In the final of our AI webinar series, we walk you through how to successfully roll out AI to a business, without it failing as just another IT project.
Welcome everyone, and thank you for joining the fourth and final session in this AI webinar series.
In our earlier sessions we covered AI basics, how to prepare your business for AI, and we took a deeper look at what Copilot can do.
Today is about the practical rollout of AI: how a small business can move from interest and experimentation to safe, consistent, everyday adoption.
By the end, you should have a clear 90-day plan you can take away and start using.
Here is the route we will follow over the next 30 minutes. We will start with why AI rollouts fail, then move through the adoption journey. We will finish with a 90-day rollout plan and time for questions.
I’m probably not going to tell you anything earth-shatteringly new, but sometimes it just helps to have someone remind you about stuff you already should know.
Let’s start with talking about why AI rollouts fail.
They often do. There’s plenty written about this at the moment.
The issue is rarely the tool. It is usually the adoption model.
There’s a common pattern that has emerged: the business starts with excitement. A few people try it, there are some impressive early examples, and there is a feeling that this could change how work gets done.
But very often, the organisation never gets beyond experimentation.
There is no follow-up or reinforcement of the messaging. The platform is not built out to support and encourage use.
People may use it sporadically, but not in a meaningful and measured way.
Leaders then struggle to prove value because usage is patchy and inconsistent.
The project falls flat on its face, costing money and giving no return.
This is the headline for today. Your takeaway.
AI does not create value when you buy it; it creates value when your people use it consistently.
Rolling this out is a people thing, not just a tool thing. It is really about people management: encouraging staff to engage with AI and helping them build it into their day-to-day work.
The technology matters, but adoption happens through your teams, your processes and your habits.
If people do not know where AI fits into their day, they will not keep using it.
How do you encourage them to use it?
The way to ensure your rollout is successful is to follow this four-step adoption journey:
1. Build a foundation
2. Run a pilot
3. Embed habits
4. Scale it
As the slide says, the earlier webinars covered everything in the foundation step in detail. This one is all about steps two, three and four.
The pilot stage is about setting up a limited group who try the pre-identified use cases, supported by champions, and gather examples of what works.
The habits stage is about attaching AI to repeatable tasks they stick. AI becomes the way of doing things, with human review and simple measurement.
Finally, the fourth stage, scaling, is about extending what works to wider teams, reviewing governance, and choosing the next wave of use cases.
Although we have covered step one extensively in previous webinars, I just want to touch on those use cases again.
I’ll quickly recap that because it has an impact on your rollout success.
You want to identify bottlenecks: repeated admin, delays, duplicated effort, and on. Once those bottlenecks are visible, choose the quickest and lowest-risk opportunities where AI can help immediately.
Those wins help drive excitement and build traction with people.
Good use cases to start with include drafting the first email response, summarising meetings, and creating first-draft documents. Those quick wins prove the value, help people build confidence, and give you evidence before you automate more of your processes.
Avoid starting with large, business-critical, end-to-end automation projects. They will fail.
Let’s move onto the first of the main topics for today: the pilot.
You need to figure out who is going to be the guinea pig.
You want to identify a small group of users, maybe even just one team to start off with. That will vary depending on your size, of course. Pick a pilot team that you think can benefit from the simple quick wins you identified.
Then you want to appoint your AI champions.
They do not need to be AI experts or technical specialists.
In a small business, the best champions are often the people who like trying new things, are comfortable sharing examples, and enjoy helping colleagues.
They do not need to be technical. They need to be passionate, enthusiastic people who are happy to help others. They are your superheroes in this process: the people who can train, engage and keep the momentum going.
Ideally, if you are big enough, choose one champion per department or function: finance, sales, service, operations and leadership.
Their job is not to police AI use. Their role is to share tips, gather feedback, test new capabilities and promote success stories.
As the slide says, people learn best from colleagues. You can write as many policies or guidance notes as you like, but people are often more likely to pay attention when a colleague says, “I used AI for this, and it worked really well.” Those practical examples help create engagement and excitement.
Champions also become the natural owners for the prompt bank and other AI-related assets, which we’ll come onto in due course.
You want them to be your AI evangelists, spreading their enthusiasm for it and bringing everyone else along for the ride.
You’ve identified what you want to do with AI, and you’ve identified your champions. Next, you need to help set them up to succeed.
To really get AI working for your pilot, you need it to move from giving generic answers to useful, business-supported answers specific to you. To get there, you need to give it context.
There are two levels of context to think about.
At company level, the organisation decides the approved sources, terminology, tone, policies and instructions that should be common for everyone. This is the company-wide context you want AI to refer to when people ask it to do work.
Unfortunately, you cannot set these to apply at a global level. Each individual user will need to specifically tell Copilot to always reference that repository when answering or carrying out work for them.
This same process can be done at the individual level. You want to tell Copilot what your role is, what your responsibilities are, how you like to interact with people, and anything else you feel is important that will help it work better for you.
The simple message is that company context sets the baseline, and personal context makes the output from AI reflective of the person asking it to do something.
How do you actually do that?
I’ve got a couple of screenshots here from Copilot.
The first is to tell you about Remember.
Remember is how you tell Copilot to make memories. This stores information in Copilot’s memory, just like a real person. It moves it from being a transient piece of information in a chat to a fixed one: a memory.
Just start your interaction with “Remember” and then give it the information. For example: “Remember to always reference the IT Foundations tone of voice guide when drafting any response or document.”
You can also use it to get Copilot to remember your context. For example: “Remember I’m the chief wrapping elf at the North Pole.”
Incidentally, if you want Copilot to forget a memory, you can just ask it to forget it. You can also ask it what memories it has stored. Or, if you prefer the old-school way, you can go to the three dots at the top right, select chat settings, personalisation, then manage saved memories.
You might also want to create different pieces of context for different types of work: how you write customer emails, how you write blogs, how you respond internally, or how you respond externally. You do not need to have just one memory; you can create several and ask Copilot to refer back to them.
The other action illustrated here is saving prompts.
Staff should save prompts they use regularly, especially if they are long, detailed or particularly useful. That way they do not need to keep typing them out each time.
To do that, simply run the prompt, then hit the save banner at the bottom right and it will add it to your library.
How do I access the library?
Open Copilot and use Prompt Lab.
With Microsoft 365 Copilot, users can save and reuse prompts through Prompt Lab.
To access it, open Copilot, click the three dots, open Prompt Lab, and select from your saved prompts. You can also access prompts saved at team level, which is especially useful where a whole team should be using the same prompt regularly.
If you do not want to store the prompt bank in Copilot itself, use OneNote, SharePoint, a Teams channel, or another shared location. The exact location matters less than making sure people know where to find it quickly.
Start with around 10 prompts linked to real workflows. Each prompt should include the purpose, what input to provide, what good output looks like, who owns it, and what a human must review.
If you want more on prompting, take a look at the last webinar when Pax8 touched on this topic for us. We might do another deeper dive into prompting in the future.
Now you have your pilot group, they know what they are using AI for, and they have given the AI context.
The next step is to embed it. Make it a natural part of people’s days.
The habit of using it sticks best when AI has a clear trigger, an AI-assisted step, a human review point, an output and a measurement.
Take customer emails as an example. The trigger is the incoming message. AI helps create the draft response. A person reviews tone and accuracy. The reply is sent. The business can then measure whether response times improve.
For meetings, the trigger is the meeting ending. AI creates a summary. The owner confirms actions. The summary is shared.
To encourage this behaviour with the pilot group, be specific with the language you use when briefing them and getting them started.
A common mistake is to tell people simply to use AI. That instruction is too vague. Adoption improves when AI is attached to a specific workflow.
For example, instead of saying “use AI”, say “use AI to draft the first response to every customer email.”
Instead of saying “take better notes”, say “use AI to summarise every management meeting.”
These are behaviour-based prompts. They tell people when to use AI, what to use it for, and what output is expected.
Measurement is what turns AI from a novelty into a business case.
It is easy to think of measurement as just counting the number of minutes saved, but it is worth measuring adoption as well as outcomes. If only one person is truly using it frequently, you will be able to see that and factor it into decision-making.
Regardless of how you measure it, start by recording a baseline for the activities or workflows you plan to bring AI into. How long does it take to write emails? How long does it take to write up meeting notes? Forgetting that step will derail all your metrics, and you will not be able to get a good handle on your ROI.
For Microsoft Copilot, use products like Viva Insights, included with Business Premium, for reporting on usage and adoption. Also, do not forget to gather feedback from champions.
For other platforms such as Claude, or systems with built-in AI like Xero, use whatever admin reporting or activity history they offer.
Where the platform does not give you analytics, measure the workflow instead: how many meetings were summarised, how many documents were drafted, and how many customer replies were AI-assisted.
Measuring time spent on tasks and therefore time saved is difficult. People do not like doing it, you need to make it as easy as possible. That can be in a ticketing system like a PSA, or it could be a simple spreadsheet — even paper and pen.
AI is valuable when people use it well, but it can quickly become wasted spend if adoption is low or inconsistent. That is why ROI matters: it tells you whether the rollout is genuinely creating value or simply adding cost.
We have run our pilot. People used it. We have demonstrated ROI. Now we need to scale it.
We need three things: knowledge libraries, adoption assets, and governance and support.
Let’s start by making sure people know where to find resources — making sure you have a knowledge library.
A successful rollout normally needs a small set of shared resources. Start with a simple AI hub in Teams or SharePoint. It should point people to the approved document libraries, the prompt bank, process documents, FAQs, templates, and any tone-of-voice or customer communication examples.
“Adoption assets” is a rubbish term, but it is accurate.
Your champions can reinforce the message about where to find these documents and literally repeat them to people who do not like reading.
Create a Teams channel for questions, run short training sessions, save onboarding material for new starters, and keep recording success stories.
Finally, keep the governance basics visible: who has access, what licences are in place, which tools are approved, what needs human review, where issues are reported, and how the content will be refreshed. If someone wants an agent built, for example, they need to know who has access to the right tools or licences, such as Copilot Studio.
This is what stops the rollout depending on people’s memory or enthusiasm.
It gives staff a starting point and makes the right way of using AI the easiest way.
Let’s talk a bit about training after it has rolled out.
AI skills will not be a one-off training event.
The tools are changing too quickly for that.
A simple continuous learning model works better: a quarterly AI lunch-and-learn, champions sharing real examples, a monthly tip of the month, and an internal knowledge base of approved prompts and use cases.
Keep the learning practical and tied to real workflows.
The aim is not to turn everyone into an AI specialist. The aim is to make AI literacy a normal part of professional development.
The behaviour-change piece is what keeps adoption alive after the launch.
It is all about comms, comms, comms, comms, comms.
Managers should make AI part of normal team conversations. In team meetings, ask where people have had AI wins. In one-to-ones, ask how people are getting on with AI and whether they need training in any specific areas.
Remember to keep reinforcing your AI position on which tools are approved. You do not want people sloping off into shadow AI.
Celebrate the wins. Share one practical success story each month, such as a proposal drafted faster, meeting actions captured more consistently, or a customer response improved.
Monthly AI challenges also work well. Give teams one small challenge, ask them to try it, and share what worked. It does not need to be strictly work-related; it could be creating an image, testing a creative prompt, or doing something fun that helps people get more comfortable using AI.
Lastly, when a platform has had its regular review, share that it has happened. Remind people that approved tools are being reviewed and that shadow AI is on your radar.
Here is the practical action plan.
In month one, build the foundation. I have included doing a Copilot Technical Readiness Assessment because it is important to understand what is being shared outside the organisation, what is being shared into the organisation, and who has access to what before AI is introduced more widely.
In addition, select approved tools, define the first use cases, pick your guinea pigs, identify champions, and baseline the current processes.
In month two, run the pilot: train champions, trial AI with small groups, gather success stories, build the prompt libraries, start manager follow-up, and measure outcomes.
In month three, scale what works: extend to wider teams, share wins, refine governance, communicate clearly about approved tools and shadow AI, and plan the next wave of use cases.
Keep it manageable. A small number of well-supported use cases will create more value than a large AI programme that no one uses consistently.
That brings us to the end.