A8gent

Planner

Workflow Automation Mapper

Describe your workflow steps in plain text. See which ones are automatable, the automation percentage, and which AI tool fits each step.

Workflow planner

Map workflow automation potential

Describe 3-8 steps and see which can be automated.

Free

Steps (3-8)

1
2
3
1

Send follow-up email

AutoAutonoly
95%
2

Update CRM with notes

AutoZapier
94%
3

Generate weekly report

AutoPandaDoc
85%

Automation potential

91

Pilot-ready

Average across steps.

Overall

91%

3/3 automatable.

Guided action plan

1Automate highest-% first
2Keep human review on low-%
3Connect via Autonoly or n8n
4Measure after 2 weeks

Recommended course

Auto guided

AI Agent Bootcamp

Highly automatable. Build and launch.

Open course path
Contents
  1. Fast answer
  2. Quick facts
  3. What this covers
  4. Why it matters
  5. Where to go next
  6. Outcomes
  7. Step by step
  8. The methodology: twelve patterns and one average
  9. How to read your automation percentage
  10. A worked example: a salon's rebooking process
  11. How to read your automation percentage
  12. The coordination tax your map will expose
  13. Sequencing: which mapped steps to automate first
  14. What changed in 2025-2026 for workflow automation
  15. The limits: keyword matching is not process analysis
  16. When not to use this tool, and what to do instead
  17. At a glance
  18. Mistakes to avoid
  19. FAQ
TL;DRAnswer-first

Fast answer

The workflow mapper takes three to eight steps written in plain English and matches each against twelve keyword patterns, with the highest match winning. Sending email scores 95, data entry and CRM updates 94, notify or follow up 93, scheduling 92, copy or sync 88, triage or route 87, generating reports 85, logging 85, drafting content 72, replying 72, verifying 70 and research 60. A step matching nothing scores 30. Fifty or above is labelled Auto, below is Human, and the headline figure is the plain average of the steps. Those percentages are editorial constants rather than measurements, and a low score means no pattern matched rather than that the tool detected human judgement. Read the per-step column instead of the average, because a workflow runs at the speed of its slowest human step, and rewrite any step as a verb plus an object before trusting its reading.

Facts last verified 2026-07

Quick facts

Work activities automatable
About 50% with demonstrated tech, McKinsey Global Institute
Occupations fully automatable
Under 5%; but 60% of jobs have 30%+ automatable activities, McKinsey
Work about work
58-60% of the workday, Asana Anatomy of Work Index
Coordination waste
103 hrs/yr unnecessary meetings, 209 hrs/yr duplicative work per person, Asana
Interruption rate
A ping every 2 minutes in core hours; 117 emails and 153 chats daily, Microsoft 2025
RPA expectation hit rate
85% met or exceeded expectations; payback 12 months vs 9 expected, Deloitte

On this page

What this page covers

A tool visitor should leave with a decision, not just a number: build now, prepare first, choose another workflow, or follow a course path.

  1. 01Workflow input
  2. 02Analysis
  3. 03Scores
  4. 04Recommendations

Why does this matter now?

Most automation projects are scoped from a description of the process that nobody has checked. Forcing yourself to write the steps as verbs, in order, exposes the missing ones, the duplicated ones and the ones that only exist because of a data problem upstream. That is worth more than any percentage the tool prints, and it takes five minutes.

Internal path

Where to go next from this page

These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.

Start with readiness

What you should be able to do after this

  • Step analysis
  • Automation scoring
  • Tool matching
  • Implementation plan

How do you do it, step by step?

1. Write the process as it actually runs

Not the official version. Include the waiting, the chasing and the step someone does from memory. Three to eight steps, each a verb and an object.

2. Read the per-step column first

The average hides the constraint. A workflow moves at the pace of its slowest human step, so find that step before you look at the headline number.

3. Rewrite anything that scored 30

A 30 usually means the phrasing did not match a pattern, not that the step needs a person. Name the system and the action - query the client list, update the booking record - and read it again.

4. Add the exceptions

Missing data, a customer who replies with a question, an unusually large amount. If your score does not fall when you add these, you have not added the real ones.

5. Move the human step to the end if you can

A human step at the end is an approval gate. In the middle it is a queue that needs an owner, a timeout and an escalation. Redesign for the first shape wherever the work allows.

The methodology: twelve patterns and one average

You type three to eight steps in plain English. Each step is tested against twelve regular expressions, and the highest-scoring match wins. The patterns and their percentages are: sending email 95, data entry or updating a CRM or database 94, notify, remind or follow up 93, scheduling 92, copy, move or sync 88, sort, categorise, triage or route 87, generating a report, invoice or summary 85, log, record or track 85, writing, drafting or creating content 72, respond, reply or answer 72, check, verify or validate 70, and research or analyse 60.

A step matching nothing scores 30. Anything at 50 or above is labelled Auto and anything below is labelled Human. The overall figure is the plain average of the step percentages, and the meter is that average clamped between 20 and 96. Each matched pattern also carries a fixed tool name, which is a hint about the category of tool rather than a recommendation for your business.

Two consequences follow and both matter. First, the percentages are editorial constants, not measurements: nobody has established that emailing is 95 percent automatable and drafting is 72. They encode a defensible ranking - mechanical actions above generative ones, generative above research - and the ranking is the useful part, not the digits. Second, a low score means no pattern matched, not that the tool detected judgement. Write negotiate contract terms and you get 30 because none of the twelve expressions fire, which is the same 30 you get for typing something meaningless.

That second point is the one to hold on to. The mapper recognises verbs it knows and shrugs at everything else, so it can only ever confirm that a step looks like familiar automation work. It cannot tell you that a step requires human judgement; it can only tell you that it did not recognise the step. Those are very different claims and only the first one is supported by the code.

How to read your automation percentage

Read the per-step column, not the headline. The average across steps is arithmetically fine and operationally misleading, because workflows fail at their weakest link rather than their mean. A five-step process at 88, 95, 30, 93 and 85 averages 78 and looks strong, but that 30 in the middle is a person, and every step after it waits on them. The realistic throughput of the workflow is set by that step's cadence, not by the average.

Where the human step sits changes the design completely. A human step at the end is an approval gate, which is cheap and usually correct. A human step in the middle is a queue, and queues need someone watching them, a timeout and an escalation. A human step at the start means the trigger is manual, which is the version most likely to quietly stop being used within a month.

Phrasing changes the answer, which is a limitation and also a technique. Write update the CRM and you match at 94. Write make sure the client record is right and you match nothing and score 30. Rewrite each step as a verb plus an object, in the language of the system it touches, and you will both get a better reading and produce a description a builder can implement. If a step cannot be written that way, that is a genuine finding: it means the step is not yet specified.

Finally, be suspicious of a very high overall score. Real processes have exceptions, and a workflow that reads as 90 percent automatable usually means the exceptions were not typed in. Add the awkward steps - the ones you do when the data is missing, the customer replies with a question, or the amount is unusually large - and see what the number does. It should fall, and the fall is the honest part.

A worked example: a salon's rebooking process

A salon owner maps the process for filling next week's gaps. Six steps: check the diary for open slots, pull the list of clients past their rebooking window, send a message offering a time, log the reply, update the booking system, and send a confirmation the day before. Against the twelve patterns those score 70 for check, 30 for pull the list because nothing matches, 95 for send, 85 for log, 94 for update, and 93 for the confirmation reminder.

The average is about 78, and the label on the 30 is Human. But the step scoring 30 is not judgement work at all: pulling a list of clients past a service-specific rebooking window is a database query, which is one of the more automatable things on the page. It scored low because the phrase did not match any expression. Rewrite it as query clients whose last visit is past the rebooking interval and the reading changes entirely.

That is the mapper's characteristic error and it is worth seeing once, because it cuts both ways. It also labels send a message offering a time at 95, which is technically correct and operationally incomplete: the salon workflows we describe on the salons page depend on a waitlist backfill that has to handle a cancellation arriving by phone, and on rebooking prompts that vary by service. The sending is easy; the deciding who to send to, and the not sending twice, is the build.

The useful output for this owner is therefore not 78 percent. It is the discovery that five of six steps are mechanical, one was badly described, and the real work is the state handling nobody typed in: who is on the waitlist, who already got a message, and what happens when someone replies with a question instead of a yes. That is the specification the workflow planner would have asked for, and it is the artefact worth having.

How to read your automation percentage

The mapper labels each step of your workflow as automatable, assisted, or human, and the resulting percentage tells you which category of workflow you are holding. The research says a middling score is normal: McKinsey Global Institute's analysis of more than 2,000 work activities across 800 occupations found about half of all work activities are technically automatable with demonstrated technology, roughly 60 percent of occupations have at least 30 percent of activities that could be automated, and fewer than 5 percent of occupations can be automated entirely. If your workflow scores 40 to 60 percent, you are looking at the standard human-machine split, not a disappointing result.

Interpret the step types, not just the total. Steps that move data between systems, apply fixed rules, or draft text from structured inputs are the automatable core. Steps that require judgment on ambiguous inputs, negotiation, or accountability - approving an exception, signing off on a client deliverable - stay human, and the tool marks them that way deliberately. The middle category, assisted steps, is where AI agents changed the picture: drafting, summarizing, classifying, and researching used to be human-only and now run with a human reviewing rather than producing.

A low score is information too. Workflows dominated by judgment calls, relationship work, or physical activity resist automation, and forcing agents into them produces the failure cases that populate the cancellation statistics. The better play with a low-scoring workflow is usually to automate the coordination around it - the scheduling, the status updates, the handoffs - which is where a surprising share of the time actually goes.

The coordination tax your map will expose

When operators map real workflows, the shock is rarely the task time; it is the coordination overhead between steps. Asana's Anatomy of Work Global Index found knowledge workers spend around 60 percent of their time on work about work - status updates, chasing approvals, searching for documents, duplicating effort - rather than on the skilled work itself, with the 2023 edition still putting it at 58 percent. The same research quantified the waste at 103 hours per year in unnecessary meetings, 209 hours on duplicative work, and 352 hours talking about work, per person.

Microsoft's telemetry paints the interruption side of the same picture. Its June 2025 Work Trend Index special report found workers are interrupted roughly every two minutes during core hours - up to 275 times a day for the heaviest-pinged fifth of users - while receiving 117 emails and 153 Teams messages on an average weekday. Its earlier research found employees spend 57 percent of their digital time communicating and only 43 percent creating. Every handoff in your workflow map is a place where that interruption economy taxes the actual work.

This is why the mapper scores handoffs and waiting states, not just tasks. An agent that watches a mailbox, updates the CRM, notifies the right person, and prepares the next step's inputs does not just save the five minutes of the task; it collapses the hours or days the item would have sat in a queue. When you present the map to your team, the automation percentage will get attention, but cycle time through the workflow is usually where the business case actually lives.

Sequencing: which mapped steps to automate first

Start with the deterministic, high-frequency steps that the map marks fully automatable - data entry between systems, notifications, document generation from templates, status synchronization. These have decades of evidence behind them: Deloitte's global RPA survey found 85 percent of adopters said robotic process automation met or exceeded expectations on accuracy and timeliness, though actual payback averaged 12 months against an expected 9, a useful calibration for your own timeline promises.

Take the assisted steps second, and design them as drafts plus review rather than full autonomy. An agent that drafts the client email, proposes the ticket classification, or pre-fills the report gets 80 percent of the time savings while the human veto keeps error cost near zero. This staging also builds the trust and the evaluation data you need before removing the review step on the narrow slices where the agent proves reliable.

Leave the judgment steps alone, but instrument them. Even when a decision stays human, an agent can assemble the context - pull the account history, summarize the thread, flag the anomalies - so the human spends their time deciding instead of gathering. Teams that sequence in this order compound wins safely; teams that start by automating the judgment step star in the post-mortems.

What changed in 2025-2026 for workflow automation

The step-type boundaries moved. Language-heavy steps that scored as human two years ago - summarizing a meeting, triaging an ambiguous email, extracting terms from a contract - now score as assisted or automatable, because LLM-based agents handle unstructured inputs that rule-based automation never could. When you re-map a workflow you last assessed in 2023 or 2024, expect the automation percentage to rise materially even though the workflow itself has not changed.

Leadership expectations moved with it. Microsoft's 2025 Work Trend Index found 82 percent of leaders expect to use digital labor within 12 to 18 months, and 24 percent say AI is already deployed organization-wide. The gap between that intent and the McKinsey finding that only about 7 percent of organizations report fully scaled AI is exactly the space where a concrete workflow map beats a strategy deck: it converts ambition into a sequenced list of steps with owners.

Tooling converged on the map-then-build pattern. n8n, Make.com, and Zapier all now embed AI steps natively alongside their integration catalogs, so the artifact this mapper produces translates almost one-to-one into a buildable workflow: automatable steps become nodes, assisted steps become AI nodes with human approval, and human steps become notifications and task assignments. The distance from map to running automation is now days for a simple workflow, which makes keeping the map current worth the effort.

The limits: keyword matching is not process analysis

The mapper has no model of your business, your data, your systems or your volume. It matches twelve keyword patterns against short strings, so it cannot see that two of your steps are the same step, that a step is only needed because of a data problem upstream, or that the step you left out is the one that takes the longest. Nothing in it does process analysis in any meaningful sense.

It cannot see risk. Send email scores 95 whether the email is an internal note or a price commitment to a customer, and the percentage carries no implication about whether the step should run unattended. Any step that reaches a customer, moves money or touches personal data needs the risk scorer regardless of what number it shows here.

It cannot see value. There is no frequency input and no duration input, so a step that runs 400 times a week and one that runs monthly contribute identically to the average. A high overall percentage on a workflow that runs twice a month is a technically automatable process with nothing much to gain from it.

And the tool names attached to each pattern are category hints stored as constants, not recommendations matched to your stack. Treat them as a reminder that a category of product exists for that step, then choose the actual tool by connector coverage and billing unit on the platform pages.

When not to use this tool, and what to do instead

Do not use it to decide whether to automate. It has no cost, no frequency and no risk inputs, so it cannot answer that question. Use the business automation audit to rank candidates by frequency and manual effort, then the ROI calculator to put hours and money against the winner.

Do not use it on a process nobody has documented. Typing six guessed steps produces an analysis of a guess. Watch the person doing the work for one cycle, write what actually happens including the waiting, then map that. The gap between the official process and the real one is where most automation plans fail.

Do not use it as a build specification. It produces labels, not a design. The workflow planner asks for the trigger, the trusted source, the review rule and the output, which is the minimum a builder needs, and the workflow builder turns that into something you can hand over.

And if the mapping shows a human step in the middle that nobody wants to own, that is the finding of the session. Either the step gets an owner and a service level, or the process gets redesigned so the human step moves to the end as an approval. Automating around an unowned queue just makes the queue fill faster.

At a glance

Where knowledge-work time actually goes (published research)

Time sinkMeasured valueSource
Work about work (coordination, status, searching)58-60% of the workdayAsana Anatomy of Work Index
Unnecessary meetings103 hours per person per yearAsana
Duplicative work209 hours per person per yearAsana
InterruptionsEvery 2 minutes in core hours; up to 275/day for heaviest usersMicrosoft Work Trend Index, June 2025
Daily message load117 emails, 153 Teams messagesMicrosoft, June 2025
Communicating vs creating57% vs 43% of digital work timeMicrosoft Work Trend Index 2023

What mistakes should you avoid?

  • Reading the average instead of the slowest step.
  • Treating a 30 as evidence that a step needs human judgement. It only means no pattern matched.
  • Mapping the official process rather than the one people actually follow.
  • Leaving out the exception paths, which produces a flatteringly high score.
  • Taking the tool name beside a step as a recommendation. It is a fixed category hint, not a match to your stack.

FAQ

How does the workflow mapper decide if a step is automatable?
It tests each step against twelve keyword patterns and takes the highest match. Sending email scores 95, data entry or CRM updates 94, notify and follow up 93, scheduling 92, copy or sync 88, triage or route 87, report and invoice generation 85, logging 85, content drafting 72, replying 72, verifying 70 and research 60. Anything matching nothing scores 30, and 50 is the line between the Auto and Human labels. The numbers are editorial constants that encode a ranking, not measured automation rates.
Why did my step get marked Human when it is clearly automatable?
Almost certainly phrasing. The tool has twelve patterns and a step that does not match any of them defaults to 30, which falls below the Auto threshold. Pull the client list matches nothing; query clients past their rebooking interval matches the data pattern. Rewriting each step as a verb plus the object it acts on, named in the language of the system it touches, fixes most false Human labels and simultaneously produces a description a builder can implement without asking you three follow-up questions.
Can I trust the overall automation percentage?
As a rough indicator of how mechanical your process looks, yes. As a plan, no. It is the plain average of the step scores, so it hides the constraint: a workflow with one human step in the middle runs at that step's cadence no matter how high the average is. It also has no frequency or duration input, so every step counts equally whether it happens 400 times a week or once a month. Read the per-step column, find the slowest human step, and design around it.
What does it mean if my score is very high?
Usually that the exceptions are missing. Real processes branch: the data is incomplete, the customer replies with a question, the amount is unusual, the record already exists. Those branches are where the build time goes and they rarely match a keyword pattern cleanly. Add them as steps and watch the number fall. A workflow that still reads high after the exception paths are in it is genuinely a strong candidate, and one that only reads high before them is a demo waiting to disappoint.
Should I use the tool names the mapper suggests?
Treat them as category hints. Each pattern carries a fixed product name stored in the code, chosen to illustrate the kind of tool that handles that class of step, and none of it is matched to your systems, your volume or your budget. The choice that matters is the workflow platform in the middle of your stack, and that should be decided on billing unit and connector coverage for the systems you already run. The platform pages carry dated pricing and honest limitations for that decision.
How many steps should I enter?
Between five and eight for a useful reading. Three steps is usually a summary rather than a process and will score high because the messy parts have been rolled up. Beyond eight you are probably mapping two workflows that share a trigger, and they should be separated because they will have different owners, different approval rules and different failure modes. If your process genuinely needs fifteen steps, map the first eight and treat the handoff point as the boundary of build one.
Does a high automation percentage mean I should automate it?
No. The mapper says a process looks mechanical; it says nothing about whether the work is worth taking, how often it happens, what it costs today, or what happens when the automation gets it wrong. A perfectly automatable process that runs twice a month is not worth a build. Rank candidates on frequency and manual effort with the business automation audit, price the winner with the ROI calculator, and score anything customer-facing or financial with the risk scorer before you start.

Sources & further reading

Was this page helpful?