One person's good AI routine helps one person. A written, owned, checked routine that moves across functions helps the organization. Here's how to build the second kind.
Individual AI use adds up, but shared use compounds. Turn good routines into workflow cards with inputs, a check, known misses and an owner. Use four handoff labels so AI output means the same thing in every function. Run a monthly session where people bring cards, not demos, and score the whole thing with a six-question health check built on use, persistence and impact.
Many organizations already have people who are good at AI, and much of that skill is invisible to everyone else. It lives in one person’s saved chats, a note on a laptop, a habit nobody wrote down. The company pays for the tool and gets the output of whoever happens to be good with it.
A finance lead has a way of turning raw variance notes into commentary her director will actually read. She keeps the instructions in a note and adjusts them each month. In another business unit, a colleague with the same job is building nearly the same thing from scratch, and the two of them will find out only if it comes up in a hallway.
That pattern is common. Microsoft and LinkedIn’s 2024 Work Trend Index, a survey of 31,000 people in 31 countries, found that 78% of AI users were bringing their own AI tools to work.[1] It’s a vendor’s survey about its own market, so hold it loosely. The shape still matches what many leaders see: people reach for AI on their own, build their own routines, and the organization sees none of it.
Individual use adds up. Shared use compounds. If ten people each save an hour a week, you have ten hours. If one of them writes down how, and the other nine start from her version, the next improvement gets made on top of the best version instead of beside it. I’d argue that’s the advantage hardest for a competitor to copy, because they can buy the same tool next week but not your colleagues’ accumulated, checked ways of working.
This piece is about building that on purpose: what to share, how to carry it across functions, and who owns it once more than one person depends on it. Our guides to building an AI champions network and building a prompt library cover two parts of this. Here the frame is the whole organization.
People already trade tips, in chat threads and over lunch, and that’s worth encouraging. The trouble is how much of what made a result good doesn’t survive the trade.
| What gets shared | What gets lost on the way | What to add |
|---|---|---|
| A prompt, pasted into a chat | The situation it was written for, and what the author fixed by hand afterwards | The inputs it needs and one example of a good result |
| A tip (“it’s great for summaries”) | Which summaries, and how anyone checks them | The task, named, and the check |
| A workflow, shown over a screen share | Steps get remembered differently each time, and the check is the first thing dropped | A written card: steps, inputs, check, owner |
| A standard of good | What the author would have accepted and what they’d have sent back | An example of an approved result |
| A warning, as a story | Usually never recorded anywhere | A “known misses” line on the card |
Future Factors working view of what travels between colleagues. Not a research finding.
What’s worth spreading is how your strongest people actually work, and that’s usually the part nobody has written down. A study of customer support agents found the biggest gains went to the least experienced: 34% more issues resolved per hour for novice and low-skilled workers, against 14% on average across the group.[2] The authors offer suggestive evidence that the tool carried the practices of more capable workers to newer ones.
That was one setting, with a system built to learn from top performers’ conversations, and the authors call the mechanism suggestive. But it points at the right target. The thing to move around an organization is the way your best people work, and in many workplaces nothing moves it unless someone designs for it.
If a colleague can’t reuse it without calling you, you haven’t shared it yet.
The smallest thing that passes the Travel Rule is a workflow card: one page that lets a colleague in another team run what you do, including how to tell it worked. It isn’t a prompt-library entry. A library entry stores words. A card stores the situation, the inputs, the check and the owner, which is the part that goes missing. Here’s one filled in; swap the right-hand column for your own work.
| Field | Variance commentary draft |
|---|---|
| Job | Turn the month’s variance notes into a first draft of commentary for the director’s pack (used today by FP&A analysts in one business unit) |
| Inputs | The variance table, the analyst’s one-line note on each large variance, and last month’s approved commentary as a style example |
| Steps | Paste the inputs. Ask for two sentences per large variance, using only the analyst’s notes for reasons. Ask it to list any reason it wrote that isn’t in the notes. Edit. Check every figure against the table |
| What good looks like | Reads like last month’s approved version, and every reason traces to a note |
| The check | The analyst ticks each figure against the source table. The controller reads one section at random |
| Known misses | Invents a plausible reason when a note is missing. Rounds differently from the pack |
| Must never | Be given unreleased forecasts or personal data, or go to the director before an analyst has checked it |
| Owner and review date | FP&A lead; review at next quarter-end close |
The two rows people tend to skip are Known misses and Must never. Fill those in first.
Writing one is slower than it looks, because the author has to say out loud what they’ve been doing by feel. That’s where AI can do real work. Paste in your steps, your instructions and an example you were happy with, and let it draft the card.
| Below are the steps I take, the instructions I give an AI tool, and an example of a result I was happy with for [TASK]. Draft a one-page workflow card with these fields: job, inputs needed, steps, what good looks like, the check, known misses, must never, owner and review date. Use only what I’ve told you. Where I haven’t said what the check is or what has gone wrong before, write ASK ME instead of guessing, and list those questions at the end. Do not invent statistics or examples. |
The ASK ME lines are the real work. A card with none of them answered is a prompt in a nicer layout.
Cards tend to fail in two ways. Some run so long that people go back to pasting a prompt. Others are so tidy that the Known misses row says “none,” which usually means nobody has looked. Keep the card to one page and treat an empty misses row as a reason to ask.
A sales manager forwards a call summary to the product team with a one-line note: AI summary, checked. The product lead reads “checked” as every quote verified against the recording. The sales manager meant she’d skimmed it and nothing looked wrong. Both are being reasonable. Nobody has told the other what the word means.
Shared workflows only work across functions if the words attached to AI output mean the same thing on both sides. You don’t need a glossary of AI terms for that. Agree on four labels for handoffs and put them in every card.
| Label | What it means | Fine for | Not fine for |
|---|---|---|---|
| AI draft | Generated. The sender hasn’t read all of it | Brainstorming, a rough starting point | Anything that leaves the team |
| AI draft, read | The sender read all of it and fixed what was wrong | Internal notes and working documents | Figures or quotes nobody has traced |
| AI draft, sourced | The sender traced every figure and quote to a named source | Reports and decisions | Legal or regulatory filings without a second reviewer |
| Approved | A named person owns it as if they had written it | Sending outside the organization | Nothing, but the owner’s name goes on it |
Future Factors working view. Rename them to suit your organization, but keep what each one promises.
This matters more now because AI makes it cheap for one person to step outside their own function. In a field experiment with 776 professionals at Procter & Gamble, individuals using AI matched the performance of teams working without it, and they produced balanced solutions whether they came from R&D or Commercial.[3] I read that narrowly. It says one person can now cover ground that used to need a handoff. It doesn’t say the handoffs go away. Work now moves between people as AI-assisted drafts, and the label on the draft is what the next person has to trust.
The labels land differently by role, so each card should say which label its output needs before it goes to which function:
Say what was checked, not just that AI was involved.
The first time an organization sets up an AI show-and-tell, it fills with the most impressive thing anyone built. That’s the wrong thing to reward. The impressive build is rarely the one a colleague in another function can reuse on Monday.
A better format is a monthly session where people bring cards, not demos. Forty-five minutes is enough, with people from at least three functions, or it turns into a team meeting.
Which ones got reused, by whom, and what they changed.
The author shows the card and one real output beside its source, then says where it missed.
People from other functions say which card they’ll try this month and what they’d change for their data.
The program owner promotes any card two or more people reused, and retires one.
Future Factors working view of a monthly session. Adjust the timings to the size of the group.
The second half of that agenda does most of the work. Borrowing is what turns one person’s card into the organization’s, so ask a new person, ideally from a different function, to run it before it’s called shared. Before anyone builds something new, they check the register for the closest existing card. The register near the end of this guide is where that check happens.
People trade notes better when they share a baseline. Without one, tips don’t land, because colleagues disagree about what a good result looks like. BCG’s 2025 survey of more than 10,600 workers found that those who received five or more hours of training, especially in person and with coaching, were significantly more likely to become regular users.[4]
In the terms we use at Future Factors, training builds the capability, and workflow design (the card and the register) gives it somewhere to be used. Reinforcement, which is the monthly session and a manager who asks, turns it into behavior. Adoption is what you see when that holds. Our guide on moving from awareness to fluency covers that chain in more depth.
Teams can start this internally, and many do. Expert-led workshops shorten the learning curve and make the shared baseline reach the people in finance, HR and operations who weren’t the keen ones, not only the people who already were. Our corporate workshops are built for exactly that gap, and our guide to training your team on AI covers what to teach first.
Sharing creates a problem solo use doesn’t have. When five people run one card, a mistake in it is five people’s mistake, and nobody feels responsible for fixing it. A card without an owner tends to go the same way: its author changes roles, the data source changes, and for a quarter people keep following steps that no longer fit.
AI can also carry much of the paperwork inside the sharing system itself, as long as it’s clear who decides.
| What AI does | What the human still owns | How it gets checked |
|---|---|---|
| Drafts a workflow card from a colleague’s pasted steps and a good example | The colleague fills the gaps and signs the card. The card owner decides it’s ready to share | The owner runs the card once on real work before it enters the register |
| Compares new cards with the register and flags likely duplicates | The owner decides whether it’s a duplicate or a different job, and what to merge or retire | Functional leads review the flagged list at the monthly session |
| Rewrites a card for another function, swapping in that function’s examples | The receiving function’s owner confirms the examples and the Must never row fit their data | The receiving owner runs it once on real work before it’s listed for their team |
| Turns the month’s usage notes and misses into a one-page briefing | The program owner decides which cards to promote, narrow or retire | Spot-check three cards each month against the notes the briefing came from |
Future Factors working view. Pick the rows that match work you already do.
A card involving employee or customer data should be checked with IT, security and legal before it’s shared. Sharing a card is different from sharing the data behind it.
The manager’s seat is the one people underrate. Managers decide whether an afternoon spent writing a card counts as work. In Gallup’s May 2025 survey of 19,043 U.S. workers, only 28% of employees in organizations that had begun implementing AI strongly agreed their manager actively supports their team’s use of it. Those who did were 2.1 times as likely to use AI a few times a week or more.[5] Gallup measured use, not sharing, so read it as a reason to ask managers for the habit, not proof it works.
One more thing sits under all of this. A Known misses row only fills in if posting a miss is safe. If the first person to report one gets teased, the row stays empty. Our guides to psychological safety for AI adoption and turning resistance into collaboration cover how.
Every shared workflow has one named owner and a review date.
The easy numbers are cards written and people at the session. Those measure effort. What you want is the same lens we use for any AI rollout: use, then persistence, then impact. Use is whether anyone other than the author ran the card. Persistence is whether they were still running it a few months later without a reminder. Impact is whether some real piece of work came out different.
The evidence lives in a register: one short list of every card, with its owner, who uses it and when it was last looked at. Keep it short and let its columns do the nagging.
| Card | Owner | Used by | Last reviewed | Status |
|---|---|---|---|---|
| Variance commentary draft | FP&A lead | Finance, two business units | Quarter-end close | Promoted |
| Call summary with quote check | Customer success manager | Sales, product | This month | Trial |
| Job-ad draft against the role scorecard | HR business partner | HR, hiring managers in operations | Over a quarter ago | Needs review |
| Contract clause finder | Procurement lead | Procurement | Not yet reviewed | Hold: data rules unclear |
Statuses: Trial, Promoted, Needs review, Hold, Retired. Every card has an owner and a review date, and a card without either is a rumor.
With the register in place, each seat has a different monthly question to ask of it.
| Seat | Owns | Asks or checks each month |
|---|---|---|
| Functional lead (finance, HR, operations) | The cards for their function, including every Must never row | Which cards were reused, and which have gone stale |
| IT or security | What data cards may touch and which tools are approved | Whether any card now needs data it shouldn’t have |
| Manager | The habit: time to write and run cards | In the one-to-one: which card did you use this month? |
| Program owner | The register and the monthly session | Which cards to promote, narrow or retire |
Future Factors working view. In a small organization one person may hold two seats.
Then score your organization on the six questions below, 0 to 2 each. Use the register, the session notes and a quick ask of the functional leads.
| Question | 0 | 1 | 2 |
|---|---|---|---|
| 1. Use. Did someone run a card they didn’t write this month? | No | One person | Several, in more than one team |
| 2. Persistence. Are cards from three months ago still in use without a reminder? | No | Some | Most of the promoted ones |
| 3. Crossing functions. Has a card moved across a function line with examples rewritten? | No | Once | Regularly |
| 4. Misses. Were misses reported and a card changed because of one? | None reported | Reported, not acted on | Reported and the card changed |
| 5. Ownership. Does every card have a named owner and a review date that hasn’t passed? | Under half | Most | All |
| 6. Impact. Can an owner point to a decision, deliverable or turnaround that changed because of a shared card? | No | Vaguely | Yes, with an example |
Bands: 0 to 4, a library with no system behind it (fix ownership first). 5 to 8, working in pockets (focus on crossing functions). 9 to 12, compounding. Future Factors working view, not a validated scale.
A zero on question 4 is worrying, not reassuring. It usually means nobody is checking, or nobody feels able to say so. If your score is stuck in the middle, the usual cause is that the cards exist but the session doesn’t happen, or nobody from a different function is in it.
To keep the session light, let AI prepare the briefing and keep the judgment for people.
| Below are this month’s usage notes and reported misses for our workflow cards. For each card, list who used it, which function they sit in, and what went wrong. Then list the three cards with the most reuse across functions and any card with a reported miss that has no change logged. Use only the notes. Do not estimate time saved or invent counts. If a note is unclear, list it under QUESTIONS instead of guessing. Keep the briefing to one page. |
Read the briefing against two or three of the original notes before the session. It will miss things, and the owner is the one who notices.
The first session needs an invitation that sets the bar low and the usefulness high.
| Subject: 45 minutes next month: bring one AI routine you’d hate to lose We’re starting a monthly session where people from different teams trade the AI routines that save them time, written up as one-page cards. Nothing needs to be polished. Bring one routine, one real result next to its source, and one place it got something wrong. [DATE, TIME, LINK] Please skip the impressive demo. Bring something a colleague in another team could try on Monday. |
Send it from a manager or functional lead, not a general mailbox. The ask for a miss is what makes the Known misses row fill in later.
Here’s what I’d do this month:
It can, which is why the card carries the check, the Known misses row and an owner. A card should run once on real work, with its check, before it enters the register, and a card with no reported misses after a quarter deserves a second look. Without those pieces, sharing spreads habits as they are. With them, what spreads has at least been run once and checked.
Share the pattern and keep the data out. The card names the kind of data it needs, never an example from real records, and its Must never row says what can’t go in. Check with IT, security or legal before connecting customer or employee data to any tool your organization hasn’t approved.
Not to start. One person per function who can give an hour a month is enough for a first session. A network becomes worth building when the organization is large enough that no single group can see across it, and our champions network guide covers that.
The usual reasons are credit, time and fear of looking wrong. Put the author’s name on the card, treat writing it as work, and let a manager go first with a miss of their own. Our piece on turning resistance into collaboration goes deeper on the fear.
A library stores prompts and a knowledge base stores reference material, as in our guides to prompt libraries and team knowledge bases. The cards, labels and session here are about moving how work is done across functions, with an owner for each. Many organizations will use all three.
Sources were read on October 16, 2026: Microsoft and LinkedIn’s announcement of the 2024 Work Trend Index, the NBER working paper pages for “Generative AI at Work” (Brynjolfsson, Li and Raymond) and “The Cybernetic Teammate” (Dell’Acqua and colleagues, abstract only), BCG’s June 2025 press release on AI at work, and Gallup’s November 2025 article on manager support. Microsoft and BCG sell products and services in this space, and the Cybernetic Teammate and support-agent studies are single settings, so treat all of them as evidence to weigh, not rules. The workflow card, handoff labels, session format, register, ownership table and health check are Future Factors working views based on our own judgment, not research-validated models. The scenes in this guide are illustrations, not client stories. We left out ROI, hours-saved and adoption-rate figures because we could not confirm any against a primary source. This is not legal advice.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.