Workflow automation consulting: n8n vs Zapier vs custom code

Workflow automation consulting is outside help to decide which of your business processes should run on their own, choose the tool that should run them, and build them so they keep working. The deliverable is not advice alone: a good engagement ends with automations running in your accounts, documented well enough that your team can change them.

You will see the same service sold as business automation consulting, a workflow automation agency, Zapier consulting or n8n consulting. The names mostly signal which tool the firm prefers. The more useful question is whether the consultant will pick the tool to fit your process, or fit your process to the tool they already know.

Four things, in roughly this order. If an engagement skips the first, the rest tends to automate the wrong thing.

• 01A process map from observation. How the work really flows today, including the exceptions, the shared inbox and the person who fixes things by hand.
• 02A shortlist ranked by payback. Which processes are high-volume and rule-based enough to automate, and which are not worth it.
• 03A tool decision with reasons. Zapier, Make, n8n or custom code, and why - written down, so it can be revisited when volume changes.
• 04Built, monitored automations. Error handling, alerts, logs and a handover document, running on accounts you own.

Business automation consulting that stops at the first two items is a strategy engagement. That can be useful, but price it as one, and make sure someone is accountable for the build afterwards.

Start with processes that run often, follow rules most of the time, and cost real hours when done by hand. A process that runs fifty times a day and takes four minutes each time is worth far more than a complicated monthly task that takes an afternoon. Volume beats complexity when you are choosing where to begin.

A quick way to score candidates is to ask four questions of each one:

• 01How often does it run? Daily or hourly processes pay back quickly; quarterly ones rarely do.
• 02How rule-based is it? If most cases follow a clear rule and only a minority need judgement, automate the majority and route the rest to a person.
• 03Are the systems reachable? Tools with documented APIs are cheap to connect. Systems that can only be driven through a screen are expensive and brittle.
• 04How bad is a mistake? A mis-tagged enquiry costs little. A wrong payment or a wrong message to a customer costs a lot, and needs approval steps and testing budgeted in.

Typical first candidates are lead intake and CRM updates, invoice and receipt processing, support ticket triage, onboarding checklists that span several tools, and weekly reporting that someone assembles by hand from four exports. Each of these is common enough that the patterns are well understood, which keeps the first project predictable.

Be careful with processes that nobody agrees on. If three people do the same task three different ways, the consulting work is standardising the process first. Automating it before then just encodes one person's version and annoys the other two.

Use Zapier when the people maintaining the automation are not engineers and volume is modest. Use n8n when you want more control, self-hosting or lower cost at volume. Use custom code when you need error handling, state and testing that the platforms make awkward. Plenty of businesses end up with all three, each where it fits.

The point where a platform stops being the right answer is usually one of these: you need a failure at step three to be retried without repeating steps one and two; you need per-record state across runs; you need tests that catch a change breaking something; or per-task pricing has become a meaningful line on the bill. Below that line, keep using the platform. We cover the n8n numbers in detail in what n8n actually costs, self-hosted vs cloud - self-hosting runs roughly $14 to $150 a month in infrastructure, plus your own maintenance time.

Zapier consulting pays off when you already run a tangle of zaps that nobody fully understands, when task costs have crept up, or when a flow that matters keeps failing silently. The job then is mostly consolidation: merging duplicate zaps, adding error paths and alerts, removing polling that burns tasks, and documenting what each one does.

It is also worth an outside opinion before a migration. Moving from Zapier to n8n or custom code saves money at volume, but it moves maintenance from operations staff to engineers. If you do not have engineers, the cheaper platform can turn out to be the more expensive choice.

An n8n automation consultant should be comfortable in three places at once: n8n's visual editor, the JavaScript or Python inside code nodes, and the infrastructure n8n runs on if you self-host. Plenty of people can build a working n8n flow. Fewer can run one reliably at volume.

Support ticket triage is one of the most common first automations, and a good test of the buy-versus-build decision. Zendesk's intelligent triage uses AI to detect the intent, language and sentiment of incoming tickets, which you can then use in routing rules and views inside Zendesk. If your support runs entirely in Zendesk and those three signals are what you route on, turn it on before paying anyone to build a replacement.

A custom triage workflow earns its cost when triage depends on things the helpdesk cannot see: the customer's plan and contract in your billing system, their recent orders, an open incident, or a classification scheme specific to your product. It is also the answer when tickets arrive through several channels that do not all land in one helpdesk. In that case the workflow reads the ticket, looks up context from your own systems, classifies it with a model tested against your labelled history, and writes the result back into the helpdesk as fields and tags.

Either way, measure it. Take a few hundred historical tickets with known correct routes, run them through, and compare. A triage system nobody has measured is one nobody can improve.

For a single workflow built end to end, our engagements take four to six weeks. The shape is consistent even when the tools differ:

• 01Week 1: observe. Sit with the people who do the work, collect real examples including the awkward ones, and write down where the process actually branches.
• 02Week 1-2: design. Choose the tool, decide which steps need a model and which do not, and agree what a failure should do: retry, park for review or alert someone.
• 03Weeks 2-4: build. Integrations first, then the happy path, then the exceptions. Any AI step is tested against a labelled set of real examples before it goes near production.
• 04Weeks 4-5: parallel run. The automation runs alongside the manual process, and the two outputs are compared. This is where most surprises surface, and it is much cheaper to find them here.
• 05Week 5-6: cut over and hand over. Monitoring and alerts switched on, documentation written, and a walkthrough with whoever will own it.

The parallel run is the step most often cut to save time, and the one we are least willing to drop. An automation that has never been compared with the manual process is an automation you are trusting on faith.

Our workflow automation builds start at $4,500 for a single process and $4,500 to $7,500 for a suite covering one workflow end to end, delivered in four to six weeks. Running costs for most automation at moderate volume are $60 to $250 a month, paid on your own accounts.

Hourly consulting rates in the wider market vary enormously by region and seniority, so compare on delivered scope rather than rate. A cheaper consultant who builds a flow without error handling costs more once you count the hours someone spends checking it. Published rates and what moves them are on the pricing page.

Choose the one that asks the most about your process and the least about which tool you want. Then check these:

• 01Tool neutrality: they recommend Zapier, n8n or code case by case, and explain why
• 02Ownership: workflows, credentials and code live in your accounts, not theirs
• 03Failure design: they can show you what an alert looks like and how a failed run is replayed
• 04Handover: documentation your team can use to change a workflow without calling them
• 05AI restraint: models only at the steps where judgement is genuinely needed

When the work outgrows workflows - multi-step reasoning, tool use across several systems - it becomes agent development, which is a different engagement. If you are not sure which side of that line your process sits on, book a call and we will tell you.