Zapier
Line-of-business glue. Form → CRM. Calendar → SMS. Email → spreadsheet. Fast to build, fast to fix, predictable cost. We pick Zapier when the workflow is linear and stable.
Zapier when it fits. Make.com when the logic gets gnarly. n8n when you want to own the stack. Custom code when nothing else clears the bar. We don't have a horse in the race — we have a deliverable: a workflow that runs when nobody's watching.
We map the actual workflow on paper first. Then we pick. About 60% of our builds end up on Make.com because of how it handles branching and arrays. Zapier wins for simple line-of-business. n8n wins when the client wants to host it themselves. Custom wins when no platform can do it well enough.
Line-of-business glue. Form → CRM. Calendar → SMS. Email → spreadsheet. Fast to build, fast to fix, predictable cost. We pick Zapier when the workflow is linear and stable.
Branching logic, iterators, JSON parsing, complex error handling. The right pick for 60% of what we build. Visual but actually expressive. We document scenarios so the next person can read them.
Self-hostable. Open source. The right pick when you need data residency, want to own the stack, or need to scale without per-task pricing. We deploy on Railway or your own VPS.
Code-first automation. We use it when a workflow needs real code (regex, custom auth, complex transformations) but doesn't justify a full backend.
Node.js or Python on Railway, Fly, or AWS. We pick this when the workflow is mission-critical, high-volume, or needs durability the no-code platforms can't promise.
Multi-step agents on Claude or GPT — agents that reason, call tools, and decide. We build with the Claude Agent SDK or OpenAI Agents SDK, deployed against your real systems.
Online order → POS sync. Inventory threshold → reorder. Review monitoring → manager Slack. Catering inquiry → CRM + 24h follow-up.
New-patient intake → PMS. Recall automation. Insurance verification queue. Treatment-plan follow-up sequences.
Referral intake → EHR. Pre-visit form completion → SMS. Lab result review queue. PA submission tracking.
Lead enrichment → CRM tag. Listing → MLS + portals + social. Closing checklist orchestration. Sphere reactivation sequences.
Job request → schedule + invoice + tech dispatch. Review request after job complete. Stripe → QuickBooks + Slack alert. NPS survey on completion.
Contract intake → e-sign + folder + matter setup. Time entry rollup. Client communication digests. Trust accounting reconciliations.
2–4 hours of working sessions. We sit with the person who actually does the job today. We draw it on a whiteboard. We find the broken handoffs and the duct tape. This is where most automations fail — wrong workflow.
Fixed price, fixed scope. We tell you which platform, why, and what the recurring platform cost will be (Zapier task counts, Make ops, OpenAI tokens, etc.). No surprises.
We ship the first piece in week 1 and let you run it before we build piece two. Most automation projects fail because the whole thing is built before anyone uses it.
Loom walkthrough, written runbook, and a "what to do when X breaks" cheat sheet. You should be able to fire us and keep the automation running.
Automation work is scoped per project. Below is what most engagements cost, by complexity. Recurring platform costs (Zapier, Make, n8n hosting, OpenAI tokens) are passed through with no markup and disclosed in the proposal.
Pricing in review — final ranges pending principal approval.
Because picking the wrong tool is the expensive part — and that's the part where the wrong tool ruins the project. We'll happily train your person on the tool we picked. Several engagements end with us writing the runbook for the client's ops manager who then owns it.
Yes — discovery is a paid fixed-fee block ($500–$1,500 depending on scope) that's credited toward the build if you proceed. We've found that "free discovery" results in shallow analysis and bad scopes. We'd rather charge for the thinking and quote the build accurately.
Three options: (1) you own it after handoff and call us when something breaks (T&M at $250/hr); (2) monthly retainer ($500–$1,500/mo depending on surface area) for monitoring, tweaks, and small additions; (3) Fractional CAIO if automation is one of several AI workstreams you want owned.
Agents make mistakes. We design for that: read-only when possible, human-in-the-loop on irreversible actions (refunds, sends, deletes), explicit confirmation patterns. The right pattern depends on the stakes. We won't ship an agent that can do damage without a confirmation step in front of damage.
If it has an API, yes. If it has a database we can query, yes. If it's a 1998 desktop app with no API, we might be able to bridge it with a watcher process, or we'll tell you the truth: an MCP server build or migration is the right answer first.