In this episode of In-Ear Insights, the Trust Insights podcast, Katie and Chris discuss the release of Microsoft Copilot Cowork and its hidden financial implications for your business. You’ll learn how to calculate potential costs by categorizing your daily tasks into light, medium, and heavy workloads. You’ll discover how to apply the 5P framework to prevent runaway AI spending in your organization. You’ll identify specific strategies to optimize your workflows by separating planning from execution. You’ll explore how command-line tools can help you maintain efficiency without burning through expensive credits.
00:00 – Introduction
03:15 – Categorizing AI tasks
08:45 – The shock of the credit-based bill
14:20 – Applying the 5P framework for cost control
19:10 – Using planning to save money
25:30 – Call to action
Watch this episode now to learn how to keep your enterprise AI costs under control before you start using Microsoft Copilot Cowork.
Use the free Trust Insights Microsoft Copilot Cowork Cost Calculator!
Watch the video here:
Can’t see anything? Watch it on YouTube here.
Listen to the audio here:
- Need help with your company’s data and analytics? Let us know!
- Join our free Slack group for marketers interested in analytics!
[podcastsponsor]
Machine-Generated Transcript
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.
Christopher S. Penn: In this week’s In-Ear Insights, let’s talk about the newly generally available Microsoft Copilot Cowork, which is a licensed version of Claude Cowork. So Katie, you have spent a lot of time with Claude Cowork. You teach for Smarter X for their AI Academy on all the different uses of Claude Cowork. You’ll be doing an entire workshop at the Marketing AI conference on the Claude ecosystem and stuff like that. So when you hear that now Microsoft, the largest enterprise AI deployment system, has made effectively a copy of Claude Cowork available, what comes to mind?
Katie Robbert: Endless opportunities. I have never met someone who is like, “Yay, Microsoft.” And we’ve talked about why a lot of companies are tied into Microsoft and a lot of it comes down to security and privacy. Chris, you have a whole series on enterprise AI, so enterprise AI not being the size of the company, but really more of the security and governance requirements needed. Microsoft as a workforce software, Microsoft 365, tends to check the most of those boxes, which is why so many large companies or companies in general tend to be tied into Microsoft. Which also means what we hear is, “Well, I can’t use Claude or I can’t use OpenAI, I can only use Copilot. I want all the bells and whistles that I’m seeing you guys talking about.”
Very quick anecdote. My husband, who I’ve mentioned numerous times, is not a technology person—that is not the nature of his job—was lamenting that the new version of Microsoft is hiding all the replies to his emails from the entry-level user to the expert user. I don’t know anyone who enjoys using Microsoft, but I’m hoping now that this little bell and whistle is something that could bring people around on the users. Because Claude Cowork has been such a literal game changer for the way that I operate. The amount of things that I can get done that I couldn’t get done before because I’m just one person is infinite. Just the other day, I’ve always done the company financial projections—it’s very laborious. I have a spreadsheet, I have to check numbers from four or five different places. That’s something that Cowork can now not only help me with, but build an interactive dashboard for. And it’s like, “Yeah, you got multiple data sets, I got this, I can build that for you.” The amount of time it saves me is immense because it unlocks my time to do things like, “Hey, what’s a new target market we need to go after? What does that look like?” I didn’t have the brain space to do that before because I was so bogged down. So when I hear that Microsoft now has their version of Cowork, I’m like, “Wow, people are going to get so much done if they want to, if they see the opportunities within the software, if they’re curious.”
Christopher S. Penn: If they can afford it. So that’s what I want to talk about on today’s show because Microsoft has released an Excel spreadsheet, of course, a calculator for how much Cowork will cost you because it is pay-as-you-go, it is not flat rate. So let’s talk about some of the tasks that you do, Katie. They define tasks in three categories: light, medium, or heavy. A light task is basically prompt and chat, no tool calls, one deliverable. And they classify this by the four different categories: corporate knowledge workers, customer-facing knowledge workers, technical workers, and managers and senior leaders. Now I would say that you are a manager and senior leader—I think that’s who you are, what you do. I am a technical worker. We have Kelsey who is a customer-facing knowledge worker—she’s our account manager—and we have John who is our corporate knowledge worker. John is our head of business development. So we actually check the box on each of these Cowork types of people.
Now on a daily basis, Katie, you for sure have at least one Cowork process that calls more than one tool because you send out a daily update. So you have at least one of those that’s a medium-level task that sends up our daily sales report. What other daily tasks do you have Coworks have to do?
Katie Robbert: I have Cowork Daily set up to send me a daily writing prompt. All it’s doing is writing to a Word document. I would imagine that’s a lightweight task. Basically, one of the things that I’m doing for my own professional development is I’m trying to make sure I don’t lose that writing muscle. As AI makes it so easy to replicate our voices, I want to make sure I don’t lose it. So I spend a few minutes every morning writing to a randomly generated prompt. So I would imagine that’s a lightweight thing.
You mentioned the update that I send to the team. This is calling on our CRM data, and that I would imagine is sort of a medium because that’s only one piece of software. But once a month, I’m calling on our CRM and our financial data and a couple of other sources, so that would be a heavy task. So on a day-to-day basis, the scheduled tasks that I have are fairly lightweight. But then when I get into the real thinking, that’s when—so I was working on something this morning on behalf of the team. I was engaging a plug-in, I was engaging the Google Drive connector, I was engaging the Google Search connector, I was engaging that deep thinking of “put all this information together,” and all of the skills that are involved: the skill of building a Word doc, the skill of building a PDF, the skill of building an HTML interactive page, the skill of building a PowerPoint—all of those in one specific task. So I would say that is a heavy task, even though it looks at the surface like a lightweight task.
Christopher S. Penn: I would say, and I think this is a fair characterization, you probably do two heavy projects a day in Cowork because you’re constantly doing deep strategy and things. So I’m going to put two a day—this is a monthly calculus—put down 60 there. Now for Kelsey, I would say Kelsey at least does at least one light and one medium task in Claude per day. I think it’s actually more than that, but I’m going to put that down as a starting point. What do you think?
Katie Robbert: I think that’s a fair starting point.
Christopher S. Penn: Okay. For me, I work in Claude code, which is slightly different, but since we’re just trying to get a sense of what Cowork will cost, I’m going to do the equivalent. On a day-to-day basis, I probably do five tasks that are light, so that’s going to be 150 of those a month. I probably do 10 tasks that are medium, so that’s going to be 300 a month. And I probably—actually, I know I do over 10 tasks a day that are heavy, that are like pure heavy code lifting. So that’s going to be another 300 there for John. John really doesn’t use Claude much at all, I don’t think. So maybe like 30 at most.
Katie Robbert: Yeah, I think so. We have a skill that was built specifically with his role in mind, and he runs it maybe once every couple of weeks. When I look at the weekly tasks—so this is looking at a month at a glance—I would actually bump up the medium tasks for me because I have weekly reports that are run that engagement, the Claude Chrome extension, the connections to our CRM, connections to our project management software. I have eight of those weekly.
Christopher S. Penn: Okay, so you’re basically running two mediums a day. Effectively.
Katie Robbert: Yeah.
Christopher S. Penn: Claude or Microsoft Copilot Cowork bills on what are called credits because why make this easy? Light tasks bill 125 credits, medium tasks bill 500 credits, and heavy tasks bill 1,200 credits. The cost is a penny per credit. So our Microsoft Copilot Cowork cost—are you ready for this, Katie? $1,600 a month.
Katie Robbert: Get out. We’re going back to candlelight and whittling pencils.
Christopher S. Penn: That is because it’s a penny per credit, which they do to make it sound cheap, not realizing that a single heavy task is 1,200 credits. So a single task is $12. So for me to do one QA run on a piece of software is swipe the credit card for $12. On a monthly basis, we are consuming effectively 657,000 credits, which is $6,570 total, all in. It’s $1,600 per user. So Katie, our Trust Insights Copilot Cowork bill is $6,570.
Katie Robbert: I have no words. That is insane. And to be fair, so you and I, Chris, I would say are power users. We are turning to these tools to do all kinds of things all day long. Even with trying to do things and schedule them off-hours to not be during peak usage, we’re still using up usage. And yeah, we are a small team. If we take out the work that Kelsey does just for the sake of this example, you and I are still eating up the majority of the cost. If we take out you, I’m still eating up a majority of the cost. I don’t know how a company or team is supposed to be able to afford to use this. It’s a real bait and switch. Shame on Microsoft.
Christopher S. Penn: Well, this is enterprise. They can do this.
Katie Robbert: Yeah, they can. It doesn’t mean they should.
Christopher S. Penn: So your usage, because a credit is a penny, your usage of Copilot Cowork a month would be $1,057.50. That is how much you consume in equivalent credits in the system. Now granted, we pay for the four of us to share a Claude Max 20 account; we pay $200 a month for it. This at the enterprise level, you’re talking four people, $1,600 for four people, one of whom barely will use it. Realistically, like you said, we’re probably going to average $3,000 an employee is what it will cost to use Cowork.
Katie Robbert: Which is an insane amount. For some companies that don’t even blink at that, but that’s a very small handful of companies who would feel that way about $3,000 a month. One of the things that we’re doing with a lot of our clients right now is trying to help them find cost savings in their tech stack—like how many tools can they reduce or licenses they can let go of and replace with things like Claude Code or Claude Cowork. But if they’re like, “Yeah, I want to do that exercise,” and what I have is Microsoft Cowork, I would say, “Cool, we’re not doing that exercise until Microsoft changes the billing,” because it’s going to cost you 10x more than it’s costing you now. It’s not worth it. Which is a real shame because Microsoft users have been waiting for this kind of functionality.
Christopher S. Penn: And so what I wanted to talk about on today’s podcast episode, now that we’ve worked out that this thing is going to cost you three grand a month—because one of the things that people have pointed out on LinkedIn is, “Oh great, you fired all these people so you can switch to AI; now AI is going to cost you more than the people did”—is how do we reduce AI costs? How do we use AI more efficiently? Because this is clearly a lot of money.
Katie Robbert: If only we had a few things to start with. I’m going to shock and dazzle everyone and say, “Guess what? Start with the 5P framework by Trust Insights.” You can learn more about it at TrustInsights.ai/5P-framework. At a high level, the five Ps are: Purpose—what the heck are you doing? People—who the heck’s involved? Process—how do you do the thing? (These are your SOPs). Platform—what tools are you using? (Not just the AI, but also your external data sources). And Performance—did you do the thing?
It sounds really straightforward because it is. However, a lot of people go straight to pushing the buttons and “vibe coding” and, “Hey, build a thing.” “What do you want it to be?” “I don’t know, you pick.” Without doing this work up front, yeah, you’re going to find yourself at $650,000 a month very quickly. There is no tool that allows you to skip over good planning upfront, good governance up front. Microsoft Cowork is no different from any other large language model in that you still need to have good requirements, you still need to have good prompting, you still need to have good governance, even if you’re just using it internally on your own systems. Enterprise companies, any company, has sensitive data somewhere within their SharePoint stack, within their databases, their document repositories. You don’t want to accidentally or carelessly give a large language model access to that because you didn’t plan ahead. So that’s my soapbox. I’m coming down off of it. Chris, what would you add to how to make AI efficient?
Christopher S. Penn: So planning, yes, 100% is going to make the most of the tools you have. The other question is, given these outlandish costs, is Microsoft the right system for you to use? Because Claude in Anthropic’s enterprise level is just as expensive. Companies have recently seen their burn through their entire Claude usage for the year, their budget in weeks. I think it’s Uber that burned their 12-month budget in a month and a half in terms of their token budget. So when we look at these prices, Katie, you remember a while back I had said, “Hey, Nvidia’s got this cool little desktop box. It’s $5,000.” You’re like, “You’re not buying $5,000 worth of hardware.” Absolutely not. Now if Microsoft or Anthropic said, “Hey Katie, you need to pay us $6,500 a month,” you’d be like, “You know what, Chris, go and buy one of those boxes; let’s buy one for each of the team and we’re going to drop Anthropic because we are not paying $6,500 a month for AI.” Right?
Katie Robbert: You know, and so it’s an interesting question because where we started the conversation was saying there’s a reason why people are wedded to using Microsoft because of the security and privacy. I don’t know that introducing an Nvidia box would comply with the regulations set forth by that company. I mean, that’s a big question. It’s an interesting workaround, but it’s not going to work for everybody, especially the more regulated the industry gets. It just might not be an option.
Christopher S. Penn: Yeah, it’s going to very heavily depend on IT. However, because it lives literally in your infrastructure, you do have a lot more governance over it because it’s literally a box that sits on your desk that you control. But more importantly, today’s top local models match a lot of the cloud foundation models and capabilities. GPU AI’s new GLM 5.2 matches Claude Opus 4.8 capabilities. Now you’re going to need a few of those Nvidia boxes to be able to load and run it well for a small cluster of employees. But for the lighter models like Qwen 3.6 or Google’s Gemma 4 if you have to, or Nvidia’s Neotron Ultra if you have to use a US-based model because of regulatory reasons—like you’re not allowed to use anything Chinese, regardless of the fact that it’s on your infrastructure—those are options that you would then use a tool like Open Cowork to handle the inference for it.
So my suggestion is that to Katie’s point, use the 5Ps and then drill down and say, “What are the things that we absolutely positively have to use Cowork for?” Or can we make that task as deterministic as possible using command-line tools and stuff that do not require AI? So for example, Katie, when you query HubSpot every day with Claude Cowork, that is using the MCP connector that uses a ton of tokens back and forth. Now we don’t see it because we’re on an individual plan. The moment we’re forced to switch to a team or an enterprise plan, we will say, “Okay, we’re going to use the HubSpot command-line tool which can fetch data in and out.” And then the AI just says, “Hey tool, give me the thing,” and it goes off and does the back and forth and brings the data back and hands it to the AI. That will dramatically cut the amount of AI usage you have because a non-AI tool is getting data for you.
Katie Robbert: As you’re describing it, I want to sort of make sure I understand because you’re making it sound like it’s an easy switch from the process that I currently have built in Cowork to, “Okay, just use a command-line tool.” I’m not someone who’s well-versed in command-line tools. You’re someone who is. However, you have your own set of things to do right now. So it’s time. It’s internal resources to make those switches to make the cost savings. I just want to be clear about that; it’s not a, “Oh well, in order to save money, let me just go ahead and use a command-line tool.” Like you still have to set it up.
Christopher S. Penn: Yes, and corporate IT will be very busy doing that. However, corporate IT also likes us because they can then govern it. They can say, “Okay, we will ensure that this suite of 10 command-line tools is installed on every computer in the company, and there’s a joint service key that we can maintain programmatically and rotate every 30 days and stuff like that.” So that infrastructure, which corporate IT is very well-versed in, is going to be much happier with that than kind of like the whole shadow IT where people are like, “Oh, I’ll just have Claude make me this thing.” No, they would much rather say, “I would like to have control over the command-line tools that are installed on every machine in the company.”
Katie Robbert: So work that out. You’ve worked with IT teams before. How likely is it that they’re going to—if you say, “Hey, I would like to have control over the command-line tools on every machine in the company,” they’re like, “Yeah, sure, Chris, no problem. Let me bump you to the top of the list. You’re a priority now.” I think you’re going to have a hard time. Like, we see the value in it, we know that it’s a useful thing. I just want to be realistic, and I’m trying not to derail the conversation too much, but I just want to be realistic that, like, yes, that’s the thing. If you have the skills to do it and if you don’t have to go through your IT team to do it, absolutely do it. If you have to go through your IT team and they have to set it up, get comfy, get in line; you’re not a top priority right now.
Christopher S. Penn: Yeah, well, my perspective is IT would want to do that. It would be like, “We would love to have more control over this to stop the shadow IT that’s happening all over the place because of AI.” So IT in its MDM config would say, “Okay, these are the 10 tools that we’re going to drop on every machine, and we’re going to also programmatically alter your Claude MD files and stuff to tell Claude this is what’s installed. You must use it so that it cuts those costs.” And IT can then say, “We certify these 10 command-line applications are safe to use.”
Katie Robbert: Provided it has the time to get skilled up to do that. So yeah, I like to make sure that we’re very clear about caveats because in the 25 to 30 minutes we have for a podcast, we go through things like “do this, do this,” and then it’s, “Well, what do you mean? It said no.” So let’s get back to—Microsoft has started to release Cowork, their version of Cowork, and we’re talking about AI efficiencies. When you think about starting places for someone who’s using Microsoft, someone who’s using their Cowork version, what is the first thing you think somebody should do before they start burning tokens or usage or spending pennies?
Christopher S. Penn: The five Ps, the planning, and build all of your prompts and all of your infrastructure for Cowork in regular Copilot, because regular Copilot is very smart. Now in regular Copilot, if you go in the upper right-hand side, there’s a little menu, a little drop-down saying “models,” and you should choose for planning. Choose GPT 5.5, soon to be 5.6—”think deeper,” that’s the smartest model that’s available. And say—and that’s where you have your conversation like, “Oh, I want to do this in Cowork. I don’t want to do this, I want to do this. Help me figure this out. Ask me questions. Let’s plan this out. Here’s the Trust Insights 5P framework. Help me use this to come up with these plans.” So you do all of your planning and all that heavy token usage in regular Copilot to build the skills and the pieces that you can then drop into Cowork, so you don’t have to use Cowork to plan because Cowork is going to chew up your usage. That way, if you can use regular Copilot, it should be a little bit lighter on your budget.
Katie Robbert: And I think one of the questions that you should add into your planning is, “Can I do this in Copilot or do I need Cowork for this?” And you know, I want you to use your human judgment, but it would be a good idea to ask the large language model like, “Do you have the capabilities to do this within Copilot or do I need to bring this into Cowork to actually execute it?” Because you may be surprised. You know, to Chris’s point, the models are getting smarter every day. And so you may not need to execute what you think you need to execute in Cowork; you may be fine with using Copilot. Yes, I get it’s not as shiny and as exciting, but you know what’s also not exciting? Being told you owe the company $60,000. That’s not exciting.
Christopher S. Penn: Exactly. Even for something like scheduled tasks—Microsoft Copilot tasks are scheduled tasks—so if it’s not something that needs Cowork’s horsepower, that will obviously keep you from chewing up those extra credits over there.
Katie Robbert: Yeah, and I think that’s a good best practice for a lot of these tools, you know? So can you do your planning in Claude Chat before bringing it into Cowork? Can you do your planning in Gemini before bringing it into their version of whatever that is? And that’s just a good best practice for efficiency in general.
Christopher S. Penn: Yeah, I mean, when I do my planning for even software builds and stuff like that, the first thing I do is I have a master planning prompt. It’s actually a skill that incorporates the 5P framework by Trust Insights. And so I have the model ask me questions from the 5P framework: “What are you doing? Who’s it for? How should it work? What are the additional command-line tools that we should be using? What is the definition of done?” And all of that is stuff that if I don’t dictate it out loud, it knows to ask me for it. So I can plan first, and then the language model rebuilds the prompt into something that meets all of those conditions and produces a really solid output that I can then go use to build requirements documents and all the stuff. You will save so much time and money by investing more heavily in planning up front, and you can then hand off the execution of the plan to a very small, fast model.
Katie Robbert: And I think that’s a really good pro tip. And I just want to give a small plug—you can actually download, we have for sale in our academy at Academy.TrustInsights.ai, a “prompt-to-skill.” So basically, as Chris was just describing, he has a specific process for building those requirements. This prompt-to-skill will help you do that and get more efficient at building those requirements. And then what you may find that you have is a reusable template, and it makes that even more efficient. So start with that. Go to Academy.TrustInsights.ai, purchase the prompt-to-skill—it’s very awkward to say that—and then start building out those requirements before you bring it into something like Cowork. And you’re going to save yourself a lot of time and money, and people are going to be like, “Wow, you did that really fast. How did you do that?” And you’ll be like, “I don’t know, I’m just that good.” But in the back of your mind you’re like, “I use the 5P framework by Trust Insights. It got me there faster.”
Christopher S. Penn: Exactly. So Copilot Cowork from Microsoft is now generally available. Before you type one character into it, please take the time to use the 5P framework by Trust Insights. Take the time to understand what your company has budgeted. Take the time to understand what tasks fall in each category, and as best as you can, try to reserve it for the things that truly need Cowork’s capabilities. And don’t just make it the default. If you’ve got some thoughts about the new Microsoft Copilot Cowork that you want to share, pop by our free Slack group. Go to TrustInsights.ai/analytics-for-marketers where you and over 4,700 other marketers are asking and answering each other’s questions every single day. And wherever it is you watch or listen to the show, if there’s a channel you’d rather have it on instead, go to TrustInsights.ai/TI-Podcast. You can find us in all the places fine podcasts are served. Thanks for tuning in. We’ll talk to you on the next one.
|
Need help with your marketing AI and analytics? |
You might also enjoy: |
|
Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, INBOX INSIGHTS. Subscribe now for free; new issues every Wednesday! |
Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new episodes every Wednesday. |
Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.
Newsletter
Podcast
