Everyone on your team has a login. Roughly a third of them haven't opened it this week. Here is the difference between announcing a tool and actually driving adoption.
Marketing has adopted AI faster than most functions: The CMO Survey found 73.9% of marketing teams now use AI for content creation, up from 49.2%[1]. But the same research concludes that “technology adoption is outpacing organizational readiness” and that the barriers are “decidedly organizational”[1]. Meanwhile 30% of people who use AI at work hadn’t touched it at all in the previous week[2]. The sequence that works is train the skill, redesign the workflow, reinforce the behaviour, measure the work. This is the practical playbook for the last three: why announcements fail, how to pick and rebuild the first workflow, how to make the first win copyable, the review loop most teams skip, what to do about the quiet holdouts, and a 30-day plan.
Here is the rollout I’ve now watched maybe a dozen times, in my own teams and in clients’. Someone signs the contract. IT provisions the seats. A nicely written message goes into the marketing channel with a login link and a sentence about how this is going to free everyone up for higher-value work. Two or three people reply with the rocket emoji.
And that’s the strategy. That’s the whole thing.
Three weeks later your usage report shows four heavy users, a long tail of people who logged in once, and a handful who never activated at all. The four heavy users were going to use it anyway. They’d been paying for it personally.
The numbers back up what you’d guess from watching it happen. When the US Census Bureau asked workers who use AI on the job how often they’d actually used it in the previous week, 24% said every day, 46% said at least one day but not every day, and 30% said they hadn’t used it at all[2]. Nearly a third of people who count as AI users had a week where they didn’t touch it.
Marketing is genuinely ahead of most functions here, so I don’t want to be unfair about it. The CMO Survey is run out of Duke’s Fuqua School and surveyed 308 US marketing leaders in January 2026. It found AI use in marketing has more than tripled since 2022: 73.9% of teams now use it for content creation, up from 49.2%, with 65.4% using it for content personalization and 48.9% for marketing automation[1].
But read their own summary of what that means: “AI is accelerating and delivering, but adoption is outpacing organizational readiness… The barriers are decidedly organizational, factors such as budget, integration, bandwidth, and talent dominate”[1]. No marketing technology activity in their survey scored above 5 on a 7-point performance scale[1].
Prosci’s research across more than 2,600 change practitioners found that 88% of projects with excellent change management met or exceeded objectives, against 13% with poor change management, roughly seven times more likely to succeed[3]. And for anyone who thinks structure slows things down, the same research found projects with excellent change management were nearly five times more likely to be on or ahead of schedule[3].
This is Future Factors’ own framework and it’s the clearest way I know to explain why the announcement approach produces nothing:
Tool x Workflows x Behavior = AI-powered professional.
Think of it as multiplication rather than addition. A weakness in any one part drags down the whole result, and no amount of strength in the other two compensates for it. That is why a licence on its own, with nothing else changed, returns so little.
Tool is the licence and the access. It’s the only variable you can complete by spending money, which is exactly why it’s the one that gets completed first and celebrated loudest. A Slack announcement with a login link sets Tool to one and leaves the other two at zero.
Workflows is whether a specific, recurring, named task your team already does has actually been rebuilt around the tool. Not “you could use this for social captions if you want.” The social caption process, redesigned, written down, with the old way retired.
Behavior is whether that new way survives a week where three things go wrong and a client moves a deadline. Under pressure people revert to whatever is automatic, and if the AI-assisted version requires a decision, it loses.
| What you did | Tool | Workflows | Behavior | Result |
|---|---|---|---|---|
| Announced the tool, sent the login link | Yes | None | None | Zero |
| Announced it and ran one training session | Yes | None | Brief spike | Zero |
| Rebuilt one weekly task, no reinforcement | Yes | One | Fades | Small, then decays |
| Rebuilt one task, retired the old way, reviewed weekly | Yes | One | Becomes automatic | Compounds |
Future Factors’ adoption framework applied to four common rollout patterns. This is an author’s framework illustrating a multiplication relationship, not measured survey data.
What makes me confident this isn’t just a neat metaphor is that Microsoft’s research lands in almost the same place from a completely different direction. Its 2026 Work Trend Index found that organizational factors like culture, manager support and talent practices account for more than twice the reported AI impact of individual factors like mindset and behavior, 67% versus 32%[4]. Microsoft notes these are statistical associations rather than causal effects, which is a caveat worth carrying, but “organizational” there is largely another word for Workflows.
The same research identifies a group it calls “blocked agency”: 10% of AI users who “have built strong skills but lack the systems to apply them”[4]. If you’ve trained your team and seen nothing come back, that’s very likely who you’re looking at. Skilled, willing, and producing nothing because nobody changed the work. More generic tool training won’t solve that. The next learning intervention needs to move into the work itself: specific workflows, manager expectations, quality standards, and practice.
| Stage | What it looks like for that one task |
|---|---|
| Announced | People know it exists. Nobody’s actual task has changed yet. |
| Tried once | A few people used it for this task once, out of curiosity, then went back to the old way. |
| Rebuilt | The process has been redesigned and written down, with a quality bar someone can check against. |
| Defaulted | It’s the standard way this task gets done now, but it drifts back the moment a deadline gets tight. |
| Held | Still the default at 90 days, without anyone pushing it. |
A workflow-level tracker for the one task from this section, not a company-wide capability ladder. Author’s model for locating one workflow, not measured survey data.
The instruction “use AI where it makes sense” sounds empowering and is functionally useless, because it puts the design work on the busiest person in the chain.
Pick one task. One. Then rebuild it properly.
Four criteria, in this order of importance. There is a scored version at the end of this section if you want to compare candidates properly.
It happens weekly or more. Monthly tasks can’t build a habit, because by the time it comes round again everyone has forgotten the new way. This rules out a lot of tempting candidates like the quarterly report.
Someone currently dislikes doing it. Motivation is free if you pick something people resent. Repurposing a long-form piece into channel variants, writing the fifteenth ad variation, turning campaign numbers into a summary paragraph for the weekly.
Quality is checkable in under two minutes. If nobody can tell quickly whether the output is good, you get endless debate instead of adoption. Ad copy variants: checkable. Brand strategy: not.
It isn’t the most sensitive thing you do. Don’t make your first workflow the one that touches customer data or goes straight to a client without review. You want early wins boring and safe.
For most marketing teams this lands on one of about four things: repurposing long-form content into channel-specific variants, generating ad copy variations for testing, drafting the first pass of social captions from a brief, or turning campaign performance data into a written summary.
| Score one point for each |
|---|
| It happens weekly or more often |
| It takes a meaningful amount of time |
| Someone genuinely wants relief from it |
| The output can be checked in under two minutes |
| The consequence is low if the first version goes wrong |
| The inputs are available in an approved environment |
| Result |
| 5 to 6: strong first candidate, start here |
| 3 to 4: workable, but test carefully and expect friction |
| 0 to 2: pick something else, this one will not build a habit |
Score two or three candidate tasks and compare. An author’s selection tool, not measured data.
This is the step that separates a real rollout from a nice announcement, and it takes about ninety minutes with the people who actually do the task.
Sit down with them and write the new version of the process end to end. What the input is (a brief, a doc, an export). What the prompt is, word for word, saved somewhere shared rather than living in someone’s head. What the output looks like. Who checks it and against what. What happens to the old way.
That last one matters more than the rest combined. Once the new workflow has been tested and the quality bar is clear, retire the old default. Keep a fallback only where risk or business continuity genuinely justifies one. The point is not to strip out your safety nets on day one, it is that an optional new method stays optional forever. If both routes remain equally available, everyone under deadline pressure takes the familiar one, and you will conclude AI did not work for your team when what actually happened is you never retired the alternative.
Write the prompt down as a team asset, not an individual one. Our guide to building a prompt library your team actually uses covers doing this without creating another document nobody opens, and the twelve marketing workflows piece is a decent shortlist to pick your first candidate from.
The first genuine win often does more for adoption than another announcement, because people can finally see what the new way actually looks like. Most teams let that win happen privately and then wonder why nothing spread.
Someone on your team is going to do something genuinely good with this in the first fortnight. Your job is to make sure everyone sees it, in enough detail to copy.
Not “shout out to Dan for using the new tool.” That tells nobody anything and mildly embarrasses Dan.
Visible means the mechanics. The actual prompt he used. The actual output. What he changed about it before it shipped. How long the old version took and how long this took. Five minutes in a team meeting, or a post in the channel with the prompt in a code block.
The detail is the whole point. People copy processes they can see. They don’t copy praise.
This is the counterintuitive one and I’d argue it’s more important than the wins.
When AI produces something confidently wrong for your team, show it early, rather than months later once everyone has quietly decided they are supposed to pretend the tool works perfectly. Put it in the channel. The hallucinated statistic, the client name it got wrong, the caption that was subtly off-brand in a way nobody could quite articulate. Two things happen. Your sceptics stop feeling like they’re the only ones noticing problems, which is what usually pushes them into quiet opposition. And your enthusiasts calibrate, which is the difference between a team that uses AI well and one that ships slop.
Microsoft’s data on manager behaviour is unusually specific about this mechanism, and it is worth being precise about where it comes from. These figures are from the Microsoft People Science Agentic Teaming and Trust Survey of July 2025, covering 1,800 employees globally: 819 leaders, 520 managers and 461 individual contributors[4]. That is Microsoft’s own research team measuring a market Microsoft sells into, so I would trust the direction of travel more than the decimal places.
With that caveat: the pattern holds across the board rather than showing up as one isolated number. Employees whose managers visibly used AI themselves reported double-digit gains in how much they valued it, how critically they thought about their own use, and how much they trusted it[4]. Employees whose managers made experimentation feel safe reported a similar gain in readiness, and were meaningfully more likely to be frequent users rather than occasional ones[4].
The reason I am willing to lean on a vendor’s own numbers here is that the independent evidence points the same way. Prosci’s change-management research and Cisco’s readiness index both reach the same conclusion from outside the AI industry: the organisational scaffolding around a tool predicts the outcome better than the tool does[3][5].
Two findings from that data matter most here. Microsoft’s most advanced AI users are far more likely than everyone else to say their manager openly uses AI (85% versus 64%), and far more likely to say their manager sets quality standards for AI work (83% versus 57%)[4].
The interesting part is not that managers make people more enthusiastic. It is that managers define what good AI-assisted work looks like. That is adoption work, and almost nobody has it in their job description.
On a marketing team, setting that standard is the single most useful thing a head of marketing can do, because without it every AI-assisted deliverable turns into an argument about taste.
| Behaviour | Why it moves adoption |
|---|---|
| Use the workflow visibly | Permission is granted by example, not by announcement |
| Set the quality bar | Without it, every AI-assisted deliverable becomes an argument about taste |
| Protect time for experimentation | People will not learn a new method in the gaps between deadlines |
| Ask what failed | Makes problems reportable instead of quietly fatal |
| Remove blockers | Most quiet holdouts are stuck, not resistant |
| Review whether the workflow stuck | Adoption decays silently unless someone checks |
| Avoid rewarding raw usage | Rewarding prompt counts produces compliance behaviour, not better work |
The manager behaviours that carry the most weight, drawn from the research above and from running these rollouts. An author’s checklist, not measured data.
Here’s a question worth sitting with: on your team right now, how does someone find out whether the AI-assisted work they produced was any good?
For most teams the honest answer is that they don’t, unless it was bad enough that someone complained. That’s not a feedback loop. That’s a complaints process, and people optimise around it by using AI only for things nobody checks.
The gap is well documented. Cisco’s 2025 AI Readiness Index, based on 8,039 senior leaders at organisations with 500+ employees, found only one in three companies have a formal change management plan to guide employees through AI adoption, and only 32% have a process to measure the impact of their AI initiatives[5]. The Conference Board reported in July 2026 that 55% of workers regularly use AI, but only a third had employer-provided AI training in the previous six months, and 28% get none at all. Fewer than half say they have sufficient time (48%) or tools and access (48%) to develop AI skills[6].
A quality bar, written down, one page. What does acceptable AI-assisted copy look like here? It takes about an hour to write and it ends most of the taste arguments. Here is a version you can copy and cut down.
| Before any AI-assisted work ships |
|---|
| Every statistic has a source someone actually opened |
| Every customer quote is verbatim, or clearly labelled as paraphrased |
| Claims about a product trace to approved source material |
| Brand voice is checked against real examples, not against a description of the voice |
| A named human reviewer owns the final decision to ship |
| Confidential and customer data stayed inside approved systems |
A starting template to adapt, not a compliance standard. Cut it to the four lines your team will actually check.
Fifteen minutes in the weekly. One standing slot: what did you use it for, what did it get wrong, what would you tell someone trying this next week. Not a report. A conversation. It doubles as the reinforcement that stops the new workflow decaying.
A shared prompt file that gets edited. When someone improves a prompt, the improvement goes back into the shared version with a note on what changed. Most teams’ prompt libraries are write-only, which is why they die. The edit history is the actual asset.
One caution on the ROI conversation, because it will come. Census data on self-reported time savings is more modest than the marketing around these tools suggests. Of workers asked, 25% said AI saved them less than an hour and 31% said one to two hours. Another 15% said three to four hours and 15% said more than four. At the other end, 10% said it saved no time at all, and 3% said using it actually cost them time[2]. Those savings are useful, but hours are only one part of the value, and per person they are modest. The better question is what the team can now do differently because that capacity exists, which is where quality, throughput and decision speed show up. Either way, promise your CFO a modest number you can defend rather than a big one you can’t.
Every team has them. They didn’t object in the meeting. They attended the training. They have never once opened it.
The instinct is to treat this as a compliance problem and start reporting on individual usage. Please don’t. You’ll get compliance behaviour, which looks like adoption on a dashboard and is worth nothing, and you’ll spend credibility you’ll want later.
In my experience quiet holdouts fall into four groups, and they need completely different responses. The mistake is treating them as one group with one attitude problem.
| Type | What you’ll hear | What’s really going on | What works |
|---|---|---|---|
| Blocked | Nothing. They go quiet. | Access issue, wrong licence tier, or a feature that needs an admin setting | Sit with them for ten minutes and watch. Usually fixed same day. |
| Burned | “I tried it, it wasn’t great” | One bad early output, often from a vague prompt with no source attached | Show a failure publicly first, then pair them on one specific task |
| Craft-protective | “It doesn’t sound like us” | Often correct, and they’re your quality signal | Put them in charge of the quality bar. Genuinely. |
| Job-anxious | Vague agreement, no action | Worried this is a prelude to headcount conversations | Answer the actual question honestly. Nothing else works. |
Author’s typology based on running marketing teams through AI rollouts. Illustrative categories, not survey data.
A few notes on the harder two.
The craft-protective people are your best asset and everyone treats them as an obstacle. The person who says the output doesn’t sound like your brand is usually right, and they’re the only one on the team who can articulate why. Making them the owner of the quality standard converts your loudest sceptic into the person enforcing good use of the thing.
The job-anxious group needs a straight answer, not reassurance. If you don’t know what this means for the team’s size in a year, say that. If you do know, say that. What people cannot work with is warm ambiguity, and they will read evasion as bad news and act accordingly. Microsoft’s finding that 65% of AI users fear falling behind if they don’t adapt quickly, while 45% say it feels safer to focus on current goals than to redesign work with AI[4], describes exactly this bind. They’re anxious about being left behind and simultaneously certain that experimenting is the riskier move.
Which is a rational read of most workplaces. Only 13% of AI users say they’re rewarded for reinventing work with AI when results don’t immediately follow[4]. If you want people to try things, the reward for a failed attempt has to be visible at least once, and it has to happen before you ask for the second attempt.
Thirty days, one workflow, no new tools. This is deliberately narrower than most rollout plans, because narrow is what works.
Pick one weekly task and rebuild the process with the people who do it. Write the prompt and the quality bar down. Announce only at the end of the week, after the work is done.
Make the new workflow the default for that task. Use it visibly yourself. Share an early failure as well as the first win, and sit with anyone who has gone quiet.
Make the win copyable: the actual prompt, the before and after timing. Start the standing fifteen-minute slot. Pick the second workflow but do not start it.
Check honestly. Seven-day active users as a share of people with access, and whether the rebuilt task has drifted back. Only add the second workflow if the first one held.
The author’s recommended 30-day sequence, described in full below. This is a recommended plan, not measured data.
If you take one thing from this, take the sequence: train the skill, redesign the workflow, reinforce the behaviour, measure the work. Adoption is not something you drive by talking about AI more. It is what you get when all four of those happen to one specific piece of work, and then you refuse to let it drift back. The 30-day plan above is simply that sequence run once, at the smallest scale that still proves something.
If you want the structured version of this rather than running it yourself, our Corporate Workshops and AI Bootcamps are built around exactly this, redesigning real workflows with the team that does the work rather than presenting features. For the training side specifically, our step-by-step playbook for training your team on AI covers running the sessions, and if you want to know where your team’s gaps actually are before you start, the 20-minute skills gap diagnosis is a good first hour.
Because access is only one of three things adoption needs, and they multiply rather than add. Future Factors frames it as Tool x Workflows x Behavior: the licence is the Tool, a specific recurring task genuinely rebuilt around AI is the Workflow, and reaching for it without being reminded is the Behavior. A rollout that provides access and nothing else sets two of the three to zero, so the result is zero. Microsoft’s 2026 research points the same way, finding organizational factors account for more than twice the AI impact of individual ones, 67% versus 32%[4].
Pick one task that happens at least weekly, that someone currently dislikes, where quality is checkable in under two minutes, and that isn’t your most sensitive work. Then spend ninety minutes with the people who do it rebuilding the process end to end: the input, the exact prompt written down and shared, the output, who checks it and against what. Then retire the old way explicitly. If both routes still exist, anyone under deadline pressure takes the familiar one, and you will wrongly conclude AI did not suit your team.
It states that Tool x Workflows x Behavior = AI-powered professional. The multiplication signs matter: there is no partial credit, and one zero produces a zero overall. Tool is the licence and access, the only part you can buy. Workflows is whether a named recurring task has actually been redesigned. Behavior is whether that new way survives a bad week, since under pressure people revert to whatever is automatic. It explains why a Slack announcement with a login link produces nothing, and why a single training session produces a spike that flattens.
Stop treating them as one group. Quiet holdouts usually fall into four groups. The blocked have an access or licence problem, fixed in ten minutes of sitting with them. The burned had one bad early output, and are won back by pairing on a specific task after you have publicly shown a failure of your own. The craft-protective are often correct, and are best converted by putting them in charge of the written quality standard. The threatened are worried about headcount, and only a straight answer works. Individual usage reporting produces compliance behaviour that looks like adoption on a dashboard and is worth nothing.
About thirty days to get one workflow genuinely embedded, and two to three quarters to have four or five running properly. Expect the second workflow to be roughly three times easier than the first. The test that matters at day 30 is not the licence count but whether the rebuilt task is still running the new way, and the share of people with access who used it in the last seven days. Prosci’s research found projects with excellent change management are around seven times more likely to meet objectives and nearly five times more likely to be on schedule[3].
This playbook comes from running marketing teams through AI rollouts, combined with primary research from The CMO Survey (Duke Fuqua), the US Census Bureau, Microsoft’s Work Trend Index, Prosci, Cisco and The Conference Board. Every figure was traced to the publishing organization’s own report rather than to secondary coverage. Self-reported time-saving figures are included at their real, modest size rather than the larger numbers vendors quote, because a defensible small number is more useful to you than an impressive one you cannot stand behind.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.