How to Build a Concrete Automation List for Your Business
The most common automation failure in mid-market companies is not bad technology. It is no list. They automate whatever feels urgent in the moment; a tool someone demoed at a conference, a workflow someone heard about on a podcast; and miss the workflows that actually compound. “We should automate more” lives in every founder’s head. A written, prioritised, scored list of exactly which workflows to automate first lives in almost none of their documents.
This article produces that list. One 90-minute working session. A specific methodology, a scoring rubric, and a Sprint 1 you can act on this month.
“The businesses pulling ahead are not automating more things. They’re automating the right things first; the ones that run daily, cause real friction, and compound when they disappear.”
Why “we should automate more” never turns into anything
Three failure patterns explain why most automation initiatives stall at the intention stage. Each one is recognisable.
Automating the interesting task, not the important one
The CEO sees a demo of AI generating marketing copy and puts “automate content” on the roadmap. Meanwhile, the ops team spends 14 hours a week on invoice reconciliation. The interesting task gets attention. The high-friction task gets ignored.
The roadmap reflects what sounded good in a meeting; not what the business actually needs.
Starting with the tool, not the inventory
A new AI tool gets purchased. The team is asked to “figure out how to use it.” Nobody maps the workflows first. The tool gets used for ad hoc tasks and never deployed at scale. Twelve months later: renewal cancelled. The problem was not the tool. It was the missing inventory underneath it.
No decision criteria
When someone suggests automating a workflow, there is no framework for evaluating whether it belongs on the list. Discussions become debates. Nothing gets prioritised. The list stays in everyone’s head and nobody’s document.
The solution is a method: a structured workflow inventory, a two-axis scoring system, and a forced ranking that produces a list you can act on this month.
The two axes that determine what gets automated first
The automation prioritisation framework runs on two axes. Score every workflow against both before making any build decision.
Axis 1 — Frequency
How often does this workflow actually run?
| Frequency | Score |
|---|---|
| Daily | 4 |
| Weekly | 3 |
| Monthly | 2 |
| Quarterly | 1 |
| Rarely | 0 |
Axis 2 — Friction
How much pain does it generate when it runs?
| Pain level | What it looks like | Score |
|---|---|---|
| High | More than 2 hours per run; consistent complaints; error-prone | 4 |
| Medium-high | 45 minutes to 2 hours per run | 3 |
| Medium | 15 to 45 minutes per run | 2 |
| Low | Under 15 minutes | 1 |
| None | No real friction | 0 |
Total score = Frequency + Friction
| Score | Decision |
|---|---|
| 7–8 | Automate first; Sprint 1 |
| 5–6 | Automate next; Sprint 2 |
| 3–4 | Quick win when bandwidth allows |
| 0–2 | Deprioritise |
The top-right quadrant; high frequency, high friction; is always the starting point. The bottom-left is always skipped. The scoring exists specifically to prevent the interesting-but-infrequent task from displacing the boring-but-daily one that would actually free up your team.
The 90-minute workflow inventory; how to run it
One founder or ops lead. A blank document. The scoring rubric above. No committee. No consensus required at this stage. Committees produce compromised lists. One person with the right framework produces a useful one.
Step 1 — Block 90 minutes, no interruptions
Calendar it. Close Slack. Treat it as seriously as a board prep session. The constraint is what makes the output real rather than aspirational.
Step 2 — List every recurring task in the business (30 minutes)
Start by department. List every task that happens on a recurring basis; daily, weekly, or monthly. Do not filter yet. Just list.
Starter prompts to unlock the inventory:
- “What does the team do every Monday morning without exception?”
- “What tasks generate the most complaints in weekly team meetings?”
- “What work gets pushed to Friday because nobody wants to do it?”
- “What would fall apart if one specific person left tomorrow?”
- “What did we do manually last week that felt like it should not be manual?”
Target: 20–40 tasks across the business. Fewer than 15 means you are filtering too early. Let the list be messy; you are not deciding anything yet, you are inventorying.
Here is what a partial inventory looks like for a $12M professional services firm:
| Task | Department | Rough frequency | Notes |
|---|---|---|---|
| Invoice reconciliation | Finance | Weekly | 3–4 hours; error-prone; crosses QuickBooks and email |
| Pipeline summary report | Sales | Weekly | Manual export from HubSpot; formatted in Google Sheets |
| Meeting action items | Ops | Daily | Often missed; ownership always unclear |
| Vendor follow-up emails | Procurement | Weekly | Repetitive; template exists but still manual |
| Customer follow-up emails | Sales | Daily | High volume; tone varies by rep |
| Expense categorisation | Finance | Weekly | Manual; CFO reviews exceptions |
| Job posting formatting | HR | Monthly | Template-based; always takes longer than it should |
| New client onboarding checklist | Ops | Variable | Inconsistent; relies on one person’s memory |
Step 3 — Score each task (20 minutes)
Apply frequency + friction scoring to every item. Aim for 30 seconds per task. A fast instinct score is usually more accurate than a committee-debated one.
After the initial score, run two additional filters:
- AI-readiness check. Is this task based on text, data, or structured information that AI can read and process? Tasks requiring physical presence or real-time human judgment in every instance score lower regardless of frequency and friction.
- Human checkpoint check. Can a clear approval gate be defined; a moment where a human reviews before output goes external? If not, the workflow needs redesigning before automating; not removing from the list.
Step 4 — Force-rank and pick Sprint 1 (15 minutes)
Take every task scoring 5 or higher. Rank from highest to lowest. Where scores are tied, use this tiebreaker: “Which one would my ops team celebrate most if it disappeared from their week?”
The top three are Sprint 1. Not the top ten. Three. Every item added beyond three reduces the probability that any of them ship properly.
Step 5 — Assign ownership and define the human checkpoint (15 minutes)
For each Sprint 1 workflow, decide three things before leaving the session:
- Who owns the workflow documentation?
- At what moment does a human review before output goes external?
- What is the 30-day build target?
If you cannot answer all three, the workflow is not ready for Sprint 1. Move it to Sprint 2 and pick the next item on the ranked list.
The “keep human” list; and why building it matters as much as the automation list
Every automation list needs a “keep human” column. This is not a concession to sceptics. It is good system design; and it makes the rest of the list more credible to the people most affected by it.
The fear is rarely “AI will replace me.” The fear is “AI will make decisions that should be mine and nobody will notice until it goes wrong.” The “keep human” column addresses that directly; and in writing.
Tasks that belong there:
- Customer escalation responses where relationship stakes are high
- Vendor negotiations and contract term decisions
- Pricing decisions and discount approvals
- Final hiring decisions (resume scoring: fine to automate; the hire decision: human)
- Board communications and investor updates
- Any decision with legal or compliance consequences in your jurisdiction
Here is what a complete, honest automation list looks like:
| Workflow | Score | Decision | Human checkpoint |
|---|---|---|---|
| Invoice reconciliation | 8 | Automate — Sprint 1 | AP manager approves exceptions |
| Pipeline report | 7 | Automate — Sprint 1 | Sales lead reviews before Monday meeting |
| Meeting action items | 7 | Automate — Sprint 1 | Team confirms in Slack before tasks are created |
| Follow-up email drafts | 6 | Automate — Sprint 2 | Rep approves before sending |
| Contract review summaries | 6 | Automate — Sprint 2 | Senior review required before sharing |
| Expense categorisation | 5 | Automate — Sprint 2 | Weekly spot check by finance lead |
| Pricing decisions | 5 | Keep human | — |
| Client escalations | 6 | Keep human | — |
| Vendor negotiations | 4 | Keep human | — |
| Board reporting | 5 | Keep human | — |
When team members see the automation list sitting next to the “keep human” list, the conversation changes. It stops being “how much will AI replace us?” and becomes “which parts of Thursday stop being our problem?”
The five workflows that almost always top the list
These five workflows consistently score highest across $5M–$25M non-tech businesses; not because they are technically interesting, but because they run constantly, generate real friction, and are well-suited for AI. Validate your own list against this pattern.
1. Invoice reconciliation
Frequency: weekly. Friction: high (3–5 hours, error-prone, cross-references purchase orders and email threads).
What automation looks like: AI reads incoming invoices, matches line items against open POs in QuickBooks or NetSuite, flags discrepancies for human review, drafts the vendor exception email. The AP manager approves exceptions rather than spending the afternoon finding them.
2. Meeting action item extraction
Frequency: daily. Friction: medium-high (items fall through the cracks; ownership is always unclear).
What automation looks like: AI reads the call transcript, extracts action items with owners and due dates, and creates tasks in Monday or Asana before the meeting host has closed their laptop.
3. Weekly pipeline or operations report
Frequency: weekly. Friction: high (2–4 hours of manual compilation from CRM, spreadsheets, and notes).
What automation looks like: AI pulls data from the CRM and project system, formats the weekly summary, highlights anomalies, and delivers to the relevant lead before the Monday meeting; without anyone compiling it.
4. Customer follow-up drafting
Frequency: daily. Friction: medium (repetitive, high volume, tone varies by rep).
What automation looks like: after a call, AI drafts the follow-up in the rep’s voice using the call notes and CRM history. Rep reads, adjusts if needed, approves, and sends. From 20 minutes to 3 minutes per follow-up.
5. Job posting and intake document formatting
Frequency: monthly. Friction: medium (template-based but always takes longer than it should; formatting inconsistency is chronic).
What automation looks like: the hiring manager fills out a brief role requirements form; AI produces a formatted job posting in the company’s standard structure and voice. Consistent every time, for every role.
The most common mistakes when building the first list
Scoring aspirational workflows instead of real ones
The list should reflect what actually happens in the business today. “AI-generated strategic analysis” sounds impressive and runs never. Invoice reconciliation runs every Thursday. Score what runs.
Putting too many workflows in Sprint 1
Every item added to Sprint 1 reduces the probability Sprint 1 completes. Three workflows is a real sprint. Ten workflows is a project that never finishes. Constrain first; expand after Sprint 1 ships.
Skipping the human checkpoint definition
Every automated workflow that ships without a defined review gate will eventually produce a bad output at the wrong moment. Define the gate before you build; not after the first incident breaks team trust.
Building automations before the context layer exists
Automating a customer follow-up workflow without a company voice guide produces generic outputs. The team stops using it within two weeks. The problem is not the automation; it is the missing foundation underneath it. The sequence matters:
- Build the context pack first
- Load it into the shared workspace
- Deploy the automation on top of it
Reversing that order is the most common reason Sprint 1 gets quietly abandoned.
Treating the list as permanent
The automation list is a living document. Workflows get added as the business changes. Some get retired. Review the scoring every quarter. A list that hasn’t been touched in six months is probably not being used.
Common questions from founders building their first list
”What if my business is too unique to use standard examples?”
The five examples in this article are starting points, not constraints. The 90-minute session works regardless of industry because it starts with your actual workflows; not a generic template. Run Step 2 without filtering and the right workflows for your business will surface from your own team’s patterns.
”Do I need to automate all five top workflows at once?”
No. Sprint 1 is three workflows maximum. Attempting five simultaneously reduces the probability that any of them get properly documented, deployed, and adopted. Ship three well before adding more.
”How do I know if my data is clean enough?”
For most workflows; invoice reconciliation, meeting summaries, follow-up drafting; data cleanliness is not the gating factor. AI handles messy email threads, inconsistent spreadsheets, and PDFs with varying formats. Start with one workflow, let it surface the data issues, and clean as you go. Waiting for clean data before automating is how automation never starts.
”How long before we see time savings?”
For well-documented, high-frequency workflows like invoice reconciliation or meeting action items: measurable within the first two weeks of deployment. The 30-day window gives you enough data to make the case for Sprint 2 with real numbers rather than estimates.
”Can I run Sprint 1 without a technical team?”
Yes. The workflows that top most $5M–$25M automation lists can be built using Make, Zapier, Claude, or ChatGPT without a developer. The constraint is documentation quality, not technical skill. If the workflow is well-mapped and the human checkpoint is defined, the build is straightforward.
The list is ready. The foundation is what ships it.
A concrete automation list is a one-session deliverable. The 90-minute session gets you the list. Sprint 1 follows from it. The founders who run this session once are months ahead of the ones who keep saying “we should automate more.”
Path one: start this week. Run the 90-minute session. Score your top 20 workflows. Force-rank. Pick your Sprint 1. You will know more about where to start after that session than after six months of researching tools.
Path two: bring in a partner. If you want the workflow audit, the automation list, the context packs, and Sprint 1 deployment handled in a single engagement; that is the work Phos does. The fastest way to know if it’s the right fit is a conversation. Thirty minutes, no deck. Start here.