During this session, we explored how Copilot can be used to help recruitment and staffing firms throughout the organisation, with some practical examples and demonstrations. Here’s a summary from the session host, Daniel Fox:
For the last couple of years, most conversations about AI have really been about personal productivity. Write this email. Summarise this document. Research this company. Create this presentation. All useful, but saving somebody ten minutes writing an email is very different from changing how a staffing business actually operates.
That is why I think Microsoft Copilot has become much more interesting. It arrived a couple of months later than ChatGPT and some of the other AI tools, but Microsoft has been steadily connecting AI to the place where most businesses already do their work: Outlook, Teams, Word, Excel, PowerPoint, SharePoint and the rest of Microsoft 365. And that changes the equation.
Most people are already using AI somewhere in their working day. The problem is that much of that usage is still individual. One consultant has a favourite ChatGPT prompt, someone in marketing uses Claude, a manager has discovered Perplexity, and somebody else has built their own collection of prompts. Everyone might be getting slightly more productive, but the business itself hasn't necessarily changed.
The bigger opportunity comes when AI becomes an organisational capability. When it understands your data, your meetings, your emails, your documents and your processes, it stops simply helping you create things and starts helping you do things. I think that distinction matters.
Microsoft hasn't always made Copilot particularly easy to explain, but there is a relatively simple way of looking at it.
Traditional generative AI has always had a context problem. You can ask it to prepare you for a meeting, but first you need to tell it who you are meeting, give it the previous emails, upload the proposal, explain what happened last time and tell it what you are trying to achieve. At some point you start wondering whether it would have been quicker to prepare yourself.
Microsoft's advantage is that much of that context already exists inside Microsoft 365. That is the thinking behind Work IQ. If I ask Copilot what meetings I have tomorrow, it already has my calendar. If I ask it to prepare me for a conversation, it can potentially use the emails, meetings and documents connected to that person.
The AI hasn't necessarily become dramatically more intelligent. It has become dramatically better informed. And in business, that might be even more important.
We demonstrated this using some dummy placement data. Around 735 placement or job order records containing consultants, candidates, salaries, revenue, placement dates, invoice dates and payment dates. Basically, the sort of spreadsheet that could come straight out of a recruitment CRM.
The instruction was deliberately vague: act as an experienced recruitment market analyst, review the data and tell me what I should know.
The analyst agent worked through 119 separate analytical steps. It found the obvious numbers: placements were up 77%, revenue was up 79%, average fees were around 19% and average lead time to placement was 35 days. But then it went further. It identified that some consultants appeared to be discounting more heavily and that margin could be leaking as a result.
That's much more useful. Calculating a percentage is automation. Spotting something you might want to investigate is insight.
Knowing your own numbers only gets you so far. The obvious next question is whether those numbers are actually good.
So we gave the findings to Microsoft's research agent and asked it to benchmark the business against high-growth staffing companies. It carried out 42 checks and produced an eight-chapter report comparing the performance with external market data.
Suddenly that 77% placement growth had context. So did the 35-day time to fill. And the output didn't stop at reporting what had happened. It recommended what management should look at next.
This is where I think AI starts to become genuinely valuable to leadership teams. Not because it gives you more data, because most businesses already have plenty of that. It helps turn data into questions worth asking.
Doing something impressive once is nice. Doing it every month without rebuilding the process is considerably more useful.
So we built an agent. Its job was simple: take placement and revenue data from a specific Excel file, analyse it, compare the results and create a two-page CFO report in Word with charts. We defined the source it was allowed to use, the type of analysis we wanted and what it should do if the information wasn't available.
A task that could easily take someone a couple of hours took around five minutes. Next month, instead of repeating the process, the instruction can simply be "run the report". Better still, it can be scheduled to happen automatically.
Even on a small scale, that is the difference between using AI and redesigning a process around AI.
Cowork operates far more as an assistant, or as the name suggests, a coworker. Give it the outcome and go and do something else.
Before one of our Copilot events in Manchester, we asked Cowork to help prepare the sales team. The brief included researching Copilot adoption, creating a presentation for the team, finding a suitable meeting time with several colleagues and preparing the relevant materials.
It built a plan and checked details when it wasn't certain. At one point it found several people internally with the same first name and asked us which one we meant rather than guessing. It then found a suitable meeting slot, prepared the research, created the deck and drafted the communication around it. The whole process took about 40 minutes, time I was able to spend doing something else. This starts to feel less like operating software and more like delegating work to a colleague.
There is understandably a lot of discussion about whether AI will replace recruiters. I think that is the wrong objective. The better question is how much of a recruiter's working week actually requires a recruiter?
Meeting clients does. Understanding a brief does. Interviewing candidates does. Building trust, negotiating and making placements do. A surprising amount of the administration surrounding those things doesn't.
If AI can remove more of that work, the opportunity isn't necessarily to employ fewer consultants. It's to give consultants more capacity to do the things that actually generate fees. A consultant who can effectively manage more relationships, more vacancies and more candidates without creating more administration becomes considerably more productive.
You can't simply buy Copilot licences and declare yourself an AI-enabled business. The quality of the output is still heavily influenced by the quality of the environment underneath it.
If your information is spread across seven systems, duplicated in three places, inconsistently structured and badly maintained, AI doesn't magically fix that. In some ways, AI makes good data architecture even more important.
Which is why staffing firms thinking seriously about AI should also be thinking about consolidating their technology, cleaning their data and reducing unnecessary silos. Machines can only reason effectively from the information they can access.
This might be the biggest management challenge. If ten consultants each have their own AI tools, prompts and ways of working, you might have ten slightly more productive consultants. You haven't necessarily built a more productive business.
The real opportunity is identifying the things AI does well and making those capabilities part of the process. Build the agent once, share it, improve it and measure it.
And then ask the question that actually matters. Did it improve the outcome? Did consultants make more placements? Did managers make decisions faster? Did salespeople spend more time talking to the right prospects? Did margins improve? Did administrative time fall?
AI usage isn't the KPI. Business performance is.
A couple of years ago, businesses could reasonably argue that AI wasn't ready. That's becoming much harder to say. The technology is already sitting on millions of desktops. The bigger challenge now is deciding what to do with it.
My advice wouldn't be to start by asking, "How do we get everyone using Copilot?"
I'd start with a much more useful question:
What work are our people doing today that they shouldn't still be doing themselves?
Find that, then give it to AI.