Free YouTube Transcribe

Video transcript

Real-world agent case studies: how organizations are using AI agents

Microsoft Power Platform · 7,343 words · 34 min read

Want to search this transcript, jump the video from any line, or download it as TXT, SRT, or VTT?

Open in the transcript tool

Full transcript

0:00[MUSIC]

0:05First of all, thank you to Lupe and Sameer and

0:08the Microsoft team for having this session and

0:10allowing us to talk about our exciting transformation journey.

0:14We're partway through that right now,

0:16so we're excited to share where we are at,

0:18what we've done so far,

0:20and certainly there's so much more to come for us as well.

0:23I'm Steve Goodwin, Senior Director of

0:25Enterprise Applications at Graebel Companies.

0:27I've been in post for about six and a half years now,

0:30so quite a long time in technology,

0:33particularly with the advent of AI and how

0:36quickly everything is moving in that area.

0:39Shaun, pass it over to you.

0:41Yeah. Again, just echo that thanks to Lupe and Sameer for hosting.

0:46I'm Shaun Eades. I'm the Senior Director of

0:48Process Improvement over our Client Financial Services Department.

0:52Overseeing some of the accounts payable,

0:55just a number of areas,

0:58and that process improvement aspect has definitely taken on

1:03a new avenue with the implementation of

1:07Copilot and Cowork and all of those things.

1:10The things that I thought were automation years ago,

1:14Excel formulas are now yesterday's news.

1:17We love the direction we're going.

1:20In 14 and a half years with Graebel,

1:22it's just been one advancement after another,

1:25so we're really excited where this journey is going to continue to take us.

1:30Thanks, Shaun. Just some housekeeping.

1:32I have three different screens at the moment,

1:34and my chat and everyone's faces are on the far right.

1:37If you see me looking that way, that will be why.

1:40What are we going to do today? What are we going to talk about?

1:42First of all, we're going to introduce

1:44Graebel Companies and just talk about our journey so far as an organization.

1:48I think that's the important point to make,

1:51is how we've moved from what we were to what we'll be coming today,

1:55and Shaun will cover some pieces of that.

1:57We'll talk about our Be Bold initiative and what that really means for us.

2:02It's how our business transformation is being branded internally,

2:06but certainly Be Bold carries so many different positives for us,

2:10and we want to talk about that.

2:12Getting into the solutions, of course,

2:13is the key aspect of why we're here,

2:15talking about some of the solutions that we

2:17have deployed and are utilizing right now.

2:20That'll be a mixture of Power Platform,

2:23Copilot Studio, and some of the more

2:25advanced features around Copilot Studio as well.

2:28Then we're getting to key takeaways,

2:30which is our experience,

2:32what went well, what did not go so well,

2:34and where did we get those key learnings,

2:36and how did we bounce back from some of the failures that we did have.

2:40Then pass it back to Lupe and Sameer,

2:42we'll have a panel discussion around that.

2:45Shaun, I'm going to pass it over to you for the introduction to Graebel.

2:49Thanks, Steve. Graebel is a 75-year-old company.

2:54The name is familiar to some,

2:55not so much for others.

2:57Graebel started in Washtenaw, Wisconsin,

3:00Little Town Central Wisconsin with our founder,

3:03Dave Graebel, and he had an idea of taking a truck,

3:08and some friends, and helping people move across town,

3:12moving their actual household goods.

3:14That initial dream has expanded into what we are today.

3:21Not just dealing with individuals' household goods,

3:24but we are a global partner for a lot of

3:28very influential customers around the world,

3:32helping them with their client relocations,

3:36or I'm sorry, their employee relocations.

3:38As you can see, we cover 195 plus countries that changes every day.

3:44We're doing about 85,000 plus relocations of employees every year.

3:51It's a journey.

3:53There's more than just moving household goods involved in a relocation,

3:57especially when it's international.

3:59We assist our clients and their employees

4:02with everything from language training to visa and immigration,

4:06home sale, school tuition coverage,

4:10and then area tours, cultural training.

4:13You name it, a move, if anyone's ever done it,

4:18is one of the most stressful things that you can do in your lifetime.

4:23Graebel's goal, one of our goals is to make

4:26that as least stressful as possible for you.

4:32Lupe referenced in the beginning that we're a

4:35managed service provider with global mobility.

4:38What really is that?

4:40It's global mobility as a service.

4:43What do we provide?

4:44Well, these seven items on the screen are

4:47everything that get packaged together for us.

4:49It's the standard mobility service of cost estimates and surveys

4:55and all of that into a strategy conversation with our clients of

5:01how do they want their mobility and their services to be managed?

5:04What matters to them and what can we provide for them?

5:07We ensure that we take that with a compliance vision to make sure that

5:12we're managing and mitigating risks that may come up from their journeys,

5:18actively monitoring things,

5:20and we do it with a supply chain network that is very near and dear to our heart.

5:25We are a homegrown grassroots organization,

5:29and that's who we look for in a lot of our

5:31partners to make sure that the same service

5:34that Dave and his son Bill,

5:37who we have now in our chair and Ron as our CEO,

5:42envision an experience being.

5:44We want to make sure that who we put out on the ground is sharing that same vision

5:49and is doing things as we would like to do it.

5:53Now you bring into that,

5:55"That's great, we can do that, but what are we doing it with?"

5:57It's the technology, it's the innovation

5:59aspect that is really starting to set us apart

6:02from others in this space.

6:04The experience that our clients and their employees have when they're going through

6:08this process with us is top of the bar.

6:14Our expectation is continually improving that

6:16experience and improving the backend as well

6:19for the entire process on both ends.

6:25Thank you, Shaun. One of the key things that Shaun's touched on there is

6:30Graebel is a company that is family-owned.

6:33It started from very humble roots and it was very asset-driven.

6:37It was very much a man in a truck,

6:40going around and moving people's furniture and helping them relocate.

6:45We fast-forward 75 years now to where we are today

6:49and we find ourselves in a position where the world has changed.

6:52Graebel as a company has changed.

6:54Over time, we've diversified into different portfolios,

6:57we're attacking different marketplaces.

7:00One of the key things that we're having to really deal with,

7:03and I'd be interested in chat if other end users like us are in the same boat,

7:08clients are asking for more things,

7:10clients are asking for different things.

7:12The market is dictating to us that we have to change.

7:15What we have done as a company is started to

7:19really embrace that business transformation piece.

7:22We've been doing it for a while,

7:23but the injection of technology and the advancement of technology

7:27only accelerates that journey for us.

7:29We're trying to make ourselves as frontier as possible to use

7:33Microsoft terminology and really embrace that technology and take it

7:37into all of the experiences that we have.

7:40It really started with Ron Dunlap's,

7:43Ron Dunlap is our CEO, I should say.

7:45Started with Ron's initiative and challenge to

7:48the organization that we do have to be bold,

7:51that we need to look at different ways of doing things.

7:54The way we did it yesterday is not necessarily going to lead to success tomorrow.

7:58That is something that we need to really embrace.

8:01He's challenged the whole organization,

8:03not just the senior leadership,

8:05not just levels of management,

8:07everybody in the organization to be bold and to

8:10think differently and to challenge the way we do things,

8:13which I think is a really innovative approach and

8:16certainly lends itself to that frontier thinking as well.

8:19This is a short video from Ron just explaining what that means for us.

8:23Right now, this industry is ripe with challenges.

8:26That means it's ripe with opportunity and that means being focused,

8:30having a clear strategy,

8:31understanding where you're going to go as an organization,

8:34having the right teammates to get there.

8:36That means challenging yourself and being bold and thinking outside the box,

8:40trying new things and not fearing failure, actually celebrating it.

8:43If you don't try, you'll never succeed.

8:45I can tell you the greatest successes you have come with the greatest risk.

8:49If you're not willing to take them,

8:50you'll never have that great success story.

8:55Great words from Ron there and that certainly is

8:57echoed across all of our leadership group as well.

9:00Everybody, as I mentioned,

9:01is being challenged to think differently.

9:03We've started an initiative internally called Be Bold,

9:06which is very much a whole cultural shift for

9:10us because we haven't really done that before.

9:12Anyone who's been in an IT team before will know

9:14that you get lists and lists and lists of requests.

9:17You can only do so much to change and to make those things relevant.

9:22With this, we're trying to really push

9:25some of that power back into the organization.

9:29Shaun is a great example of that.

9:31His role is continuous improvement,

9:32but he's now got tools that five years ago, Shaun, you would never have.

9:36Let's be honest about that.

9:38Let it all.

9:38Shaun's doing great things with it.

9:40He's able to just run a little bit of IT overhead, of course,

9:44and making sure that we've got the right governance structures in play.

9:47Shaun's able to go and do things in his area and make changes

9:51that are improving the way that we work greatly in his domain.

9:57That's the type of initiative,

9:59that's the type of cultural impact we want to see more of across our organization.

10:04We're doing it through Microsoft Frontier Framework as well.

10:08Anyone who's seen that before will know that the four quadrants on the screen

10:11are very much the way Microsoft have approached how you do AI,

10:15how you do business transformation.

10:17We shamelessly borrowed that.

10:19In our ambition to be more frontier,

10:22we are really using these as the pillars in which we are operating.

10:26Everything we touch is, are we improving the employee experience?

10:30Are we making life easier for our people to

10:32perform the duties we're asking them to do?

10:35Are we reinventing the customer engagement process?

10:38Are we making sure that people are getting that personal experience,

10:42that it is hyper-personalized to them,

10:43that it is successful and that they do

10:45leave us with a sense of greater satisfaction?

10:49Are we reshaping our business processes?

10:51We are a company of massive manual processes and backlogs.

10:57What we deal with and the way we've done it in the past,

11:00again, is not going to be the success that carries us forward.

11:04So challenging that thinking, why do we do everything the way we do it?

11:08Why have we got so many manual processes that people take a long time to do?

11:11Can we bring technology in to change that and make that more seamless for us?

11:17Then lastly, bending the curve on innovation,

11:19that is the key piece around Be Bold.

11:21How do we get that creative problem solving as a signal from the organization,

11:27that it isn't just coming from the IT team,

11:29or it isn't coming from the senior leadership group?

11:32Again, to go back to Ron on this,

11:34he has said in numerous occasions,

11:36the greatest ideas for how this company is going to work in the future,

11:39are going to come from people lower down the organization.

11:42They're not going to come from the C-suite.

11:44It's going to be someone who is working in these business processes,

11:49who is in the trenches every day,

11:51is going to be the person who comes up with

11:53the best ideas on how we can do things differently.

11:56So that's just a glimpse into what we are as a company aiming for.

12:01This whole mantra around Be Bold,

12:03is certainly it's taken over a lot of our thinking internally,

12:07and it's how we're trying to position our company moving forward.

12:13Shaun, over to you.

12:15Thanks, Steve.

12:15Just to expand on a couple of things that Steve said,

12:18we have manual processes.

12:21So I've said I've been with the company 14.5 years.

12:24When I started, there were paper time sheets that were put on my desk,

12:28that I had to audit by hand and then manually key into a system.

12:34At that point, we didn't even necessarily have imaging to compile that stuff.

12:38So 14 years ago,

12:40we were doing things manually compared to how we're doing them today.

12:43I can't even think as to how they were done 75 years ago.

12:48The other aspects that really grabs me from what Ron said is the failure aspects.

12:57So some of the things that you're going to see,

12:59we didn't create immediately and have wonderful success at them.

13:04We failed fast.

13:05That's the key.

13:06Have a Be Bold idea and don't be afraid.

13:10Somebody had it in the chat or feel the fear and do it anywhere.

13:14Anyway, make sure you fail so that you can learn it.

13:18That failure, especially if you're doing it at a leader level,

13:21helps empower your frontline people to realize it's okay to make mistakes

13:25and it's okay to challenge the normal.

13:28We don't ever want to hear that, "Well, that's how we've always done it."

13:31If we have that idea that we're never going

13:33to move forward with some of these things.

13:36So the other aspect about those Be Bold ideas is AI is the buzzword.

13:44Everyone wants AI, but not every solution is AI.

13:48It doesn't have to be.

13:50Agents have a place, but they're not needed for everything.

13:55So a couple of the things that we'll show you on the screen now

13:57are things that I've built as a citizen deveLuper.

14:02I'm not IT, I have an accounting background,

14:04but I had a curiosity to do things differently

14:07and provide an opportunity for my team to be more efficient.

14:13So hearing the background of what Graebel is,

14:16you can imagine the number of invoices

14:19that we get from suppliers on a regular basis.

14:22If it's tied to those services have to be compliant with what

14:26our clients need from regulatory purposes,

14:29tax guidelines, just their internal HR policies.

14:33There's an opportunity for things to be incorrect on those invoices.

14:39Those invoices sometimes need to be rejected back to the supplier.

14:42So one of the first things that you'll see here is

14:45actually a process built entirely in Power Platform right now.

14:50The team is finishing up testing on,

14:52and we hope to deploy probably in the next couple of weeks,

14:55maybe a month.

14:57Steve, if you want to play the video, I'll kind of narrate over the top of it.

15:02So we're looking at D365 Dynamics 365 Finance Operations,

15:07pending vendor invoices.

15:08So the one highlighted on the screen will use as an example.

15:12Something was wrong with the invoice when we received it,

15:14so we need to reject that invoice.

15:17The process that the teams do today is very manual.

15:22I'm not sure it's playing anymore, Steve.

15:25There we go.

15:26It's a very manual process.

15:28So what we built is a Microsoft form that allows them to identify the reasons

15:33why that invoice needs to be rejected.

15:36And we view standardization.

15:37There's very specific reasons that could be selected for that rejection,

15:42whether it's externally back to the supplier.

15:45It could be internally.

15:46Maybe we've just allocated it wrong in our system through automation.

15:50The analyst populates a few other pieces of information,

15:55who should get the rejection notification,

15:58and then where inside dynamics it exists from a legal entity perspective,

16:02identifying the invoice number, the line reference,

16:05and then having the opportunity to add some additional comments.

16:09So once they take the time to fill this out, which is about 30 seconds,

16:13they'll hit Submit, and then all the magic happens.

16:17So what we'll see is that there are a few things that will happen.

16:25They get the notification that the response was submitted.

16:29Power Automate flow begins to execute.

16:31We can monitor that as it comes through.

16:35Notice that it's running.

16:36You'll see the history of the duration on there.

16:39Most of them are 30, 40 seconds, maybe a minute.

16:44That's all the time it takes for the entire automation to be submitted.

16:45It takes for the entire automation to run.

16:47And that automation will do a few things.

16:50It will go into dynamics, into that vendor invoice record,

16:56and it will standardize an on hold reason within the record

17:01so that it makes it easier for us to report.

17:04It identifies who did the rejection, the date, external,

17:07all of the reasons that they selected.

17:09It places the invoice itself on hold so it can't be processed.

17:13It updates who was the person that was doing that rejection

17:18and identifies that it's rejected.

17:21It also generates an email that then gets attached

17:25to that vendor invoice record so that we have an audit trail.

17:28We have confirmation of this invoice was rejected,

17:32and an email notification was sent out to that supplier,

17:36giving them the information of the invoice, the reasons,

17:39and what to do if they're submitting a correction,

17:44all noting that this is an automated generated email

17:49so that they don't reply directly to that.

17:52And then we go into SharePoint in that same process

17:55and we create an entry that logs that rejection

17:57so that we have a history of all those projections

18:01using that information.

18:03So, Steve, if you want to pause it for one second.

18:06So, that whole process, it took me longer to explain it

18:10than it actually takes for it to run.

18:13But for someone to do that manually,

18:15depending on the complexities of that invoice,

18:17is an 8 to 12 minute process, creating an email,

18:21generating a separate rejection letter, updating dynamics,

18:24sending it out, updating a log.

18:28So, we've essentially reduced, on average,

18:31a 10 minute process down to 30 seconds.

18:34This next one is also kind of related to vendor invoices.

18:40So, hundreds, thousands of invoices come in daily

18:43and our suppliers want to know where it's at in the system.

18:46So, we asked them to send us an aging statement

18:48on a monthly basis.

18:50So, we've got some mocked up data here

18:52with a number of records.

18:56And if anyone's ever received a supplier statement,

18:59he'll know that it's not consistent

19:01across your entire network.

19:02They, especially from an international perspective,

19:05there may be different date formats or number formats,

19:08or one supplier may call a column invoice number,

19:13but another may call it a factura

19:15because it's international or it's invoice

19:17abbreviated with a hashtag behind it.

19:20So, there's all kinds of variations.

19:22So, it's a very manual process today

19:24for our statement reconciliation team to go through these.

19:28It's a lot of V lookups.

19:29It's a lot of exports, other dynamics,

19:31the various pending posted paid grids.

19:34So, what we've done here is we've taken their statement

19:37and we've allowed Power Automate to reconcile it for us.

19:43Hang up in the video.

19:47So, we've created an intake library.

19:50So, when we received that statement,

19:51we simply drop that file into this library

19:55and it's going to ask for two pieces of information.

19:57It's going to ask for the dynamics vendor number

20:00and then the received date

20:01so that we can record that information.

20:03So, once those two pieces of information are populated,

20:07it's the trigger for Power Automate

20:09to continue on to the next processes.

20:14So, what will happen during that whole process

20:16is it's going to standardize

20:19the information in the header of the statement.

20:22So, whereas we said one could be invoice abbreviated

20:26with a hashtag, another could be a invoice number

20:29fully written out, the automation,

20:32which we're just hung up a little bit here.

20:34Sorry about that.

20:36We'll standardize all of that

20:38and we've created a normalization registry

20:42to help with that standardization.

20:49Those.

20:50So, we'll just let the video catch up just a tad.

20:54So, again, this is the part where we're putting in

20:56the account number and the date.

20:59[AUDIO OUT]

21:05The automation will trigger based off of that entry

21:08and then this should be our normalization registry.

21:12So, at the supplier level, we identify

21:16what should be in those various column headers.

21:19So, what do they say is in invoice column?

21:21What do they say is in currency or invoice date?

21:24It allows us to identify

21:27if their dates are European format,

21:30if the numbers are European format,

21:32or if their statement is broken up into sections

21:35with multi-currency.

21:37So, as the flow triggers, the first thing it does

21:41is it goes out and finds this vendor record in the registry.

21:45If it can't find it, we get an email basically saying,

21:48"Hey, you don't have this one set up yet.

21:50We can't do that reconciliation."

21:53And provides a link back to the file,

21:55a link to the SOP for us to do the correction,

21:58and a link to the registry.

22:00All things being good, the automation then continues

22:03and does a number of queries directly in the dynamics

22:08and standardizing things like the invoice number,

22:12which may come across with spaces,

22:16hashtags, special characters, as you can see them

22:18in the various cells in the slide right now.

22:22And it does that normalization of the records in D365

22:25as well.

22:26So, we're comparing apples to apples.

22:30Once all the reconciliation is done,

22:33it creates an original statement in a supplier folder

22:37and then a reconciled statement.

22:38So, the original is exactly what they sent us

22:41so that we have proof of that original.

22:44And then the reconciled statement is in a completely

22:46different format that we've standardized across the board.

22:49So, what we give out to our suppliers is easier for us

22:54to digest.

22:56It contains all the information that they've given us,

22:59invoice number, amount, invoice date, status.

23:03So, we've standardized our status,

23:04whether something is pending payment, it's rejected,

23:08exception, there's an amount mismatch, it could be paid.

23:11There could be a missing on here as well.

23:14And as this is pulled in, it includes the D365 on hold

23:19reason to why it was rejected if that was the case,

23:22or it shows the paid date, the paid amounts,

23:24currency, so on and so forth.

23:26Also identifies via conditional formatting

23:28if something is extremely aged on that statement.

23:33And this automatic ultimately is what our statement analysts

23:38get via an email notification.

23:42It also logs at a line level in a SharePoint list,

23:45every single line that was on that statement

23:47so that we have a full audit history.

23:49Ultimately, when the analyst gets that confirmation

23:54that the statement's been reconciled,

23:56all they have to do today is double check things

23:59that maybe have a mismatch or are missing.

24:01Everything else is ready, and then they can immediately

24:03send it out to the supplier.

24:05So, between those two solutions, all driven in Power Platform,

24:10we've had a huge time savings, minimizing human error

24:14through the process, and everything is standardized.

24:17There's logic behind it now.

24:18So, it makes it a lot easier for external parties to consume.

24:25Cool. Thanks, Shaun.

24:28Let's move on to the next one.

24:33This is James Welsh, Shaun.

24:35This is James Welsh, yes.

24:37Yembo is a partnership that Graebel just recently undertook.

24:43We have become one of the first to actually use this.

24:47So, if we think about the initial part of Graebel

24:51of moving somebody's household goods,

24:55as things are being packed, is there a possibility

24:58that something gets dinged, something gets broken

25:00in transit or in placement?

25:02Maybe something got left in a room,

25:06and we didn't realize it, right?

25:07Mistakes happen.

25:08Movers are human.

25:11In the beginning, as the move is getting set up,

25:15there's a visual, a video survey of all of the rooms

25:19in the home that are done that feed into Yembo,

25:23so that at the time of the move completion,

25:27if something, a dresser drawer is broken,

25:30the employee can take pictures of that,

25:33and it will associate directly back to that original survey

25:37and be able to facilitate faster decisions

25:41for that individual, as well as our claims adjusters,

25:44to go, "Is this a replacement? Is this a repair?"

25:47It creates a huge open communication for them

25:53with transparency.

25:55Everybody can see the same thing,

25:57and it also is learning.

25:59So, there's definitely AI behind Yembo of,

26:02"How can this be prevented in the future,

26:04and what's the best case outcome for this?"

26:08So, the teams are loving using this right now.

26:11It's made a lot of improvements in their process

26:14and allows them to address claims a lot more timely.

26:18So, huge benefit, huge win for this partnership for us.

26:24Excellent. Thank you, Shaun.

26:27So, let's move into some of the other agentic solutions

26:31that we have built internally.

26:33Just to follow up on what Shaun was talking about

26:36on Power Automate first, actually.

26:38We've done a lot more around that area as well.

26:41We've got an AP invoice automation procedure,

26:43which we're not going to show today,

26:45but that deals with so many exceptions

26:48and so many crazy things that our company does,

26:52and Shaun was instrumental in that as well.

26:54But we are trying to use Power Automate and Power Apps

26:57and Power Pages too, across multiple areas

27:00of our organization to just make life easier.

27:04Process data quicker, get access to information

27:09and get access to decision-making much,

27:10much faster than we've done previously.

27:12And it ties nicely into what we're doing here

27:14with what we call knowledge scouts.

27:18We've got multiple of these across our organization right now,

27:20but the one you're seeing at the moment

27:22is very much focused on a test client that we have

27:26and test data that has been created

27:28just to show how could you surface information

27:31from policies, from exception reporting,

27:35anything that basically supports

27:37that whole transfer process.

27:40We get a lot of questions from a lot of clients

27:42as to what is going on with the move,

27:44how much have we spent so far,

27:46how does that come up against the budgetary aspects.

27:51A lot of information comes into our consultants

27:53and a lot of requests.

27:55Previously, this could be very much a,

27:57if you're on a call with a client,

27:59it could be, "I need to get back to you.

28:01I don't have that information right now."

28:04We're finding with the use of AI tools like this,

28:06where we can surface that data very quickly,

28:09we can positively impact those customer experiences

28:12because we can get the data straight away.

28:15It's a simple question to a chatbot basically,

28:18which is hosted on our SharePoint sites

28:21to pull that client level information

28:23and give that information very, very quickly.

28:26What you see on the screen,

28:27it's just a repeat of the same thing over and over and over,

28:30but hopefully it gives you an understanding

28:33of how we're deploying things like this

28:35across multiple areas of our organization.

28:38And now our supply development group, for example,

28:41have something on their sites as well

28:42as to getting to the information really, really quickly.

28:46We've got HR-related ones.

28:48We've got stuff in more of the operations areas as well,

28:51some in IT too, just to surface that information

28:55and try and get to the point of information

28:57much, much quicker than we've done previously.

29:00Again, coming back to what we mentioned earlier,

29:03we live in an on-demand world

29:04and people are expecting answers immediately.

29:07They don't want to wait for anything anymore.

29:09So deploying tools like this has really helped us

29:12to surface that data quickly,

29:15because as Shaun can tell you,

29:17we deal with masses and masses and masses of data

29:20every single day.

29:21Just a single move for a client could hit on multiple policies

29:26and each of those policies could be 10 to 20 pages deep

29:31with various information in there

29:32about eligibility criteria,

29:35what this person can get

29:36if they are of a certain grade in their organization.

29:39It's a lot of data and expecting our people

29:42to go through that and really know it in depth,

29:45probably an ask too far,

29:46but surfacing information like this using Copilot

29:50helped us tremendously.

29:52We actually had some surveys done with the teams

29:55that are using this and they found that they were responding

29:58far, far quicker, that they were able to get

30:00to that information far quicker.

30:02The trust in this actually increased

30:04because of the citations and the ability

30:06to actually get to the source data very quickly

30:08and double check, yes, it's accurate.

30:11That increased very, very quickly for us.

30:14One of the funniest things I found with this,

30:16we deployed it for a particular client

30:18and one of the comments on the survey results

30:21really interested me.

30:23A guy had responded,

30:25"The agent was able to find an answer to a question

30:27I've had for three weeks."

30:29Which is great, but my question backwards

30:32was why did you wait three weeks to get an answer?

30:34That just doesn't add up to me.

30:37But it shows you that it's finding things

30:38that some of our people maybe can't find

30:41because there's just too much information.

30:43So again, great use case that we've got out there.

30:47We're deploying more and more of these as we go.

30:50We see this as being a big part of our service delivery

30:52moving forward as well.

31:00Moving on to the next one,

31:00which is our request for proposal agent.

31:03So we have built this with our partner, GetAI,

31:07and they've really helped us to tap into that

31:10more heavy duty pro code solutioning

31:13using the Azure framework.

31:15So this is using, I think it's based on Claude now,

31:18did start off on OpenAI models,

31:20but we've migrated over to Claude

31:22because we just felt it was giving us a better experience.

31:25But the technology around this is great

31:27and certainly GetAI have helped us to approach that quicker.

31:32That learning curve is pretty steep

31:34if you're trying to do this on your own

31:35and you don't have much experience.

31:37So having a vendor like GetAI,

31:39or a partner I should say like GetAI,

31:41help us on this journey has really accelerated

31:44our use and adoption of some of these tools.

31:47But what is this?

31:49For us, requests for proposals are a very commonplace activity.

31:54We get over 100 of them every year.

31:57They come from various clients from various industries

32:00and every single one of them is very detailed.

32:04You're talking hundreds and hundreds of questions

32:07that need to be answered.

32:08And they cross multiple domains.

32:10It's not just about service delivery.

32:12It can be about financial strength.

32:14It can be about information security policies.

32:17It certainly is of late.

32:19How do we handle AI?

32:21How does it handle data?

32:22What provisions you have in place?

32:25You get the point.

32:25It crosses multiple areas of our organization.

32:29Finance, IT, operations.

32:32Everybody gets a piece of this.

32:34And the way we've been doing it previously is very much

32:38we receive a tender response or a request for proposal.

32:43We will form that out to multiple people

32:45to say help us answer these questions.

32:47Yes, we will leverage what we've done previously

32:49on other RFPs if they are within industry

32:52or very much within a similar mindset to the RFP

32:56that we have open.

32:58But it's a lot of time spent.

33:00And the other piece around this is we normally have

33:02very, very tight deadlines to go and respond to this.

33:06And obviously, as an organization,

33:08we want to put our best foot forward.

33:10We want to make sure that we answer everything

33:12as best as we can to give ourselves

33:14the best opportunity of winning that business.

33:18But it takes up a lot of time.

33:20And a lot of our people do not have that time either.

33:23A lot of other things to do.

33:25So what we've deployed here,

33:26and I'll just kick this video off actually

33:28while we talk through it.

33:30What we've deployed here is a tender agent

33:32that takes care of that for us.

33:33It's designed to create that first draft

33:37and move our people away from creating information

33:40and filling out questions to reviewing

33:43what the agent has done for us based on historic data,

33:47based on prior RFPs, based on industry knowledge,

33:51based on searches that it's done,

33:52and the ability that it can compile all of this information

33:56and really complete these questions for us.

33:59You can see on screen, it starts with SharePoint.

34:02It goes through all of the different formatting

34:04and that is there.

34:05It is linked to AI Foundry.

34:07We are utilizing multiple models to push this through

34:10and use content understanding.

34:13The real benefit of this is it's a chat interface.

34:18Our RFP team can just interact with the agents,

34:21ask it questions, ask it to refine documentation

34:24if we wanted to,

34:26ask it to reference other RFPs if we wanted to,

34:30and it will take that action for us.

34:32So again, moving away from that whole,

34:35we've got a new RFP.

34:36I need to give this to the InfoSec team.

34:38There's 20 questions on here that we need to answer.

34:42Instead, now we can say the 20 questions have been populated.

34:46It's now a verification of is the information correct?

34:49Is there anything we want to add to it?

34:51And that's a huge game changer for us.

34:53The ability to get this done quickly,

34:56the speed at which we can respond, it's huge.

35:00You can see here, it's going back and saying,

35:02please recreate the response for question two

35:04and reference something else, which it didn't do before.

35:08So really, really powerful.

35:10You can see the level of the information

35:11that it's putting together.

35:13Anyone who's worked with chatbots before

35:15or AI in general, which ensures everyone on this call,

35:19it gives you good information.

35:20It's how we want to provide that information

35:23and present it back to the client that is important.

35:27It's recommending a SME review here, which is great,

35:29because again, it gives us that targeted information

35:32to say there's something more that needs to be done.

35:35But this is a huge help for all of us

35:37who are involved in these processes.

35:39You can see more of the architecture here.

35:46It is using multiple components within the Azure AI Foundry.

35:52And then as you say,

35:53we get to that collaborative documentation

35:55where it is very much, we have an output file,

35:58but we can still talk to the agents

36:00and really verify and refine as we go.

36:05So a great experience for us for sure.

36:05So, we're going to go back to the next slide. So a great experience for us for sure.

36:12And one of the things that we found really interesting

36:15with this is the ability for it to take information

36:19from different industries and apply it.

36:22So we may get an RFP that is from a company

36:25who are very sustainability focused.

36:29We can actually have the tone of the RFP

36:31created with sustainability as the number one priority,

36:34just ask the agent to really infuse that throughout,

36:37which is again, if you're doing that manually,

36:39it can be difficult.

36:41But having AI generate that first draft for you

36:43is super, super powerful.

36:47So that's the first of the more,

36:48what we call the pro code solutions here.

36:52The other one is one we're very proud of.

36:55I think Graebel are one of, if not the first company

36:58to deploy computer use agents with Microsoft

37:01in the US at least.

37:03And we're using this to supplement

37:06a very heavy manual effort right now.

37:10So we work with a legacy system that doesn't really

37:14have modern integration capability.

37:17And it is difficult to work with.

37:19And we've had a lot of people undertaking manual processes

37:24for the creation of what we call service orders,

37:26which is effectively an instruction

37:28to our supply chain base to perform services

37:32on behalf of our clients.

37:34So I'll manage through this one legacy system.

37:37Again, our people are manually keying the states are in,

37:40or certainly were until we put this solution in play.

37:44Manually keying that data in,

37:45there's the opportunity for failures in there.

37:48Everybody gets something wrong every now and then.

37:52We can get incorrect information going in there,

37:55requires further review, et cetera, et cetera.

37:57But the biggest part of this is the volume.

38:00We can deal with over 150,000 service orders.

38:04Per annum across multiple clients with multiple

38:07suppliers attached to them.

38:09It's a lot of manual work.

38:10And the accuracy around it is very important.

38:13So what we chose to do here, with the guidance of GetAI

38:17again and the support of GetAI in building the solution,

38:20we deployed what we call our service order agents.

38:24So let's take a look at this one.

38:28So our existing process starts with a trigger from Outlook,

38:32which will pick up Power Automate and move that information

38:35into a solution with the agent.

38:39From there, we use content understanding to scrub that data

38:43and get it over to Cua, computer use agents.

38:47You can see on the right-hand side of the screen,

38:49it performs the steps that a individual will perform

38:52based on an instruction set configured with natural language.

38:57So everything that it does, it logs into the VM,

39:00it logs into our legacy system, it performs the exact same steps

39:04that a user would in creating a new entry,

39:07what we call an A-Pro.

39:10It'll create that new service order structure.

39:12It will go through the screen using the instruction set

39:14and using the information that's provided from the email.

39:18And it will start to populate all that information

39:20directly into the legacy system.

39:23Once it's done all that, the most important part for us

39:25is the human in the loop content here.

39:28So we will, the agent will spit out what it's done over email to say,

39:33perform the duties, I was able to get 90% of it done.

39:37Here's the piece that I was not able to complete,

39:39or it'll give you a full success and just say,

39:41everything's been completed.

39:43It'll also put all of that information into the legacy system.

39:47So we actually have a shell that can be sent directly

39:50to the supplier that we're going to issue this to.

39:53But we have not taken that step of allowing the agent

39:57to make those decisions for us.

39:59It's very much a data population effort for us right now.

40:02We have multiple different types of service order

40:05which have different structures, different areas

40:08within the system that it will contact.

40:10But the computer use agent is taking all of that manual effort away

40:14and really populating this information for us,

40:16which is a huge win for us.

40:19And it comes back to how are we deploying our resources?

40:22Can we take away the transactional elements

40:26and focus them more on client interactions

40:28and helping to improve that client experience?

40:33So really cool technology, it's been a tremendous help to us.

40:40So that's the end of the demos that we wanted to cover.

40:42We do have other ones that we, as I mentioned earlier,

40:46multiple applications of Power Automate

40:48and more of the Power Platform across our organization.

40:51We are doing more and more with agents every single day.

40:54But we did want to talk a little bit about where we see this going,

40:59some of the problems that we've had,

41:00and I'll pass it over to Shaun to take us through the key takeaways.

41:04Yeah, thanks, Steve.

41:05The first key takeaway is critical.

41:09AI is only as smart as your data.

41:11So we did not just create these solutions one day

41:15when we woke up and they worked perfectly, right?

41:18AI is only good as the data you have.

41:22So we've gone through a lengthy period of time

41:25where we've had our departments and key stakeholders

41:28go and look at the data that we have,

41:31make sure that it's not in four different places

41:33and with different file names.

41:35So we've done a data cleanup effort,

41:37which is really one of the first steps in the journey, right?

41:40You have to have that clean data

41:42because the AI is only good as what you have.

41:45So if you've got bad data in, you're going to get it out as well.

41:49So I can add a point there, Shaun.

41:51Sorry to interrupt.

41:52Couple of things on this one.

41:55The knowledge scout that we showed earlier,

41:58when we did the first iteration of that,

42:01it was a train wreck and it wasn't because of the AI.

42:05It was because our data was in such a terrible state.

42:09What it led us to believe though,

42:10or sorry, what it led us to realize is

42:14AI is great at exposing where you are weak.

42:16[LAUGHTER]

42:18So we were able to build that muscle and really say,

42:20how do we do this in other areas and clean up our SharePoint sites?

42:24I'm sure there's multiple people on this call who have this right now.

42:28Anyone in your organization have folders that say, do not use.

42:32Well, the AI is seeing those things.

42:34Archive, great.

42:36Let's put it somewhere else where the agent doesn't need to see it,

42:39or let's go into SharePoint and turn those things off

42:41so the agent doesn't see it.

42:44But we had a lot of that.

42:45Let's just plug it in, it'll work.

42:48Yes, it will work to an extent,

42:49but it will start to expose where you've got problems,

42:52which is good and bad.

42:54If you don't want to know where the problems are,

42:56then don't deploy it.

42:57But if you do, you best go and fix that to make this usable.

43:02The other thing that we've done as well,

43:03which we weren't going to touch on today,

43:05but Data Lakehouse is one of our major projects

43:08that is coming to a close right now.

43:11And the whole basis of that is we would have a complete clean

43:14gold-level data set that we can plug our AI tools into

43:19and really rely and trust that data to be fully accurate

43:23from all of our systems.

43:25So we're on the precipice of having that,

43:27and we're really excited about the capabilities

43:30that we'll have as soon as that is available to us.

43:33Sorry, Shaun.

43:35That's okay.

43:36And I'll wrap up quick because Lupe's playing the

43:39"get off the stage" music for me.

43:40[LAUGHTER]

43:43But all of this that we've shown has taken time.

43:49And it's all with the backstory of AI is so the front line

43:53story of everything that it's changing every day.

43:56We've had conversations with Microsoft of,

43:58"Hey, this is what's coming next,

43:59but it will change tomorrow."

44:02So making sure you start to develop an idea of what you

44:04really want to accomplish and then finding a partner

44:06to help you with it internally grow it.

44:09And making sure that you've got governance.

44:12The data is key, but also governance.

44:14If you just give AI to everybody in your team,

44:17who knows what's going to come out?

44:19So make sure that you've got those controls.

44:21And then finally, which is probably the most important

44:23takeaway, is AI is no good or automation is no good

44:28unless you've got the people behind it

44:31that understand the processes that support the implementation.

44:34So combining people and purpose with AI

44:37is how you're really going to achieve success.

44:40So we appreciate the time.

44:42Thank you for that.

44:43And Lupe, it's all yours.

44:46Thank you so much, Shaun and Steve from Graebel

44:50for sharing your AI journey and the real world

44:52innovation stories here with us at CAT AI webinar audience.

This transcript was generated from the captions YouTube publishes for this video. Get the transcript of any YouTube video atfreeyoutubetranscribe.com: free, unlimited, no sign-up.