Two AI Rollouts Ran the Same Playbook. Only One Kept Growing.

The AI rollout you funded this year will be decided by someone who was not in the room when you funded it: the direct manager of the team that has to use it.

Marie Rodgers, contributing author of this piece, led the enablement work at two mid-size companies that put AI into daily work with substantially the same playbook. Custom tools built for real tasks. Hands-on training. Peer champions. Drop-in office hours. Checkpoints written into the process. Both companies had an executive sponsor who did every set piece asked of them. Both teams had skeptics in the room.

At the first, a vertical-market technology provider, the sales team cut call prep from 25 minutes to 5. The gains held for the people who engaged. Then the rollout stopped spreading, and a meaningful share of the team never changed how they worked.

At the second, an industrial manufacturer, 12 people trained over eight weeks. They built two AI assistants with the program and at least five more on their own, without anyone asking. The person who arrived with the least AI experience was one of the most engaged users within weeks.

The variable was not the technology, the training, or the sponsor. Both rollouts had all three. It was whether the direct manager owned the change with their team, and whether anyone had set that manager up to win. That is the risk sitting inside your rollout right now, and you cannot see it from the approval meeting. A manager who nods and a manager who owns it look identical in that room.

More training has a ceiling, and one team found it

When a funded rollout is not sticking, the reflex is more training and firmer mandates. The technology provider did exactly that for months, so you can see where it stops.

The tools were good, and they got better. The revenue team of about a dozen had little formal sales process, qualification framework, and very few templates. The new AI tools were built into the sales process: a pre-call research planner, AI meeting summaries for discovery and demo calls, and structured notes ready to drop into the CRM. The AI helped build the process and sped it up at the same time.

For the people who used it, it worked. Average prep time for call planning, discovery, and demo prep dropped from 25 minutes to 5 minutes per instance. At real usage volume that is around an hour saved per person, per week. The two tools logged 200 net hits in the first eight weeks, up from zero. Output quality became far more consistent. A team-member champion, trusted by both middle and senior management, became the reference point the rest of the team measured themselves against.

Over roughly six months, including a full relaunch that Marie joined for, the rollout added, one after another:

1. A peer champion program. More people to copy.

2. Mandatory hands-on workshops. Nobody could skip the learning.

3. Drop-in office hours, reference guides, and a feedback channel. Help on demand.

4. Public win celebrations and a habit tracker. Usage made visible.

5. A CRM rule, last of all. The AI summary became mandatory to advance a deal.

The later versions of the tools were well ahead of the early builds, so some of the early friction was the tools maturing, not people resisting. That is worth saying plainly before anyone reads the rest as a story about a reluctant team.

The executive sponsor did their part: delivered the launch message, approved the mandate, and publicly backed the peer champion.

The direct manager never took ownership. Not out of hostility. It was daily behaviour that never showed up. The manager was not visibly present during training, did not learn the tools, and did not coach. The 1:1 check-ins and pipeline-review reinforcement designed for the manager to run, by every indication, never happened.

The result is a ceiling, not a failure. Real, lasting gains for the people who engaged. Enforcement that lapsed when the engagement wound down. A meaningful share of the team that never shifted behaviour. Everything the relaunch added was compensating for one person who never took full ownership.

The manufacturer's manager was inside before the tools existed

Same playbook at the industrial manufacturer, with one difference in sequence.

The manager was inside the program before there was anything to approve. They mapped their team's own workflows. They personally supplied the example documents the two AI assistants learned from: one assistant for marketing content, one for corporate documentation. A leader and a team member stood both up together, rather than receiving them finished.

Gartner finds organizations are 14 times more likely to succeed at a change when employees help build it rather than being told what is coming. That was the sequence at the manufacturer, and the manager kept going after launch:

1. Delivered the "why" personally, instead of leaving it to the enablement team.

2. Kicked off the training and called out, live, where the tools would matter for this team's work.

3. Checked in with the enablement team without being asked.

4. Kept visibly building their own use of the tools.

5. Carried reinforcement week over week in 1:1s. The same reinforcement the technology provider designed and never ran.

A senior team member became proficient well past the training minimum, specifically to model the behaviour for the group.

The training itself was modest. Twelve people over eight weeks, across marketing, HR, and IT: one two-hour in-person workshop, three 30-minute office-hours sessions, and one self-paced module. Office hours were optional and still drew seven or eight people every session.

What happened next is what the technology provider never got. Usage spread past the prescribed use cases. People invented their own applications. At least five more AI assistants were built without anyone asking. The person who came in with the least AI experience in the room was one of the most engaged users within weeks. Not everything worked: one person's attempt at a complex multi-part report failed and was finished by hand.

The evidence behind this story is thinner than at the technology provider. The enablement team did not have the same monitoring access at the manufacturer, so there is no usage survey and no system data. The stronger result has the least measurement behind it. Read it as momentum, not metrics, and do not carry the technology provider's numbers across to it.

Three things a manager needs, and approval is none of them

Set the two managers side by side, and the difference reduces to three conditions.

1. Desire. The manager sees the initiative as a benefit and a priority for their own team, not one more thing being done to it.

2. Working knowledge. The manager understands the tools well enough to coach them, not just endorse them. That means having used them on their own work.

3. Line of sight to the business result. The manager can see past the time saved to where the recovered hour goes and what result it should produce.

A willing manager without the second and third stalls almost as reliably as an absent one. Enthusiasm at the kickoff is not coaching in the 1:1.

Gallup put a number on how much the manager matters back in 2015 and still stands behind it: managers account for at least 70% of the variance in employee engagement across business units. AI adds a second requirement on top, because the manager now also has to learn a new tool well enough to coach it. Only 8% of HR leaders believe their managers have the skills to use AI effectively, in a Gartner survey of 114 of them in 2025. The unready manager is the base case, not the exception.

Equipping them is a design requirement of the rollout. Almost everything else in the playbook can be supplied from outside. This cannot.

What two rollouts can and cannot tell you

Two rollouts is a field observation, not a controlled study. This is the pattern Marie observed. The two engagements were not matched: the technology provider ran roughly six months with a relaunch, on a team with little formal sales process underneath, while the manufacturer ran eight weeks of structured training across three functions on content and documentation work. And nobody assigned the managers. Manager readiness was observed, not randomized, so the comparison cannot fully separate the effect of equipping a manager from the effect of already having a good one.

You cannot tell in advance which kind of manager you have, and the approval meeting will not tell you. So the only move available is to equip them deliberately before launch rather than hope. Both teams had skeptics. Neither started as a room of enthusiasts. The manufacturer's manager did specific, repeatable things: they were inside the build, and then they coached.

Change the sequence before the next launch

Three changes follow from the two rollouts.

1. Equip the manager before the build, not after the launch. The manufacturer's manager mapped workflows and supplied documents before a tool existed. Everything the technology provider added afterward was compensating for that sequence being reversed.

2. The executive sponsor must hold managers accountable for reinforcement. The 1:1 and pipeline-review check-ins existed on paper. Nobody held the manager to running them.

3. Empower the manager to change the workflows as the team learns. A mandate moved behaviour at the technology provider, but only while someone enforced it. When the engagement wound down, enforcement lapsed with it. The manufacturer's team kept refining outputs and building new assistants, with a manager who was inside the work from the start. A change the manager can adjust can keep growing after the launch team leaves.

Four questions to answer before you launch

Before the next AI rollout goes live, or before this one gets a second budget, answer these about the direct managers of every team it touches.

1. Do they see it as a benefit and a priority, or as one more thing being done to their teams?

2. Do they understand the tools deeply enough to coach them, and have they used them on their own work?

3. Who is responsible for equipping them before launch, and who checks that it happened?

4. Can they say where recovered capacity will go, and what business result it should produce?

If you cannot answer all four, the initiative is not ready to launch, no matter how good the technology is.

If the workflow has not been chosen yet, that decision belongs in the same room as the manager. An AI Jumpstart is three weeks, led by your executives, that picks the first workflow worth redesigning with AI and ends in a go/no-go with the economics attached. Put the direct manager of the team that owns that workflow at the table from day one. Architech advises on where AI pays off, builds it, and runs it in the live operation, the same team across all three.