scheduled poll: what to pick?

Accepted Answer

1. Check if source system offers webhooks
2. Use triggers if latency must be under 1 minute
3. Use schedule if batch is fine and webhook is missing
4. Set polling interval realistic, not every minute

When to bring us in: 
Hybrid: schedule as fallback for missed webhooks.

Question 7

Same webhook arrives twice, customer gets duplicate email

Accepted Answer

1. Identify unique ID in payload (event_id, order_id)
2. Store processed IDs in a table or cache
3. Skip if ID already seen, log the skip
4. Cache TTL long enough for retry windows

When to bring us in: 
On financial duplicates: rollback and manual reconciliation.

Question 8

An API blips and my flow fails outright

Accepted Answer

1. Add retry on external step, exponential, 3-5 attempts max
2. Catch error, push to DLQ table or channel
3. Alert only when DLQ grows past threshold
4. Build a replay mechanism to re-feed DLQ items

When to bring us in: 
Persistent DLQ growth points to upstream issues, not flow bug.

Question 9

Sales wants HubSpot deal updates in Slack

Accepted Answer

1. Create Slack channel #sales-deals
2. Build flow: trigger on deal-stage change
3. Filter only interesting stages (closed-won, lost)
4. Format compact message with deal link

When to bring us in: 
Too noisy: split into threads per deal owner.

Question 10

Incoming Typeform answers into a Notion database

Accepted Answer

1. Create Notion database with right property types
2. Build Typeform > Make/Zapier > Notion add-row flow
3. Test each field type, especially date and multi-select
4. Add fallback for empty fields

When to bring us in: 
High volume: write to Postgres, sync to Notion periodically.

Question 11

Notify new Calendly meetings in Slack or Teams

Accepted Answer

1. Try native Calendly Slack or Teams app first
2. Otherwise: Calendly webhook > flow > channel message
3. Show name, time, and calendar link
4. Add cancel-event as separate trigger

When to bring us in: 
Privacy concern: filter external domains out of the channel post.

Question 12

Stripe payments automatically into accounting

Accepted Answer

1. Configure endpoint in Stripe, copy signing secret
2. Verify signature in flow before processing
3. Map to ledger account and VAT code
4. Skip refunds and chargebacks or handle separately

When to bring us in: 
High volume: log to audit table and reconcile monthly.

Question 13

Edit content in Airtable, display in Webflow

Accepted Answer

1. Model Airtable as source-of-truth
2. Build Make/Zapier flow on record-update
3. Map to Webflow CMS collection fields
4. Trigger publish after successful sync

When to bring us in: 
Two-way conflicts: pick a master and stop fighting it.

Question 14

Auto-log Outlook meetings into Toggl or Harvest

Accepted Answer

1. Standardize title prefix or Outlook category per project
2. Flow reads event-create and event-update
3. Map prefix to project ID in time tracker
4. Skip all-day events and OOO

When to bring us in: 
Messy titles: ask user to classify after the fact via Slack buttons.

Question 15

Connector says Premium, flow refuses to run

Accepted Answer

1. Check connector page for Premium label
2. Decide: per-user plan or per-flow plan
3. Per-flow often cheaper for shared flows
4. Ask CSP for the right SKU

When to bring us in: 
Surprise premium flag in prod: cover flow with per-flow license temporarily.

Question 16

Approvals needed for procurement, expenses, leave

Accepted Answer

1. Design flow with approval step, parallel where needed
2. Write request and outcome to SharePoint list or Dataverse
3. Add reminder after N days of no response
4. Build quarterly audit report

When to bring us in: 
Financial approvals: log username and timestamp explicitly.

Question 17

SharePoint-list trigger reacts only after minutes

Accepted Answer

1. Accept polling delay as a design reality
2. For faster: use Dataverse or HTTP trigger
3. Do not split items into many small triggers
4. Document SLA back to business

When to bring us in: 
Strict realtime required: migrate entity to Dataverse.

Question 18

Dataverse or SharePoint list for flow data?

Accepted Answer

1. Under 5000 items, no relations: SharePoint list works
2. Above that, or relational data: Dataverse
3. Budget Dataverse capacity and per-user license
4. Late migration is painful, choose early

When to bring us in: 
Unsure: prototype in SharePoint, migrate to Dataverse when limits hurt.

Question 19

Flow stalls due to API limits

Accepted Answer

1. Check Action and Connector limits in Microsoft docs
2. Cache frequently read data in a variable or list
3. Spread peak processing across the day
4. Consider per-flow plan for heavy flows

When to bring us in: 
Persistent throttling: request expansion via M365 tenant admin.

Question 20

Legacy Windows app has no API

Accepted Answer

1. Install PAD on a dedicated VM or Cloud PC
2. Build flow with UI element selectors, not coordinates
3. Plan unattended runs with a service account
4. Capture a screenshot on each error for analysis

When to bring us in: 
Brittle scripts: ask vendor for an official API or CLI.

Question 21

Webhook never arrives, source says 'sent'

Accepted Answer

1. Test endpoint with webhook.site to see if payload arrives
2. Compare URL exactly, watch trailing slash and path
3. Check signing secret and algorithm
4. Open egress from source IP range if firewall is on

When to bring us in: 
Vendor can show delivery log with response code.

Question 22

Flow fails with 401, worked yesterday

Accepted Answer

1. Open connection in flow tool and re-authenticate
2. Check refresh-token policy isn't shorter than assumed
3. Use service account, not a personal one
4. Add monitoring on 401 spikes

When to bring us in: 
Repeated expiry: ask IdP admin for longer refresh window.

Question 23

Flow updates a record and re-triggers itself

Accepted Answer

1. Identify the flow's service account
2. Filter trigger on modified-by != flow account
3. Or set updated_by_flow field and skip if true
4. Test on 1 record before production rollout

When to bring us in: 
If loop is running: stop flow, clean residual records, restart.

Question 24

Scheduled flow runs twice or not at all on DST switch

Accepted Answer

1. Set all schedules explicitly in UTC or Europe/Amsterdam
2. Avoid 02:00-03:00 as run time
3. Test around DST shift (last Sunday March and October)
4. Log start time in both UTC and local

When to bring us in: 
Cross-timezone teams: standardize on UTC for all scheduled work.

Question 25

Flow changed, can't revert to working version

Accepted Answer

1. Enable built-in version history where available (n8n, Make)
2. Export flow JSON weekly to a Git repo
3. Maintain a per-release changelog
4. Test rollback procedure once a quarter

When to bring us in: 
Critical-flow incident: build rollback as a feature, not ad-hoc.

Question 26

Flow fails and nobody notices

Accepted Answer

1. Enable built-in fail notification per flow
2. Add a heartbeat flow that runs hourly
3. Send fail events to a dedicated #ops channel
4. Tier alerts: per-execution failure vs trend failure

When to bring us in: 
Silent-failure mode: alert on missing runs, not only on errors.

Question 27

We have dozens of old flows and nobody knows what's still in use

Accepted Answer

1. List all flows with last-run date
2. Mark 90+ days idle as retirement candidate
3. Send owner a disable-warning email
4. Disable, wait 30 days, delete

When to bring us in: 
No owner: route via team lead, don't delete silently.

Question 28

Employee left, their account still owns critical flows

Accepted Answer

1. Inventory flows on leaving account
2. Transfer ownership to team or service account
3. Re-authenticate all connections from new owner
4. Add 'rotate owner' to offboarding checklist

When to bring us in: 
Personal account deleted before transfer: ask IT for recovery via tenant admin.

Question 29

Zapier bill higher than expected

Accepted Answer

1. Review Zap history for task-heavy Zaps
2. Place Filter right after the trigger
3. Use Paths instead of duplicate Zaps
4. Migrate high-volume Zaps to Make or n8n

When to bring us in: 
Sudden spike: check loops, self-triggers, or new data source.

Question 30

Make operations quota gone early in the month

Accepted Answer

1. Review top-3 scenarios by operations
2. Move filters before Search steps
3. Replace loops with bulk API calls where possible
4. Schedule low-priority scenarios instead of realtime

When to bring us in: 
Structurally over plan: upgrade or split across two accounts.

Question 31

How do I know if a flow is worth its license?

Accepted Answer

1. Estimate manual time per run in minutes
2. Multiply by runs per month
3. Convert at internal hourly rate
4. Subtract license and build amortization

When to bring us in: 
Negative ROI after 3 months: pause or simplify the flow.

Question 32

iPaaS is overkill for simple scripts, what then?

Accepted Answer

1. Cloudflare Workers for light HTTP glue
2. AWS Lambda if you're already on AWS
3. Azure Functions if M365 tenant is central
4. Add secrets store and logging from day one

When to bring us in: 
Glue grows past 10+ functions: consider Workflow DevKit or Step Functions.

Question 33

Sheet automation in Google: Apps Script or external tool?

Accepted Answer

1. Simple Sheet-Sheet or Sheet-Mail: Apps Script
2. External SaaS in the mix: Make or Zapier
3. Apps Script in a shared project, not personal
4. Keep code in Git via clasp

When to bring us in: 
Apps Script quota exceeded: split across Workspace domains or migrate.

Question 34

Incoming invoice PDF must land in accounting

Accepted Answer

1. Mailbox rule routes invoices to a dedicated folder
2. Flow pulls attachment and sends to OCR (Document AI, Klippa, Moneybird AI)
3. Match vendor by VAT number or email domain
4. Push to accounting for approval, don't post directly

When to bring us in: 
Unknown vendor: park in review bucket and alert finance.

Question 35

Customer onboarding takes a day of manual work each time

Accepted Answer

1. Trigger on DocuSign-completed or CRM stage
2. Create accounts in needed tools (provisioning)
3. Send welcome email with magic link
4. Open internal Slack channel or project board

When to bring us in: 
Provisioning fails on a system: park in DLQ and alert customer success.

Question 36

New hire, 8 systems, half a day of manual work

Accepted Answer

1. Fill HR system with start date, role, team
2. Flow runs each morning and picks today's starters
3. Provision per system, idempotent
4. Send manager checklist and welcome email

When to bring us in: 
Provisioning failures: alert IT and HR together, don't skip silently.

Question 37

Departed employee still had access

Accepted Answer

1. HR termination event triggers flow
2. Disable account in IdP, don't delete
3. Archive mailbox and OneDrive
4. Transfer flow and doc ownership per list

When to bring us in: 
Dispute with employee: pause offboarding flow manually until resolved.

Question 38

Vacation requests are scattered between email and a sheet

Accepted Answer

1. Build form (Forms, Typeform, Power Apps)
2. Approval step to manager
3. On approve: write to HR system and team calendar
4. On reject: return to requester with reason

When to bring us in: 
Labor agreement constraints: have HR advisor review the rules.

Question 39

API keys sit in plain step fields

Accepted Answer

1. Move keys to the tool's credentials feature
2. Sensitive systems: integrate Azure Key Vault, AWS SM, 1Password
3. Rotate keys after every team change
4. Audit search 'sk_', 'Bearer', 'API-Key' in flow exports

When to bring us in: 
Suspected leak: rotate the key first, root-cause second.

Question 40

Flow app asks for Mail.ReadWrite, Files.ReadWrite.All for one mailbox action

Accepted Answer

1. Read app scopes before consenting
2. Where possible, scope down via tenant admin to specific mailboxes or sites
3. Use service principal with fine-grained role assignment
4. Review connected apps every quarter

When to bring us in: 
Broad scope not narrowable: pick another connector or build custom.

Question 41

Flow handles customer data, GDPR status unclear

Accepted Answer

1. Inventory which personal data flows through
2. Get vendor DPA and sub-processor list
3. Check data-residency setting (EU region)
4. Document processing in your record

When to bring us in: 
Special-category data: minimize or encrypt before transmission.

Question 42

Zapier Zaps scattered across employees' personal accounts

Accepted Answer

1. Create workspace on business plan
2. Invite employees, share Zap ownership
3. Connect SSO and disable personal login
4. Audit quarterly for orphaned Zaps

When to bring us in: 
Personal account refuses transfer: bring in manager, not IT alone.

Question 43

Three iPaaS tools, no idea which fits which flow

Accepted Answer

1. Low volume, broad integrations, non-technical admin: Zapier
2. Mid volume, complex data transforms and routing: Make
3. High volume, custom code, sensitive data that should not sit at a US SaaS: self-hosted n8n
4. A mix is normal: Zapier for sales flows, n8n for production flows touching customer data

When to bring us in: 
With more than a handful of critical flows lacking an owner, fix ownership and monitoring first, otherwise you only shift the problem.

Question 44

Booking incoming invoices manually, considering an OCR link

Accepted Answer

1. First test the OCR your package ships with, often fine for NL invoices
2. Set up a dedicated mailbox (boekhouding@) for suppliers, separate from the general inbox
3. Match rules on VAT number and email domain, not vendor name (varies too much)
4. Keep a review step, OCR is not 100%, especially on discounts, reverse-charge VAT, and EU invoices

When to bring us in: 
Above thirty invoices per week or multiple books, an external scan-and-recognise solution with approval workflow earns its keep.

Question 45

Mailbox doubles as todo list, things fall between the cracks

Accepted Answer

1. Pick a task source: Microsoft To Do, Todoist, Asana, or your ticket tool
2. Build the flow on a specific label or a manual action (flag, category), not on all inbound mail
3. Keep the mail link in the task so you can jump back to context
4. Agree which subject triggers, otherwise you get duplicate tasks on every reply

When to bring us in: 
If the same request (quote, support) keeps arriving by mail, automate the intake (form or customer portal). Mail-to-task fixes the symptom, not the cause.

Question 46

IT mailbox full of "I forgot my password" tickets

Accepted Answer

1. Enable SSPR in Entra ID, pick auth methods (Authenticator + phone, or Authenticator + mail)
2. Force registration on next sign-in, not "later"
3. Turn on writeback for hybrid environments, otherwise it does not sync back to on-prem AD
4. Communicate once with screenshots, otherwise people still call IT

When to bring us in: 
Without M365 Business Premium or Entra P1, SSPR is not included. Weigh the licence cost against your current IT hours on resets.

Question 47

Hours in Toggl or Harvest, payroll in Nmbrs or Loket, retyped manually every month

Accepted Answer

1. Standardise project codes and cost-bearer fields, otherwise nothing matches
2. Build a monthly flow that pulls hours, maps to cost bearer, hands off as CSV to payroll
3. Have the manager approve before the export hits payroll, not after
4. Store the export in a shared folder with the date, that is your audit trail

When to bring us in: 
With multiple collective agreements, shifts or allowances, do not roll your own, buy a time-and-payroll product that already handles it.

Question 48

Accounting leaks no signal to the team, payments and reminders stay invisible

Accepted Answer

1. Decide first which events the team must see, max five, otherwise it is noise
2. Trigger via webhook or poll flow from the package (Moneybird and Exact offer webhooks, Yuki via API)
3. Send to a dedicated Slack or Teams channel, not #general
4. Include the customer link and the amount, not just the invoice number

When to bring us in: 
With sensitive amounts or customer names in public channels, make the channel private and log membership, otherwise you have a GDPR debate.

Question 49

Zapier trial hits the monthly task cap mid-month

Accepted Answer

1. Open Zapier > Usage and read the count per Zap, not only the total
2. Place a Filter right after the trigger so a Zap only spends tasks on relevant items
3. Split multi-step Zaps so an expensive step (Search, Formatter) does not run unnecessarily
4. Plan the upgrade before you go to production, not under it. A paused Zap due to over-quota is silent downtime

When to bring us in: 
If volume is structurally above trial level, compare Make or n8n. For a few critical Zaps a paid plan is fine; for dozens of flows another platform is often cheaper.

Question 50

Webhook secret or API key visible in a screenshot in a ticket or doc

Accepted Answer

1. Rotate the key at the vendor now, not tomorrow. Treat a shared screenshot as a leak.
2. Find where the screenshot landed: Slack channels, ticketing tools, email threads, notes apps. Remove or replace.
3. Update the flow with the new key via the credential store, not in plain step fields
4. Agree as a team: redact secrets in screenshots before sharing when troubleshooting flow steps

When to bring us in: 
If it was a production key with access to customer data, this is a potential breach. Bring in your DPO or compliance owner to assess AP notification.

Question 51

Make scenario fails, retries, fails again, and burns through operations

Accepted Answer

1. Add an Error Handler route on the failing module, type Resume or Commit, not Rollback
2. Write the failed payload to a Data Store or your own "failed_runs" table
3. Filter the retry: do not retry on 4xx from the source; do retry on timeouts or 5xx
4. Alert on the "failed_runs" counter, otherwise the silent growth goes unnoticed

When to bring us in: 
Loop has been running for hours and operations quota is close: pause the scenario first, fix after. An empty quota mid-month also stops every other flow you have.

Question 52

Where should alerts go: Slack, Teams, email, or SMS?

Accepted Answer

1. Define three severities: info, warning, page
2. Route info to a separate chat channel that does not ping
3. Route warnings to your IT channel with @mention
4. Route pages to SMS or phone-call with an escalation policy

When to bring us in: 
For 24/7 paging with escalation, pick a tool with on-call rotation, like Better Stack or PagerDuty.

Question 53

How do I make sure an incoming webhook is really from the source?

Accepted Answer

1. Ask the source (Stripe, GitHub, Mollie, Shopify) to enable signing-secret and store the secret in a vault, not in the flow step itself.
2. In the first flow step compute HMAC over the raw body with that secret and compare to the signature header. Mismatch means stop immediately.
3. Use a time window of for example 5 minutes on the payload timestamp, otherwise a leaked old request stays replayable.
4. Where possible add an IP allowlist for known source ranges, especially payment providers that publish their IPs.
5. Log every rejected request separately, that is your early warning someone is guessing or abusing the URL.

When to bring us in: 
If your source has no HMAC support but sends financial or personal data, we can look at a proxy layer that adds verification.

Question 54

Should I poll or use a webhook to pick up changes?

Accepted Answer

1. Check first whether the source API offers webhooks. If yes, pick it. Latency is seconds instead of minutes and you pay no quota for empty checks.
2. Polling can make sense when the source has no webhook or when you are throttled on pushes. Don't set the poll interval lower than needed, every run is a task.
3. When polling, store a cursor (last-seen id or timestamp) so you only fetch new records and don't trigger 1000 duplicates.
4. Webhook flows need an endpoint that is always live. Plan a fallback poll once per hour to catch missed webhooks.
5. Do the math in tasks or operations: 1 webhook a day is near free, polling every minute is 43,200 runs per month.

When to bring us in: 
Unsure whether webhook plus fallback poll fits your SLA, we can set up the pattern for you.

Question 55

Step 4 of my flow fails, but steps 1 to 3 already wrote things to other systems

Accepted Answer

1. Define a counterpart for each write step: contact-created gets a delete-contact, invoice-created a credit-note or cancel-status.
2. Wrap all write steps in a try-block and on error: walk through the compensations in reverse order, not just for the last step.
3. Store in a log record which step-ids already ran, otherwise the error handler can't tell what to undo.
4. Idempotent compensations: a delete on an already-missing record must not fail, otherwise the rollback itself blocks.
5. Test deliberately: inject a fake error in step 4 and check that steps 1 to 3 are clean. Only then can you trust the pattern.

When to bring us in: 
If your flows touch money or inventory without rollback, that is priority one. We can look at the concrete pattern.

Question 56

My flow ran fine for weeks, now suddenly 401 and manual re-login

Accepted Answer

1. Check first whether the user recently changed their password or reset MFA, that invalidates almost any refresh token.
2. Check whether the OAuth app in the provider (Google, Microsoft, HubSpot) hasn't been revoked by an admin or security policy.
3. Ideally bind the connection to a service account, not a person. Personal accounts expire on departure, password changes trigger refresh failures.
4. Build a separate alert for 401 or invalid_grant: send a Slack message to the owner instead of just an empty error mail from the iPaaS.
5. Document per flow which account holds the connection, otherwise nobody knows who has to re-login.

When to bring us in: 
If this comes back across different flows, the issue is in your IAM choice, not the flow. We can look at that.

Question 57

I need to update 500 records, my flow takes an hour

Accepted Answer

1. Check the docs for /bulk, /batch or /multipart endpoints, or a GraphQL mutation that accepts an array.
2. Build the flow with an aggregator step (Make has Array Aggregator, n8n has Item Lists, Zapier has Looping) that groups rows into batches.
3. Keep batch size under the API cap, often 100 or 250 per call. Too large gives 413 or timeouts, too small wastes calls.
4. Handle errors per batch separately: if one record in a batch fails some APIs roll back the whole batch, others return per-record status.
5. Keep a log of success and failure records, otherwise you don't know which to retry after a partial failure.

When to bring us in: 
If your source API has no batch and you handle thousands of records daily, we can look at staging via a dedicated DB with flush.

Question 58

Flow errors disappear into a mailbox nobody reads

Accepted Answer

1. Split errors into 3 buckets: data error (input wrong, no action needed), transient error (retry fixes it), critical error (human needed).
2. Push critical errors to Slack or Teams in a dedicated channel with payload, flow id, step, and a direct link to the execution.
3. Push transient errors only if they recur N times in a row, otherwise the channel drowns in noise.
4. Log all errors to a central log (Postgres table or logging tool), separate from the channel message. The channel is for humans, the log for analysis.
5. Schedule a weekly review on the table: which flows fail most, and is that structural or incidental.

When to bring us in: 
Want a uniform error pipeline across your iPaaS tools, then an orchestrator table is the place. We can set up schema and alerting.

Question 59

Stripe or Mollie missed a webhook, how do I get that data in?

Accepted Answer

1. In the Stripe dashboard open the Events tab, filter on type (payment_intent.succeeded etc.) and hit 'Resend' on missed events.
2. For Mollie: call GET /v2/payments/{id} for each payment id and process the response as if it were a webhook. Put that call in a recovery flow.
3. Make your flow idempotent on event id, otherwise replay causes duplicate admin actions. Dedup key on event_id is the standard solution.
4. Build a catch-up flow that fetches the last 24h of events nightly and cross-checks with your own DB. If anything is missing, trigger then.
5. Document who has authority to trigger replays, otherwise someone double-processes a batch.

When to bring us in: 
If you need replay across weeks or months and want to avoid double bookings, we can test the catch-up together.

Question 60

Sometimes a customer gets 2 or 3 of the same mail, sometimes a record is duplicated

Accepted Answer

1. Pick a natural unique key from the payload: order-id, event-id, message-id, or a hash of source-id plus timestamp.
2. Store processed keys in a DB table or a key-value store (Airtable, Postgres, Upstash, Make Data Store, n8n Static Data) with TTL of say 30 days.
3. First flow step: check if the key already exists. If yes, log 'duplicate, skipped' and stop. If no, write the key and proceed.
4. Write the key as early as possible in the flow, not at the end. Otherwise a mid-flow crash can still cause double processing.
5. Monitor duplicates per day: a healthy flow sees a few, a spike points to a source bug or replay storm.

When to bring us in: 
If you need this across multiple flows, a shared dedup table with TTL and index pays off. We can deliver the schema.

Question 61

We need to rotate an API key, but 30 flows use it and can't go down

Accepted Answer

1. Centralise secrets in a vault: 1Password Secret Automation, Bitwarden Secrets Manager, Azure Key Vault, AWS Secrets Manager, or n8n credentials / Zapier auth-store.
2. Make sure the source API allows two active keys at once. Stripe, HubSpot and Mollie support it, some don't, check the docs.
3. Add the new key in the vault under the same alias or as a new version. Update flows to the new version and monitor 24h for errors.
4. Only when all flows run on the new key, deactivate the old in the source API. Document date and operator.
5. Schedule rotation as a standard quarterly or yearly calendar item, not just reactive after a leak.

When to bring us in: 
If you still use individual keys per flow and have a leak risk, a central vault is step one. We can start there.

Question 62

Our whole operation depends on Zapier, that feels risky

Accepted Answer

1. Inventory which flows are critical (orders, invoices, payroll) and which are nice-to-have (notifications, reporting). Lock-in pain sits in the first list.
2. For critical flows: build the logic in own code (Node, Python, .NET) on a platform you control (Cloudflare Workers, Vercel, Azure Functions, own server). iPaaS stays optional as trigger or glue.
3. For nice-to-have flows: stay in iPaaS, it saves time. Lock-in only hurts where it is business-critical.
4. Document per flow where the logic lives and which APIs it touches. On a vendor switch you won't have to reverse-engineer.
5. Keep your auth store independent: secrets in your own vault (see secret rotation), not in the iPaaS tool. Then migration stays feasible.

When to bring us in: 
Want to map which flows are risk-locked and what migration costs, we can run a quick scan.

Question 63

Flow goes live, turns out field X maps empty, customer sees odd data

Accepted Answer

1. Collect 5 to 10 representative payloads from production: a normal one, one with null fields, one with long strings, one with special characters, a boundary case.
2. Run the flow in test mode against each payload and check output field-by-field in the target API or a staging env.
3. Make a mapping table in a doc: source field, target field, transform, edge-case result. That doc is gold later on schema changes.
4. Build a test-suite flow that pulls a fixed payload set through the logic nightly and alerts if output deviates from baseline.
5. For financial flows: get a second pair of eyes before go-live, not just after the first customer complaint.

When to bring us in: 
If you have complex mapping where errors cost money, we can set up a test fixtures set and assertion flow.

Question 64

Source system added a field, suddenly errors or missing data

Accepted Answer

1. Subscribe to the changelog of every source API you use. Stripe, HubSpot, Microsoft Graph, Shopify have public release notes.
2. Make your flow tolerant of unknown fields: read with defaults, not strict schemas. A missing optional field shouldn't stop the flow.
3. On every target system deploy: check whether your mapping still matches the current API version. A test payload is faster than reactive debugging.
4. Pin API versions where possible. Stripe and HubSpot let you set a version header, giving you control over when to migrate.
5. Build a nightly schema check: pull a sample, validate against an expected JSON schema, alert on deviation.

When to bring us in: 
Got flows on APIs without version pinning and frequent changes, a gateway layer with schema validation helps. We can look at that.

Workflow automation