The AI assistant we built for sales got used in that first week, but now we’re barely getting any questions at all.
This is the story of Saeki (not their real name) , who looks after digital transformation at a manufacturing-focused B2B company, and the three departments he found himself working with: sales planning, HR, and information systems. The company had set up an AI assistant for each of sales, the back office, and the training team, yet within a few weeks of launch usage had stalled. Despite an announcement on Slack and a notice on the company portal, the response from the floor was telling: “I don’t know what I’m supposed to ask,” and “I can’t see how this relates to my own work.”
So they went back and mapped out a persona and a set of work scenarios for each department, and rethought what the AI chat was for, the reference data it drew on, the route into first-time use, and how feedback would be gathered. In one model case, for instance, comparing 126 people across three departments over eight weeks after launch, the number of staff using it at least once a week rose from 31 to 68, and the sales team began posting questions about deal preparation on an ongoing basis.
This article sets out the assistant operations, announcement design, and improvement cycle that raise adoption of department-level AI assistants and help workplace AI take root. The aim is a state in which staff open the AI assistant naturally, in the course of their own work, without being told to “please use it.” That said, an AI assistant is no panacea. Greater adoption only follows once you have the work design, the voice of the people doing the job, and a habit of continual review. If this sounds familiar, do read on with your own departmental setup in mind.
A department-level AI assistant won’t get used simply because you’ve built it
Department-level AI assistants tend to draw attention right after launch, only for usage to plateau soon after.
You may have divided the roles neatly at launch into “for sales,” “for HR,” and “for information systems,” but out on the floor the message often hasn’t landed: people aren’t sure what to ask, or in which situations the tool is meant to help.
To raise adoption of an AI assistant, you need to make clear why staff would use it in the course of their work before you go adding more features.
The classic pattern: usage tails off after the initial launch
When usage struggles to grow, a few common threads tend to run through it.
First, there’s the case where the launch announcement amounts to little more than a feature tour: “We’ve built a new AI assistant.” What matters to staff is not the features themselves, but how their own work is going to become easier.
For the sales team, say, it’s hard to get started unless concrete moments of use are visible: framing hypotheses before a meeting, drafting a first cut of a proposal, organising the key issues in a client’s industry. For HR, you need scenarios close to the daily routine: writing training announcements, preparing answers to internal queries, drafting appraisal comments.
Second, a major factor is that the first-time experience simply hasn’t been designed. If, on opening the AI assistant, someone has no idea what to type first, they will close the screen.
“Feel free to ask anything” looks like a friendly invitation. Yet when it comes to embedding workplace AI, too much freedom can, paradoxically, make a tool less likely to be used.
Low usage doesn’t necessarily mean a lack of enthusiasm on the ground
When an AI assistant’s usage is low, it’s all too easy to read it as “the floor isn’t interested” or “people are wary of AI.” But that isn’t necessarily so.
People on the front line are run off their feet with their day-to-day work. With limited time to try a new tool, being told “it’s handy once you use it” rarely turns into a habit overnight.
The thing worth checking, rather, is where the AI assistant connects to everyday work.
For a salesperson, can it be used in the 15 minutes before a meeting? For the back office, can it be reached for the moment a routine query comes in? For a manager, can it be opened before a meeting or a one-to-one?
Whether a department-level AI assistant gets used is decided not by how it’s built, but by how it’s run after launch.
The starting point for greater adoption is per-department persona design
To make a department-level AI assistant stick, you first need to set out the personas of the people who will use it.
By persona here I don’t mean fixing fine details like age or job title. I mean setting out their role at work, the tasks that come up often, the moments that give them trouble, and the kind of help they expect from AI.
What each department expects of the AI differs
Even with one and the same AI assistant, the sales, HR, and information systems departments expect different things of it.
| Department | Role valued most | Example use cases |
|---|---|---|
| Sales | Speeding up customer understanding and proposal preparation | Extracting next actions from meeting notes; shaping the structure of a proposal |
| HR | Bringing consistency to internal communications and training delivery | Drafting training announcements, compiling FAQs for attendees, drafting appraisal comments |
| Information systems | Standardising query handling and internal procedures | Guidance on account requests, tool usage rules, and security reminders |
Announce the AI assistant in one company-wide voice without first sorting out these differences, and the message struggles to reach the people who would actually use it.
Four things to look at in persona design
When designing the persona for a department-level AI assistant, set out the following four:
- The work they do
- Get a feel for the kinds of tasks they handle day to day.
- Their pain points
- Tease out the tasks that take too long, the ones where quality varies, and the ones that have become one person’s preserve.
- When they’d use it
- Think through the natural moments to open the AI assistant: before a meeting, after dealing with a client, at month-end, ahead of training.
- The output format they expect
- Decide on the formats that are easy to use on the ground: emails, tables, bullet points, minutes, comparison tables, FAQs.
Persona design isn’t only for marketing campaigns. For embedding workplace AI too, it lays the groundwork for being clear about whom you’re helping, and in which situation.
Give each persona a “first question”
To prompt first-time use of the AI assistant, it helps to have a “first question” ready for each persona.
- For salespeople: “For tomorrow’s client meeting, please set out a hypothesis about the customer’s challenges and the questions I should check.”
- For HR staff: “Please draft an announcement for the new-joiner training in a form that helps attendees understand what to prepare.”
- For information systems staff: “Please explain the password-reset procedure to staff clearly, in five steps.”
Presenting a first question tailored to each department’s work scenario like this makes it easier for people to give it a try. In running an AI assistant, designing this first-time experience is a crucial gateway to higher adoption.
Use work scenarios to pin down exactly where it helps
To raise an AI assistant’s adoption, you need to convey “what it can be used for” not feature by feature, but scenario by scenario.
A work scenario is simply a mapping of the moments where the AI assistant can help, set within the actual flow of how someone gets their work done.
Place the AI assistant “either side of the work”
An AI assistant tends to stick better when placed either side of existing work rather than used in isolation.
In sales, for instance, the run-up to and aftermath of a client meeting are easy moments to grasp. Beforehand, you organise the industry’s challenges, likely questions, and angles for the pitch. Afterwards, you pull the decisions, the client’s concerns, and the next actions out of the meeting notes.
In HR, you can place the AI assistant either side of training: before, for drafting announcements and pre-work; after, for summarising survey results and drawing out points to improve.
In the back office, either side of query handling works well. When a question comes in from a member of staff, you can check the relevant rules or FAQ and draft a reply.
Building the AI assistant into the flow of work like this moves it from “a tool you go out of your way to use” closer to “a tool you open as part of the work itself.”
Start with just three scenarios
When driving adoption, presenting too many uses from the outset only leaves the floor confused. In the early days after launch, narrowing to around three work scenarios per department makes it easier to run.
| Department | Example work scenarios |
|---|---|
| Sales |
|
| HR |
|
| Information systems |
|
Narrowing the initial scenarios makes it easier both to announce the tool and to analyse the usage logs.
Choose tools in a way that supports the work scenario
There are several ways to run an AI assistant: an internal chat tool, a general-purpose generative AI, an integration with groupware, a dedicated platform. What matters is less which tool you use than whether you can build something reusable, shaped to the work scenario.
If, for example, you want to organise AI chat, prompts, and reference data by department, a tool that lets you separate uses on a per-project basis is well suited. With an environment such as Kanata, where AI chat, a prompt library, and a reference-data library can all be managed within the same unit of work, it becomes easier to set up use-specific operations for sales, HR, and information systems.
Whichever tool you choose, it will be hard to make it stick without rules for updating reference data, managing permissions, and checking outputs. The point is to design the operating rules at the same time as you select the tool.
Pitch the announcement on business benefit, not feature description
When it comes to driving adoption of an AI assistant, how you design the announcement matters a great deal.
A common misstep is to leave it at “We’ve launched the AI assistant” and “You can use it from the link below.” That gives the floor no sense of why they, personally, would use it.
Put “who, when, and for what” in the announcement
An announcement needs to carry at least these three things:
- Whom the AI assistant is for
- At what point in the work it’s meant to be used
- What you put in, and what sort of output you get back
For an announcement aimed at the sales team, for instance, you might put it like this:
We’ve launched an AI assistant for the sales team to help with meeting preparation. Before a meeting, enter the client’s name, their industry, and the service you plan to propose, and it will set out the questions to check, the likely challenges, and angles for your pitch. Do give it a go on a meeting you’re preparing for tomorrow or beyond.
Put it this way and people find it easier to map onto their own work.
Vary what you say across Slack, meetings, and the company portal
It also pays to vary what you convey by announcement channel.
- Slack or Teams
- Best suited to short, concrete examples of use. Rather than a lengthy explanation, “You can use it for this task” and “Just paste in this question” are more likely to prompt action.
- Meetings
- Opening the screen and giving a live demo is effective. Watching someone put a question to the AI assistant and seeing the answer come back lowers the psychological hurdle for the floor.
- Company portal
- Set out the usage rules, the tasks it covers, the caveats, and the common questions. Having a place people can return to later gives them somewhere to turn when they get stuck.
An announcement isn’t a one-off. It pays to re-convey the moments of use at intervals: at launch, a week later, and a month later.
Appointing ambassadors helps adoption along
In driving adoption of a department-level AI assistant, it helps to appoint an ambassador in each department.
An ambassador is someone who tries the AI assistant in their department and shares how to use it with those around them. They needn’t be a manager. Someone who understands the day-to-day work well and whom colleagues find easy to ask is a good fit.
The ambassador’s role isn’t to force people to use it. It’s to share the situations where it genuinely helped, to flag stumbling blocks to the DX team, and to suggest work scenarios the department might use.
With this kind of bridge in place, the AI assistant is more readily seen not as “a tool handed down from head office” but as “a mechanism that helps us with our own work.”
Gather feedback and keep the improvement cycle turning
Raising an AI assistant’s adoption is impossible without feedback after launch.
Usage logs alone won’t tell you why it’s being used, or why it isn’t. You need to pair the voice of the floor with the logs and keep an improvement cycle turning.
The metric to watch isn’t usage count alone
The metrics worth checking in running an AI assistant include the following:
- The number of people using it at least once a week
- Users by department
- Uses per person
- The categories of question asked most often
- Examples of where it fed back into the work
- Complaints about the answers, and what people re-asked
Look at the raw usage count alone and you can misjudge the reality.
For instance, even where the usage count is high, if people are asking the same question over and over, there may be a problem with the quality of the AI assistant’s answers. Conversely, even where the count is low, if it’s being used consistently for important work, you can read that as a sign of taking root.
What matters is looking not just at “was it used” but at “did it feed back into the work.”
Make feedback easy to give
To gather feedback from the floor, you need to cut the effort of giving it as far as you can. Start with simple questions such as these, for instance:
- What work did you use it for?
- Could you use the answer as it was?
- What did you change?
- Was there anything that didn’t meet your expectations?
- Are there example questions you’d like added next?
For the response format, it’s best to mix multiple choice with free text. Ask for a long survey every time and the feedback itself dries up.
Another approach is to set up an “AI assistant improvement channel” on Slack or Teams where people can post what they’ve noticed. The point is not to stop at gathering feedback, but to tell the floor what you’ve acted on.
Based on the comments we had last week, we’ve added example questions for meeting preparation.
A small share like this helps the floor feel that “the AI assistant is growing on the strength of our own input.”
Turn the improvement cycle monthly
The AI assistant’s improvement cycle needn’t be elaborate from the outset. Checking in lightly each week early on, then moving to a monthly review once things settle, is the realistic approach.
At the monthly review, check the following:
- Which departments are seeing usage grow
- Which work scenarios are being used
- Why the scenarios that aren’t being used aren’t being used
- What complaints there are about answer quality
- What reference data or prompts should be added
- Which work to focus adoption efforts on next month
Keep this improvement cycle going and the AI assistant won’t stay at its initial settings; it will grow to fit the department’s work.
Embedding workplace AI isn’t a build-it-and-forget-it affair. It’s an operation in which you watch how something is used and update it alongside the work.
Operational tactics for raising adoption
From here, let me set out the operational tactics that are easiest to put in place to raise adoption of a department-level AI assistant.
Run a first-use campaign
Right after launch, the important thing is simply to get people to use it once.
To that end, run a first-use campaign for each department. By “campaign” I don’t mean some grand initiative.
For the sales team, for example, it might be “for one client meeting this week, try drafting the advance questions with the AI assistant.” For HR, “draft the next training announcement with the AI assistant”; for information systems, “turn one common query into an FAQ.”
The point is not to urge use in the abstract, but to specify a concrete first go.
Share success stories
Sharing success stories is effective in driving adoption of an AI assistant.
- A in sales used it to organise questions before a meeting, which made the points to check with the client clear.
- HR used it to draft a training announcement and cut the time it took to write.
- Information systems tidied up the wording of an FAQ, making the explanation clearer for staff.
Concrete examples like these make it easier for other staff to map the tool onto their own work.
That said, when sharing success stories, it’s important not to oversell the results. Steer clear of phrases like “dramatically improved” or “anyone gets results straight away,” and instead convey concretely in which task, and how, it helped.
Turn frequently used prompts into templates
One reason an AI assistant’s adoption fails to grow is the burden of working out the instructions every time.
To ease that burden, turn frequently used prompts into templates.
For sales, templates for meeting preparation, post-meeting tidy-up, and proposal structure. For HR, templates for training announcements, internal FAQs, and survey summaries. For the back office, templates for checking rules, answering queries, and internal notices.
Set out clearly in the template what needs to be entered.
For a meeting-preparation template, for instance, make the input fields the client’s name, industry, the purpose of the meeting, the service to be proposed, and any concerns. That way users are less likely to be unsure what to put in.
Review reference data regularly
The answer quality of a department-level AI assistant also depends on the reference data it draws on.
For sales, if the latest service materials, proposals, case studies, and FAQs are left out of date, the answers may drift from how things actually are on the ground. For HR, if rules and training materials aren’t kept current, it can lead to mistaken guidance.
So take regular stock of the reference data. At least once a month, check the following:
- Whether old materials are lingering
- Whether there are duplicate materials
- Whether new operating rules are reflected
- Whether content people often ask about has been added
- Whether materials you don’t want used in answers have crept in
Before you get people to use the AI assistant, it’s important to put the information it draws on in order.
Separate out what each stakeholder should be watching
In running a department-level AI assistant, what’s worth prioritising shifts with the stakeholder’s vantage point.
The chief executive, the head of IT, the head of marketing and sales, and floor managers may all be looking at the same “adoption rate,” but what each wants to judge differs.
| Role | Main things to check |
|---|---|
| Chief executive | Beyond the adoption rate, how it is affecting the speed of decision-making, cross-department collaboration, administrative cost, sales opportunities, and staff productivity. |
| Head of IT | Permission management, the risk of information leakage, log management, usage rules, the handling of reference data: in short, safety and operational control. |
| Head of marketing and sales | Whether it’s being used in the work that raises the quality of customer contact: meeting preparation, proposal writing, email wording, loss analysis, customer understanding. |
The chief executive looks at the link to business results, not the adoption rate
What matters to a chief executive isn’t merely how much the AI assistant has been used.
It’s how that use is affecting the speed of decision-making, cross-department collaboration, administrative cost, sales opportunities, and staff productivity.
The adoption rate is no more than the way in. The chief executive needs to look at how the AI assistant connects to transforming the work of the organisation as a whole.
The head of IT looks at safety and operational control
For the head of IT, it isn’t just about how handy the AI assistant is, but about permission management, the risk of information leakage, log management, usage rules, and the handling of reference data.
The more department-level AI assistants you add, the more there is to manage. You need to design who can access what, which data is being drawn on, and how mistaken information gets caught. For wider context, the NIST AI Risk Management Framework offers a structured approach to governing these risks (NIST AI Risk Management Framework), and the UK ICO’s guidance on AI and data protection covers handling personal data responsibly (ICO guidance on AI and data protection).
Embedding workplace AI calls for both ease of use on the ground and control, side by side.
The head of marketing and sales looks at the link to action on the ground
For the head of marketing and sales, what matters is whether the AI assistant is raising the quality of customer contact.
The question is whether it’s being used in the work that ties directly to sales and marketing activity: meeting preparation, proposal writing, email wording, loss analysis, customer understanding.
Even if the usage count is climbing, if it isn’t translating into better customer handling or more opportunities, its value to the floor is limited.
Higher adoption is the way in, not the finish line
Raising an AI assistant’s adoption matters. But the adoption rate itself is not the ultimate aim.
What you should really be aiming for is a state in which the AI assistant is woven naturally into each department’s work, and staff can reach for it at the moment they need it.
From “make them use it” to “there’s a reason to use it”
The phrase “driving adoption” can, if you’re not careful, carry a whiff of “making them use it.” But what embedding workplace AI actually needs is to create a state in which the floor has a reason to use it.
Open the AI assistant and meeting preparation gets quicker. The burden of writing an internal notice eases. Query handling comes out more consistent. Framing the issues before a meeting gets easier.
With that kind of felt benefit, staff start using it of their own accord.
Roll out small departmental wins more widely
You don’t need to aim for company-wide adoption all at once from the start.
First, create a small success story in a department where the moments of use are clear. Then roll it out to other departments.
If sales managed to use it for meeting preparation, customer success might apply it to organising customer-contact histories. If HR used it for training announcements, the general affairs team might apply it to internal notices.
What they have in common is the persona, the work scenarios, the announcement design, the feedback, and the improvement cycle.
An AI assistant is something you run and grow
An AI assistant isn’t finished the moment it’s built.
Work on the ground changes. Materials get updated. The questions people commonly ask change. A department’s priority themes change.
That is precisely why an AI assistant, too, has to be grown through how it’s run.
When adoption isn’t growing, before you doubt the enthusiasm of the floor, check these questions:
- Is it clear whom the AI assistant is for?
- Are the work scenarios concrete?
- Is the hurdle to first-time use low?
- Is the announcement about business benefit rather than feature description?
- Is there a mechanism for gathering feedback?
- Are the reference data and prompts being updated?
Revisit these and the AI assistant moves from “something we built but nobody uses” closer to “an operating foundation that supports the department’s work.”
Summary
Raising adoption of a department-level AI assistant takes more than simply launching an AI tool.
What matters is to set out the persona for each department, show where it helps along the work scenario, and combine well-judged announcements with an improvement cycle.
Five things, in particular, form the backbone of the operation:
- Make clear who the users are in each department
- Narrow to the work scenarios you want people to use first
- In announcements, convey business benefit rather than features
- Make use of ambassadors and feedback from the floor
- Improve continually, drawing on usage logs and the voice of the floor
An AI assistant is no panacea. Bringing it in won’t, on its own, transform every task.
But run it to fit each department’s work, keep improving it into a form staff find easy to use, and workplace AI will, little by little, take root on the ground.
Even where you’re using a workplace AI platform such as Kanata, what matters isn’t the sheer number of features. It’s combining project design, reference data, prompts, usage rules, and an improvement cycle to create a state the floor can keep using. For broader principles on trustworthy, human-centred AI, the OECD AI Principles are a useful reference (OECD AI Principles).
Q&A
If an AI assistant’s adoption is low, what should I check first?
The first thing to check is whether the floor understands “what to use it for.” If the target users, the moment of use, example inputs, and example outputs are left vague, an AI assistant will struggle to be used even once launched. The first thing to revisit is not the features, but the work scenarios.
How does a department-level AI assistant differ from a company-wide one?
A company-wide AI assistant suits a broad range of uses, such as drafting text and summarising. A department-level AI assistant, by contrast, differs in that it tunes the example questions, reference data, and output formats to the work of a particular department, such as sales, HR, or information systems. If you’re prioritising it taking root, a design fitted to each department’s work context is important.
To drive adoption, should I present many use cases from the start?
Rather than presenting many use cases from the start, narrowing to around three work scenarios per department makes it easier to run. The reasons are that users can try it without getting lost, the effect is easier to confirm, and points to improve are easier to gather.
Can an AI assistant’s results be judged on usage count alone?
They can’t be judged on usage count alone. A high count may still not be feeding into business results, while a low count can carry value if it’s being used consistently for important work. It’s important to look at a combination: the number of people using it at least once a week, the work scenarios being used, and examples of where it fed back into the work.
What’s the single most important practice in making an AI assistant stick?
The most important thing is to keep an improvement cycle turning after launch. By gathering feedback from the floor, adding more of the example questions people use often, and continually updating the reference data and prompts, the AI assistant grows into a form suited to the work. Not treating it as build-it-and-forget-it is the prerequisite for embedding workplace AI.
