AI Workflow Automation: Where to Start When Your Tools Don't Connect
Start With the Workflow That Breaks Most, Not the One That Demos Best
You're an ops lead at a company doing $2M to $10M, and your stack has outgrown your patience. Your CRM doesn't talk to your billing tool, your project software doesn't talk to Slack, and three times a day someone copies a number from one screen to another by hand. You've decided to fix it with AI workflow automation. Good. This article answers a plainly informational question: where do you actually start when your tools don't connect?
Here's the contrarian read, and it's the spine of everything below. Most teams start with the impressive thing instead of the painful thing. They wire up an AI agent that drafts marketing copy because it demos well in a meeting, while the invoice-reconciliation task that eats four hours every Friday stays manual. That's backwards. Start with your single most-broken, most-repeated workflow: the bottleneck. Automate the painful thing first, prove the model, then expand.
This isn't a tool-buying guide. This is about sequencing: which workflow goes first, how to find it, and how to avoid the trap that kills most projects.
Why "Automate the Impressive Thing" Fails So Reliably
Picking the wrong first workflow isn't a small mistake. It's the mistake that abandons projects.
The data is consistent. 68% of automation projects are abandoned within 18 months, and a recurring root cause is automating broken or low-value processes instead of fixing and prioritizing the right ones first. Automate the flashy workflow and you get something that looks good in a demo and changes nobody's week. The painful workflow keeps hurting. Six months later someone says "automation doesn't work for us," and the project dies.
There's a credibility cost too. The impressive automation is usually the one where a wrong answer is visible and embarrassing: an AI agent that emails a customer the wrong thing, a lead router that misfiles 8% of deals. The painful workflow is internal, rules-based, and forgiving of a human checkpoint. It's the safer place to learn, and where recovered hours actually compound.
Across the scaling companies we've worked with in the $2M to $10M range, the bottleneck is almost always the same shape: a repetitive, rules-based task wedged between two tools that don't share data, with one person acting as the human glue. That person is the integration. Replace that, and you've recovered real hours. Automate the demo, and you've recovered applause.
The research agrees on what a good first target looks like. McKinsey interviewed Leslie Willcocks, a London School of Economics professor who has studied process automation for years, on which processes to automate first. His answer: "It has to be stable, mature, optimized, rules-based, repetitive, and usually high-volume." That fits your bottleneck workflow far more often than the impressive one.
Automate-the-Painful vs Automate-the-Impressive: A Side-by-Side Comparison
Put the two starting strategies next to each other. One reliably ships value. The other reliably stalls.
| Automate the Painful (Recommended) | Automate the Impressive | |
|---|---|---|
| What you pick first | The most repeated, most hated internal task | The flashiest AI feature that demos well |
| Failure if it's wrong | Mild, internal, human still checks the output | Visible, customer-facing, erodes trust fast |
| Hours recovered | Real and immediate (the task ran daily) | Often near zero (the task was rare) |
| Team buy-in after week one | High, the pain is gone | Low, nothing in the day changed |
| Maintenance risk | Lower, rules-based and stable | Higher, model behavior drifts on live data |
| Fits the Willcocks test | Yes: stable, repetitive, high-volume | Rarely |
The pattern in that table is the whole method. Start where the pain is highest and the failure cost is lowest. That combination makes a first automation both safe and worth doing. The impressive workflow can come later, once you've proven the wiring holds and your team trusts the system.
How to Find Your Bottleneck Workflow in 5 Steps
You don't need a consultant or a 40-row spreadsheet. Run these five steps in order and you'll have your answer in an afternoon.
- List every task someone touches more than three times a week. Anything repeated daily or near-daily is a candidate. Rare tasks, however annoying, are not where you start.
- Cross off anything that isn't rules-based. If a human uses real judgment every time, it's not ready. You want tasks where the same inputs always produce the same action. Those pass the Willcocks test.
- Score what's left by pain times frequency. Multiply how much each task hurts (1 to 5) by how often it runs per week. The top of that list is your bottleneck, and it's almost never the workflow you'd have guessed in a meeting.
- Confirm it spans two tools that don't connect. The highest-leverage automations live in the gap between systems: the copy-paste between your CRM and billing tool, the manual Slack ping after a form fills. That gap is where the hours leak.
- Name the owner before you build a thing. Write down the human who will fix this when it breaks. If that name is "the founder, in spare time," fix that gap first, because an unowned automation is just a future outage.
That's the method: pain times frequency, rules-based, in the gap between two tools, with a named owner. The workflow that scores highest is where you start.
Bottleneck Workflow Readiness Checklist
Before you build, walk this checklist against your chosen first workflow. If most of these are true, you've picked the right one. If they're not, you've probably picked the impressive thing.
- The task runs at least three times a week, every week.
- The same inputs reliably produce the same correct action (rules-based, not judgment-based).
- It currently spans two or more tools that don't share data automatically.
- A human is acting as the copy-paste bridge between those tools today.
- A wrong result is caught internally before it reaches a customer.
- You can name the specific person who will maintain it when it breaks.
- You're connecting tools you already pay for, not buying new ones to justify the build.
Naming the gaps here is the point. A good ops partner tells you which boxes you're missing before you spend a dollar, instead of selling you a build and walking away. If three or more boxes are blank, slow down. The readiness gap, not the tool, is what sinks first automations.
Sequencing After the First Win
Once your bottleneck workflow is live and stable for two weeks, you've earned the right to expand. Now sequence outward, not randomly.
Point the recovered hours and team trust you just banked at the next-highest pain-times-frequency task on your list. Most teams find their first three automations cluster around the same broken seam: the CRM that doesn't talk to billing, the form that doesn't trigger fulfillment, the report nobody wants to assemble by hand. Clear those in order and the compounding starts.
The platforms you'll build on already sit in your stack. For most $2-10M ops teams that want visual, multi-step builds at a predictable cost, we recommend Make as the default first canvas. Make starts at $12 per month for 10,000 credits, so your first automations build on infrastructure you can afford without a procurement cycle. The tool is the cheap part. The sequencing, wiring, and ownership are the work, which is why we build on your accounts, staging first, with monitoring on every workflow. You own everything and can cancel anytime.
The payoff for getting the order right is concrete. 94% of knowledge workers regularly perform repetitive, time-consuming tasks, and 88% of small and midsize businesses say automation lets them compete with larger companies by moving faster and spending less time on busywork. You only capture that upside if you start with the workflow that's actually bleeding hours. The bottleneck-first method is how you get there, and it's the core of broader business process automation: fix the highest-leverage seam first, then expand along the pain.
Frequently Asked Questions
How do I know which workflow to automate first?
Score every recurring task by pain times frequency, then keep only the rules-based ones that span two tools that don't connect. The highest score is your bottleneck, and that's your first automation. It's rarely the impressive AI feature that demos well. It's usually the quiet internal task someone does three times a day. Start there because the failure cost is low and the recovered hours are real.
Should I start with an AI agent or a simple connector?
Start with the simplest thing that clears the bottleneck, usually a connector, not an agent. 66% of workers say automation lets them focus on more creative tasks, and most of that comes from removing dull copy-paste work, not sophisticated AI. Add an AI step only where "roughly right, human-checked" beats "manual every time," like summarizing or classifying, and keep it inside a workflow a human still supervises.
Why do so many automation projects get abandoned?
Because teams automate the wrong thing first, then nobody maintains it. 68% of automation projects are abandoned within 18 months, often from automating a flashy or broken process instead of the painful, rules-based one. The build is an afternoon. The ownership is forever. Pick the bottleneck workflow, name an owner, put monitoring on it, and you avoid the failure mode that kills most projects.
Can I do this myself or do I need a partner?
You can absolutely start yourself, and plenty of ops leads do. The honest question isn't whether you can build the first automation. It's whether someone will fix it when an API changes or a token expires. If you have a named owner with time to maintain it, build it yourself. If that owner is "you, in spare time," that gap is where a fractional ops partner earns its retainer.
Do This Next
Pick the single task someone on your team touches more than three times a week and follows the same rules every time. Write down how often it runs and which two tools it bridges, so you can see the hours it's quietly costing you. Score your top five recurring tasks by pain times frequency, and start with whichever lands at number one. Name the person who will own that automation before you build a thing, because an unowned workflow is just a future outage. Build that one automation in Make, watch it run for two weeks, then sequence outward to the next-highest pain on your list. If the owner's name comes up blank, book a call with us instead and we'll map your stack with you.
Related guides
- business process automation: the founder's field guide
- AI integration: connecting your stack without breaking it
- AI implementation: pilot to production without the bottleneck
- AI adoption strategy: a plan that survives contact with reality