After AI training: how your company moves from using AI to running on it

// Part 2 of 2 A guide for owners and leaders of companies with 50 to 300 people. Part 1: how to train your employees on AI.

What this looks like in practice. A fractional HR firm we work with gave its RFP and marketing workflows owners and written checks. RFP responses went from days to hours, depending on the RFP, and marketing materials now take about 85% less time to build. A home builder we work with went from 40 hours per proposal to four.

The three stages of AI adoption (a simple AI maturity model)

1. 1. Using

   // First 90 days
   What's true

   Leaders use AI well. Champions are named. Most staff use one approved tool (Claude, ChatGPT, Copilot or Gemini). A one-page policy exists.

   Ready to move on when

   Every team has two or three repeatable AI workflows in use, and you've recorded "before" numbers for them.

2. 2. Trusting

   // Days 90 to 180
   What's true

   Each shared workflow has an owner and written quality checks. Champions build shared workflows as well as showing them. Results are tied to business measures. Leaders decide what saved time is for.

   Ready to move on when

   Leaders can name three business numbers AI has moved, and who signs off on each workflow's output.

3. 3. Redesigning

   // Day 180 onwards
   What's true

   Roles, processes and budgets are adjusted for the work AI now handles. Old processes are retired. Champions are a permanent part of how the company works. Annual plans and pricing account for AI.

   Ready to move on when

   New work is designed with AI from the start.

The timelines come from our practice with clients. No published study we've found measures how long companies take to move between stages. Firms with committed leaders move faster, and busy seasons and leadership turnover slow everyone down.

Stage two is where most companies stall. Stage three is where the biggest returns are, and it depends on stage two. MIT's Center for Information Systems Research found financial results rise sharply once AI becomes part of how work gets done, the third of its four stages, which lines up roughly with our stages two and three. In its survey of 721 companies, firms at its earliest stage had financial performance 12.6 percentage points below their industry average, and firms at its most advanced stage were 17.1 points above it (MIT CISR, published December 2024, from a 2022 survey).

How to know you've moved past early adoption

Most companies think they're further along than they are. In Section's 2026 survey of 5,026 US knowledge workers, only 5.5% were proficient with AI. Nearly three in four were still experimenting (Section, 2026).

• // Stage one

  Signs you're still here

  • AI use depends on who's enthusiastic. It isn't built into how the team works.
  • Nobody can say who's responsible when an AI draft goes out with a mistake.
  • You measure logins, prompts or "hours saved" estimates rather than results.
  • Each person has their own way of doing the same task with AI.
• // Stage two

  Signs you've arrived

  • The team's AI workflows are written down, shared and reused.
  • Each workflow has an owner who keeps it working.
  • There's a clear standard for when AI output is good enough to send to a client.
  • You can show a before-and-after number for each of your three to five tracked measures.

Who checks AI work, and how

Owners and written checks are most of what moves a team from stage one to stage two.

Unchecked AI work has a cost. In 2025 research by BetterUp and the Stanford Social Media Lab, 40% of workers had received low-quality AI-generated work from a colleague in the past month. Each case took almost two hours to sort out, and 42% of recipients saw the sender as less trustworthy (BetterUp, September 2025). In McKinsey's 2025 survey, only 27% of respondents at companies using generative AI said employees review all AI output before it's used (McKinsey, March 2025).

Here is how to set it up:

1. Give every shared workflow an owner. Usually the champion or the most experienced person in that task. They decide how it works and fix it when it breaks.
2. Write the checks down. For a client proposal, that might be: every number matches the source, client names are correct, nothing promised that we don't offer, tone matches our style. Keep it to five to ten checks.
3. Have the AI flag problems first. Save the checks as standing instructions so the AI flags likely problems on every draft. A person still checks against the list before it goes out.
4. Decide who signs off. Whoever sends it is accountable for what it says. For high-stakes work, a second person signs off.
5. Log the misses. When something gets through, add a check so it doesn't happen again. The checks improve every month.

Steps 2 and 3 are set up once and reused on every draft.

How to measure AI adoption: which numbers to track

Usage numbers tell you people opened the tool. They don't tell you whether the business is better. McKinsey found fewer than one in five companies track business measures for generative AI.

Track these instead:

• // Track

  Time per unit of work

  Measured before and after: per proposal, per client report, per month-end close.

• // Track

  Quality

  Error rates, rework, client revisions, win rates.

• // Track

  Capacity

  Clients served per person, response times, projects delivered.

• // Track

  Where the saved time went

  If you can't answer this, it probably went nowhere useful.

Pick three to five numbers your leadership team already cares about. Don't invent new metrics for AI.

A caution: faster isn't always better. Some AI-assisted work takes longer and is much better. Judge the outcome as well as the time it took.

How champions' roles change after 90 days

In stage one, champions show people what's possible. In stage two, their job changes.

This is also the time to refresh the group. Some first-round champions won't want a building role. Thank them and bring in people who do. Consider giving champions more protected time now, since they maintain work others depend on. Part 1 covers how many champions to start with and how much time to give them.

What leaders do differently in stage two

Leaders can't hand stage two to the champions. Three decisions belong to them.

1. Decide what saved time is for. In BCG's 2026 survey of almost 12,000 workers, 42% of regular AI users said they save at least a full workday a week. But 66% of regular frontline users get limited or no guidance on what to do with that time (BCG, AI at Work, June 2026). If leaders don't decide, the time quietly disappears. Common answers: more client work, faster turnaround, new services, or fewer late nights. Our article on who decides where the time goes sets out the options.
2. Set expectations and back them. In Section's survey, proficiency was 1.5 times higher where managers expected and required AI use. Yet only about 7% of managers require AI use and tie it to performance evaluations (Section, 2026). You don't need to force it, but "we expect you to use the shared workflows" should be said out loud.
3. Revisit strategy. Once proposals, reports or month-end close take far less time, some old constraints are gone. Ask the leadership team every quarter: what can we now offer, price or promise that we couldn't before?

Which processes to redesign first

McKinsey's 2025 research found that, of 25 factors tested, redesigning workflows had the biggest effect on profit from generative AI. Only 21% of companies had fundamentally redesigned even some workflows.

Redesign is different from adding AI to a step. It means asking: if we built this process today, knowing what AI can do, what would it look like? Often steps disappear, handoffs shrink and roles shift.

Start with processes that are:

• High volume: done weekly or daily, so small gains add up.
• Already measured: so you can prove the change worked.
• Close to clients or revenue: proposals, onboarding, reporting, sales follow-up.
• Owned by someone who wants to change it.

Redesign one process at a time, measure it, then move to the next. Retire the old version properly so people don't keep running both.

The AI governance a 50 to 300 person company needs

You don't need an AI committee and a 40-page framework. You need five things:

1. An approved tool list and business accounts for everyone.
2. A one-page policy: business accounts only, what client and confidential information never goes into AI, what must be reviewed, and who to ask. Our guide to AI policy for Canadian businesses covers what to put on it.
3. Named owners for every shared workflow.
4. Written quality checks for anything that reaches a client.
5. IT sign-off on what the AI can connect to, such as email, files and your CRM.

Review it every quarter. Most of it fits in one shared document.

Frequently asked questions

What is an AI maturity model?

A staged way to see how far AI has moved from personal use into how the company runs. Ours has three stages: using, trusting and redesigning.

How do you measure AI adoption?

With business numbers you already track, measured before and after: time per piece of work, quality, capacity and where saved time went. Logins and prompt counts only show that people opened the tool.

How long does it take to go from using AI to proficient?

If proficient means using AI well on everyday tasks, about 90 days in our experience, which is the end of stage one, provided leaders go first and champions have protected time. If it means shared workflows with owners and written checks across teams (stage two), about 90 more days. Redesign (stage three) takes a year or more. Leadership commitment matters more than company size.

What if usage stalls after the first 90 days?

Check the basics: is the program owner still active, is champion time still protected, and do leaders still use AI visibly? Then pick one shared workflow, give it an owner and record its before number this week.

How do we stop people sending poor AI work?

Written checks, a clear sign-off owner and the expectation that the sender is accountable. Make it normal to say "this reads like unchecked AI."

When should we bring in outside help?

When leaders don't have time to own the program, when quality problems keep recurring, or when you're ready to redesign processes and want someone who has done it before.

Where Sidekick fits

The first 90 days of Fractional AI Leadership work on stage one with your leaders: getting them proficient, putting champions in place, writing the policy and plan, and recording before numbers. A follow-on engagement focuses on stage two: workflow owners, quality checks, business measures and a leadership strategy that accounts for AI.

Book a free 30-minute call and we'll help you pick your first workflow to own and measure. Or take the free AI Opportunity Map, about 10 minutes, to see where AI fits your company.

Sources checked October 16, 2026: MIT CISR, Enterprise AI Maturity Model; Section, AI Proficiency Report; BetterUp and Stanford Social Media Lab, the hidden costs of workslop; McKinsey, The state of AI: how organizations are rewiring to capture value; BCG, AI at Work 2026 and press release.