Choosing Platforms & Understanding Real Costs

You can build the same automation in many tools. The choice affects cost, reliability, data privacy and how easy it is for your client to look after it when you have gone. Products and prices change often. So this lesson teaches the decision criteria and the categories. It does not give a ranking that you would have to unlearn.

Categories of platform

• Hosted no-code automation platforms: visual builders with thousands of ready-made app connectors. They are the fastest to start and the easiest for non-technical clients. You pay per task or operation. Custom logic is limited.
• Visual workflow tools with more control: stronger branching, data changes and code steps. Some can be self-hosted, which helps with privacy and with cost at high volume. The learning curve is steeper.
• Code and serverless: scripts and functions with scheduled or event triggers. They give maximum flexibility and the lowest cost per run. A developer must look after them.
• AI agent and assistant builders: tools where an AI model decides which steps to take, using tools you connect. They are flexible for open-ended tasks and harder to make predictable. Use them for tasks that tolerate variation, with guardrails.
• Built-in automation inside the apps clients already use: CRMs, helpdesks and accounting tools often include automations that may be enough.

How to decide

• Connectors: does it connect natively to every app the client uses, or will you be writing API calls by hand?
• Pricing model: per task, per operation, per execution or flat? Work out the real monthly volume, including loops and failed runs.
• Reliability features: retries, error branches, run history, alerts and replay of failed runs.
• Logic power: branching, loops, data mapping, code steps, sub-workflows and scheduling.
• Data location and privacy: where data is processed and stored, whether you can self-host, and what the vendor's terms say about AI and training.
• Team features: roles, shared workspaces, version history and separate test and live setups.
• Ownership and hand-over: can the client own the account? Can workflows be exported?
• Staying power: how likely is the vendor to still exist in three years, to keep the connector working and to keep prices sensible?

Pricing traps

The sticker price rarely matches the real cost. A workflow that handles a list of 50 items in each run may use 50 times the operations of one that handles a single item. A trigger that checks every minute uses up tasks even when nothing happens. AI steps add model costs on top. Always work it out like this. Runs per month, times operations per run, times the price. Then add the AI cost per run, and the cost of retries. Compare the total with the value the automation creates.

def monthly_cost(runs, ops_per_run, price_per_1000_ops, ai_calls_per_run=0, ai_cost_per_call=0.0, retry_rate=0.05):
    platform = runs * ops_per_run * (1 + retry_rate) / 1000 * price_per_1000_ops
    ai = runs * ai_calls_per_run * ai_cost_per_call
    return platform + ai

# 800 enquiries/month, 12 operations each, 1 AI classification (~$0.003) + 1 AI draft (~$0.01)
print(round(monthly_cost(800, 12, 2.0, ai_calls_per_run=2, ai_cost_per_call=0.0065), 2))
# Replace the prices with your platform's real numbers - they change.

A sensible default

• A small, simple, non-technical client with popular apps: a hosted no-code platform, in the client's own account.
• Higher volume, complex data, privacy needs, or a wish for lower cost per run: a workflow tool with code steps, perhaps self-hosted by you or by a managed host.
• Highly custom or high scale: code, with proper hosting, logging and tests.
• Wherever it fits, use the built-in automation of the client's existing app for the simple parts. Keep your custom workflows for what that cannot do.