Capability template for operations and process improvement
Workflow fit check: is this piece of work ready for AI?
Before you build anything, run one piece of work through six checks and get a straight answer on one short page: build it, build one part, do something simpler first, not yet, or leave it with your people. The owner decides what gets built first, with the reasons in front of them.
The outcome
You stop guessing which piece of work to hand to AI first. Each candidate gets the same six checks and one short page that says what to do, why, and what would change the answer. The work that is ready goes to the front of the line. The work that only needs a setting switched on or a simple register gets that, often within the week, at no cost to build.
Building the right thing first is the fastest way to 10X what your team can do. A fit check that says “do not build this” is doing its job: it saves the weeks you would have spent on a workflow nobody could check, and it usually finds the one step inside the request that is worth building.
The answer is advice. The owner decides.
Who starts the work, and when
An owner or operations lead starts it, when someone says “can we get AI to do this?” and before any time or money goes into building. Run it again when the facts change: new software, a new plan for your AI product, or a register that now exists.
The person who does the work today supplies the detail. The owner reads the page and decides what happens next.
What they supply
- The request, in the words of whoever asked.
- How the work is done today. Notes from a short walkthrough with the person who does it: who does it and who approves, what starts it, what a finished result looks like, where each input lives, the awkward cases and what is done with each, how they would know a result was right, how often and how long.
- Two or three past cases, including one that went wrong, with people and companies named by role or label (the owner, Sub A).
- What you run on. The software the work touches, your AI product and plan, who owns the workspace, and any rule about what data may go into the AI product and who approves it.
Anything you do not know, you say so. The check marks each gap “Not known” and turns the ones that matter into questions for a named role.
What comes out
One short page per piece of work, two to three printed pages, held to five actions and eight open questions:
- The six checks, each marked clear, not clear or fails, with the words from your notes that show why: the people; start and finish; inputs; exceptions; examples and a way to judge; platform and permission.
- The six reasons to say no, each shown as applying or not: your software already does it; it is a list, a register, a reminder or a fixed rule; checking would take as long as doing; nobody here can judge it; it needs more than a person-run workflow; the data would pass through a party nobody has approved.
- The verdict, one of five: build it, build one part, do something simpler first, not yet, or leave it with your people. A fixed order decides when more than one fits, and the page says which line decided.
- What to do, in order, each line with the role that does it, starting with anything you can do without building.
- The part worth a workflow, if there is one, with what it would never do.
- What it is worth, using only the figures you supplied, and what would change the verdict.
- Open questions, each for a named role, and a blank line for the owner’s decision.
A good page lets a reader who was not in the room see why the answer is the answer.
An invented example
Illustrative example with invented data. Tarnbrook Mechanical is an invented commercial heating and cooling contractor. Its people, its cases and its software are invented too. The owner (Marcus) asks for an AI to track subcontractor paperwork; the operations manager (Priya) runs the fit check on 4 November 2026. What she supplied comes first, in full, with people named by role as the template asks. Then the page that came back. It is a worked refusal: the answer is not to build what was asked for.
Two things to notice. The owner asks for a tracker that chases; the walkthrough describes a check before each job start. The page does not stop on that and does not choose: it checks the piece the walkthrough describes, says so, and makes the choice the owner’s first question. And two reasons pull in different directions: checking the whole would take as long as doing it, and the work is a register the job software may already hold. The order of the verdicts settles it, and the page says which line decided.
What Priya supplied
1. The request (email from the owner, 2 November 2026)
We got burned twice in the past year with lapsed insurance on subs. I want an AI that tracks all of this for us: every sub, every certificate, every expiry date, and chases them when something is about to run out. Can we get that built this quarter?
2. How the work is done today (the operations manager’s walkthrough notes, 3 November 2026)
- About forty field staff. Fifteen to twenty subcontractors in rotation.
- Before a sub’s crew starts on a job I need three things to be current: their general liability certificate, their workers’ compensation certificate, and any trade licence the job needs.
- How I do it: the week before a start date, I search email and the shared drive for that sub’s latest certificates, job by job. About 20 minutes per sub per job when the files are easy to find. Longer when they are not.
- Job starts with subs per month: six to ten, I think. I have not counted.
- Certificates arrive as PDF attachments from insurers and brokers. Some come to me, some to the project manager, some to accounts.
- There is no single list. Some of the dates are in my head.
- Job start dates and sub records are in our job software. Each sub record has a “Vendor documents” tab. We have never used it. I do not know whether it holds expiry dates or sends reminders.
- Who decides whether a job starts: the project manager. The owner, if there is a gap.
- The awkward ones: a general contractor that requires particular additional-insured wording on the certificate; a broker’s letter in place of a certificate; a sub working under another company’s policy; a renewal that is “in process”.
- Result I want at the end: for each job starting next week, a list of what is missing or expiring.
- If someone else handed me that list, the only way I could trust it would be to go and find the certificates again myself. One date is different: I can check it against the PDF in a few seconds.
3. Three past cases (names removed)
- Case 1, March 2026. Sub A, rooftop unit replacement. The liability certificate had expired 11 days before the start. The general contractor’s site supervisor found it on day two. Crew stood down for one day.
- Case 2, June 2026. Sub B, ductwork. Both certificates current. Found in 10 minutes. No issue.
- Case 3, September 2026. Sub C, controls. Broker’s letter said renewal was in process; no certificate. I called the broker. The project manager let the job start after the owner agreed. The certificate arrived nine days later.
4. What we run on
- Job software: used for scheduling and sub records. “Vendor documents” tab unused (see above).
- AI product: the owner has a single-seat plan on his own login. Nobody else has an account. No shared workspace.
- Workspace owner and data approver: nobody is named for either. Both would fall to the owner.
- Data rule: none written. The certificates show each sub’s policy numbers. They were sent to us for this purpose.
- Bar for “quick to check”: not set.
- Today’s date: 4 November 2026.
What came back
WORKFLOW FIT CHECK: Subcontractor paperwork check before a job starts
Requested by: Owner Does the work: Operations manager Decides: Owner
Prepared: 4 November 2026, from the owner's email (2 November 2026), the
walkthrough notes (3 November 2026), Cases 1 to 3 and the "What we run on"
note.
THE REQUEST
"We got burned twice in the past year with lapsed insurance on subs. I want
an AI that tracks all of this for us: every sub, every certificate, every
expiry date, and chases them when something is about to run out. Can we get
that built this quarter?"
This page checks: the check the operations manager does before each start,
with the result in the operations manager's words: "for each job starting
next week, a list of what is missing or expiring". The request names a
different result. Two pieces were set aside and are judged under reasons b
and e: the tracking ("every sub, every certificate, every expiry date") and
the chasing ("chases them when something is about to run out"). The owner
chooses in question 1.
THE SIX CHECKS
1. The people: CLEAR
The operations manager does the work: "the week before a start date, I
search email and the shared drive for that sub's latest certificates".
A role acts on the result: "Who decides whether a job starts: the
project manager. The owner, if there is a gap."
2. Start and finish: CLEAR
Start: "the week before a start date". Finish: "for each job starting
next week, a list of what is missing or expiring".
Depends on: a record of each sub's certificates and expiry dates, which
does not exist: "There is no single list. Some of the dates are in my
head." Reason b is weighed below, and creating the list is action 3.
3. Inputs: NOT CLEAR
- Certificates: "PDF attachments from insurers and brokers. Some come
to me, some to the project manager, some to accounts." Searched for
in "email and the shared drive". The form is known; there is no
single home.
- Job start dates and sub records: "in our job software". Whether they
can come out as a file is Not known.
- Trade licences ("any trade licence the job needs"): where they are
kept and how they arrive is Not known.
4. Exceptions: NOT CLEAR
Four are named: "a general contractor that requires particular
additional-insured wording on the certificate; a broker's letter in
place of a certificate; a sub working under another company's policy; a
renewal that is 'in process'". What is done is shown for one, once
(Case 3): "I called the broker. The project manager let the job start
after the owner agreed." For the other three, what is done and which
role decides is Not known.
5. Examples and a way to judge: FAILS
Three past cases supplied, one that went wrong (Case 1). The bar is "not
set", so the default was used: under half the time the work takes. The
work takes "About 20 minutes per sub per job when the files are easy to
find"; half of that is 10 minutes. For the whole list: "the only way I
could trust it would be to go and find the certificates again myself."
Step that differs: reading one date from one certificate: CLEAR. "One
date is different: I can check it against the PDF in a few seconds."
6. Platform and permission: FAILS
"the owner has a single-seat plan on his own login. Nobody else has an
account. No shared workspace." The operations manager, who would run
it, has no account. "Workspace owner and data approver: nobody is named
for either." "Data rule: none written." The certificates "were sent to
us for this purpose"; that is not an approval to put them into the AI
product.
REASONS TO SAY NO
a. Your software already does it: APPLIES. Nobody has looked. "Each sub
record has a 'Vendor documents' tab. We have never used it. I do not
know whether it holds expiry dates or sends reminders."
b. It is a list, a register, a reminder or a fixed rule: APPLIES. "every
sub, every certificate, every expiry date" is a register. "chases them
when something is about to run out" is a reminder. The weekly list
rests on a record that does not exist: "There is no single list."
c. Checking would take as long as doing: APPLIES. Check 5 fails for the
whole: "the only way I could trust it would be to go and find the
certificates again myself."
d. Nobody here can judge it: DOES NOT APPLY. The operations manager judges
a date "against the PDF in a few seconds"; the project manager and the
owner decide the gaps (Case 3).
e. It needs more than a person-run workflow: APPLIES. "chases them" is a
message sent by the AI, and "tracks all of this for us" is a record kept
by the AI. Both are taken out: a person sends and a person records.
f. The data would pass through a party nobody has approved: DOES NOT
APPLY. The inputs name no outside service. The missing approval for the
AI product itself is under check 6.
VERDICT: DO SOMETHING SIMPLER FIRST
Line 2 of the order decides it: nobody has looked at what the job software
already does (a), and what was asked for is a register and a reminder (b).
Do not build "an AI that tracks all of this".
WHAT TO DO, IN ORDER
1. Operations manager: open the "Vendor documents" tab on one sub record,
with the software's help pages beside you, and note whether it holds
expiry dates and sends reminders. If it does, the owner has it switched
on and nothing is built.
2. Owner: name one place every certificate is sent.
3. Operations manager: put every sub's certificates and licences, with
their expiry dates, in one list: the tab if it does the job, otherwise a
simple register with a reminder on each date. Keeping the list and
chasing renewals stay with you; neither becomes separate work.
4. Operations manager: for one month, count the job starts with subs and
the minutes each check takes.
5. Owner: if the part below is still wanted once the list exists, settle
questions 2 and 3 (the data approval; the account, plan and workspace)
before anything is built.
THE PART WORTH A WORKFLOW
Reading one incoming certificate. The operations manager gives the
assistant one certificate PDF; it proposes the insured company, the kind of
cover and the expiry date; the operations manager confirms each against the
PDF "in a few seconds" and enters it in the list.
It would never: record a date nobody has confirmed; say whether a job can
start; judge a broker's letter, another company's policy or a renewal "in
process"; contact a sub, a broker or an insurer.
Ready now: No, it needs check 6 to be Clear (an account for the operations
manager, a plan a team can share, an approval for this data), the list to
exist (a reading with nowhere to put its answer saves nothing) and a few
real certificates to try it on. Give it its own fit check then.
WHAT IT IS WORTH
Supplied: "burned twice in the past year"; "About 20 minutes per sub per
job when the files are easy to find. Longer when they are not."; job starts
with subs per month "six to ten, I think. I have not counted."; "Fifteen to
twenty subcontractors in rotation"; Case 1: "Crew stood down for one day."
Not known: what happened the second time the company was "burned" (Case 1
shows one lapse; nothing supplied shows the other); a counted number of
starts per month; how long a check takes when the files are hard to find;
what the stand-down day cost.
Measures to watch: job starts where a missing or expired document is found
after the crew has started; minutes per sub per job for the check. A lapse
avoided is not a time saving. No saving is estimated here.
WHAT WOULD CHANGE THE VERDICT
- The "Vendor documents" tab holds expiry dates and sends reminders: the
verdict stands, the tab is the simpler thing, and nothing is built.
- The tab does not do it, the list exists and check 6 is Clear: the
certificate-reading part can be checked again as "Build one part".
- With the list in place, the operations manager can confirm the weekly
list without finding the certificates again: reason c stops applying.
OPEN QUESTIONS
For the owner:
1. Which result do you want checked: "for each job starting next week, a
list of what is missing or expiring" (the operations manager's words,
checked here), or a standing tracker that "chases them" (yours)?
2. Do you approve certificates showing "each sub's policy numbers" going
into the AI product? If not you, who approves?
3. Will the operations manager get an account, on which plan, and who will
own the workspace?
4. What is your bar for "quick to check"? The default was used: under half
the time the work takes.
For the operations manager:
5. Does the "Vendor documents" tab hold expiry dates, and does it send
reminders?
6. What do you do with additional-insured wording, a broker's letter, and
a sub working under another company's policy, and who decides each?
7. Where are trade licences kept?
8. Can the job software give you next week's starts with subs as a file?
Also not known: what happened the second time the company was "burned"; a
counted number of starts per month; how long a check takes when the files
are hard to find; what the stand-down day cost.
Status: Prepared for the owner's decision. Nothing built, sent or changed.
Owner's decision: [blank until decided] Date: [ ]
Marcus read the page with Priya and decided to start with action 1. That decision is his, and he writes it on the last line with the date; the page only prepared it.
Where it fits
The six checks are the same in any industry. These are the places the requests pile up fastest:
- Operations and process improvement, where the list of “could AI do this?” requests lands.
- Leadership, for owners choosing where the first build goes.
- Construction, where paperwork, schedules and subcontractors generate requests faster than anyone can sort them.
- Professional services, where much of the work is judgment and the fit check separates the part that repeats.
- Manufacturing, where existing systems often already hold the feature being asked for.
- Logistics and trade, where documents arrive in every form and the question is which ones a person can check quickly.
- Growth & scaling, when the work is outgrowing the people and you need to pick what to build first.
What has to change for your company
Name who decides what gets built. Say which AI product and plan you are on, and who may add a shared set of instructions or a connection. Write down your rule for data: what may be pasted into an AI product, and who approves it. Set your own bar for “quick to check”: the template suggests a person should confirm a result in well under half the time it takes to do the work. Decide where the finished pages are kept, so the next request is judged against the last ones.
What has been checked
Fully Powered AI wrote this capability template from its own method for qualifying a piece of work before building. It was editorially reviewed on 9 November 2026. It was tried on its invented example, in Claude on 9 November 2026: one recorded run, after two earlier runs the same day on earlier wording. The record is published with it and lists the small edits made to the instructions since. Runs on one invented example are not a test. It has not yet been tested in a client environment. It has not yet been tested as a packaged version in ChatGPT, Claude, Copilot or Gemini.
Testing in your environment means running it on two or three requests whose answer you already know, including one you built and regret, and seeing whether the page reaches the same answer for the same reasons. When a piece of work passes, SOP exception capture records how your best person handles its awkward cases, and Pilot acceptance-test preparation sets up the cases that prove the build.
The full capability template
Everything above is the outline. The complete capability template is the working version: the roles, the inputs, the procedure, the exceptions and stop conditions, the output format, the reviewer checklist, the acceptance tests, the instructions for an AI assistant and what to customize for your company. It comes with a download and the link to its folder in Fully Powered’s public repository.
Free with an account. Sign in by email: no password and no card.
Next step
Bring a real piece of this work to a call. We’ll walk through what it takes to fit this capability to your rules, prove it on your work and put it in the hands of the roles that own it.