Seeing the SC analytics forest among the 'mushrooms'? and dirty data

Seeing the SC analytics forest among the 'mushrooms' and dirty data

Summary

This discussion was prompted by several separate conversations that highlighted a tension within supply chain analytics: on the one hand, there is a desire and pressure to be using data science and advanced analytics to better understand and address key challenges within supply chain; on the other hand, a lot of analytics projects don't deliver much tangible value. This was nicely put by one member describing 'analytics mushrooms’ (hence the tortured metaphor) i.e. disparate projects often creating dashboards for one particular focus, at the same time, making it difficult to pull them together to get an end-to-end view because the data behind it has to be right.

Supply chain analytics is generally recognised to be a critical capability but the resources required and, therefore, the risks involved are substantial with important dilemmas about when and where to apply advanced analytics, whether and how to focus on data cleansing first and whether to develop in-house resources or work with external partners.

Topics covered:

- what is and isn't AI?

- where & how to apply analytics, machine learning and AI to get tangible results? e.g. network design, transport / route optimisation, digital twins

- how much to leverage external partners versus build in-house capability?

- can any analytics be worthwhile until the underling data is cleansed?

- challenges with 'selling' the business case of data consolidation initiatives

- establishing the right collaboration framework with IT departments

- leveraging LSP relationships & control tower capabilities

If and when to use advanced analytics e.g. machine learning and artificial intelligence?

Sometimes it adds more complexity to the results, especially if it’s a ‘blackbox’ when it is hard or impossible to understand why or how the algorithm has produced a particular result. For example, with demand forecasting, you may get a ‘best fit’ result but, unless there’s clarity about what is driving the demand, it doesn’t give you any more actionable insight than you might get from statistical modelling.

It’s not very unusual for some analytics processes to run for 60 hours and not produce anything actionable, even with good computing power.

However, some problems are just too complex for anything other than machine learning / data science. The classic example of just a few trucks with a few delivery drops each produces millions if not billions of possible route permutations.

Data first?

Unless the data is right, the value of any analytics is limited, futile or, potentially, even counterproductive. The challenge is that data cleaning isn’t ‘sexy’ and is a challenge to secure budget and commitment for as it’s a means to an end…you don’t see an immediate return on investment.

On the other hand, you could spend months and years trying to get perfect data which is itself a moving target. The cost of not taking advantage of analytics capability until the data is ‘right’ outweighs the cost of the results not being perfect if it still brings useful insights and tangible improvements.

Is a digital twin the answer? But, a digital twin for what, exactly? Different functions will have different priorities and so might configure a digital twin differently so how do you avoid this and all the integration expense that comes with it?

Insource or outsource - what’s the right mix?

Insource pros

  • More ability to tailor the analytics capabilities to your own specific needs and circumstances;
  • It might not be data science per se but having internal team members who both understand the business and are data literate but don’t do the coding themselves can really help improve that focus on working on the right things in the right way. The challenge then is how do you build that data literacy capability?

Insource cons

  • Data science is a huge domain and very few data scientists will be able to be experts in forecasting as well as transport network optimisation, for example. Those that can, will command very large salaries;
  • It’s very difficult to hire the right people because whoever is doing the hiring needs to have a deep understanding of data science and its applications;
  • There's a risk of going to significant expense to build an AI capability on top of data, processes and systems that themselves are not optimal;
  • Even if you have good foundations and enough internal data literacy to be able to hire effectively, you still need enough scale to justify the overhead. The most likely areas where that kind of scale is possible is in analytics for forecasting, network and transport optimisation and, perhaps, inventory optimisation;
  • There are also risks around retention of talent and knowledge: you can invest quite a lot into developing people who are then headhunted for those skills, leaving a capability gap. Even if those people are retained, if they are not using those skills continuously, there can be capability attrition or knowledge obsolescence.

Outsource pros

  • Access to ‘best of breed’ technology and data science ‘know how’ that any single organisation alone is not going to be able to keep up with;
  • It is possible to successfully partner with external organisations if they have sufficient scale and structure relationships so that there is an ongoing commitment to reviewing and improving capabilities that they have supported.

Outsource cons

  • Hype: some solution providers are claiming to offer AI but, from a pure data science perspective, the claim can be challenged. Aside from, perhaps, automated reporting or advanced BI suites, AI solutions can’t be pre-programmed to deliver meaningful outcomes in many different contexts. Each problem is unique in its attributes: organisational structure, processes, how analytics feed into them and what the outcomes should be whether for inventory optimisation, network design or forecasting;
  • External solutions are typically very expensive and, unless you are very careful, any impact of their involvement quickly dissipates once they've moved on unless there is a plan for how to retain that knowledge and capability.

Summing up*

Of course, it's unlikely to be a simple either/or decision that is applied across the whole expanse of supply chain analytics. The approach should be adjusted on a case by case basis depending on the nature of the business, the resources available and the desired outcome. Some projects need continued or recurring resource, for example, planning and forecasting. Others may be one-offs such as transport network analytics ahead of a potential 3PL tender where internal resource might not have enough to do after that project is complete and so external resource may be more suitable.

*please note that this is our take on the consensus view of discussion participants. We're not experts or consultants and are not advocating one approach or another.


Anonymised discussion transcript

JP Doggett

I would like to start with N, if that's okay. N, just because we had a conversation a few weeks ago which I think framed this pretty well. Perhaps you could recap what we discussed and outline your feelings about what's the right way to do things, how things are being done at the moment, and the pressures that you feel influence the way that you go about supply chain analytics within XXX.?


N

Sure. I can quickly introduce myself and also give some background before we get in there. My name is N, 23 years of experience in supply chain across different organisations. Currently I lead XXX’s analytics for supply network design, transport optimisation, manufacturing, analytics, those are the areas which I cover and I've dabbled with analytics in different organisations, not only just with XXX, but with XXX and XXX. Although this is a new topic in many organisations, depending on the size of the organisation and the complexities, the implementations vary. So, I just wanted to gain from the team thoughts in terms of your respective organisations, how has this journey been? Are you doing things internally, are you leveraging vendors? What are some of the critical areas that you are tackling? Of course things like machine learning, artificial intelligence are getting a lot of attention.?

How much of that is hype, how much of that is real??What are some of the things that you are undergoing? Of course the other questions include what should be the ideal team size, should we build all of that internally, should there be partnerships? Those are some of the questions that I discussed and debated with JP, but more than happy to collaborate here and come to some a consensus as to how other organisations are approaching this.?


JP Doggett

Perhaps you'd like to start N by giving us your view on your own questions??


N

Sure, I could do that. I joined XXX last year. A year and a half prior to this, I was working with a company called XXX in Europe. I was instrumental in establishing an analytics team of around 125 people, largely based here in XXX, and I have also lived in XXX for almost five years. I moved out of analytics into different operational roles. It was a totally in-house team, not much supported by vendors. When I joined XXX last year, I was tasked with this whole thinking about how we should do it. Again, XXX is a different organisation to XXX. We started out by leveraging a lot of external partners and now we have built some of the skills which are essential and strategic in nature for ourselves here.?

The question which I often get asked is what's the right mix? What should we build internally, what should we have externally? Of course this is moving at a very high speed in terms of skill set so, of course, hiring is a challenge. Also, buying this from outside vendors is expensive. What we have done is we have a good mix of mainly advanced analytics like forecasting, network optimisation, inventory optimisation, transport, some of those we run internally. In the support structure (in terms of getting any technology support or surge resourcing), that we leverage externally. That's one thing that we have done and we don't plan to be a large team. We want to be a reasonably sized group.?

We are often challenged with the business needs and the technical understanding and the skills to deliver at very short notice.?That is something with which we continue to struggle. That is where I am at present. Does that help JP in terms of context?


JP Doggett

It really does. Thank you very much N. I'm going to try to step back and really just open it up to you guys to share how you're approaching it or even ask questions of N or each other. Please feel free to jump in, put your hand up either for real or in the reactions and pick up from where N has introduced us. Thanks, N.?


K?

Hi guys. K here from XXX. I think the hiring aspect is critical. I think this debate of external and internal, I think in some sense, it's actually not relevant to apply that judgement uniformly across analytics. It depends on the actual business case that you're looking to explore, what outcome you're looking to achieve. Certain projects would maybe require, I would say continued or recurring resource internally within the organisation, for example, maybe work around planning and forecasting. At the other end, maybe looking at a transport optimisation analytics project in the lead up to a 3PL tender, for example, would maybe require maybe less internal resource because you're just going to end up with a lot of latent time on the analytics team's hands after the project is done.?It may make sense to leverage more external capability to do that one off piece of work.?

I think analytics itself is such a big domain. N touched upon this topic of hype and I think that the problem that we have in the machine learning / AI space is that the discipline itself, there's absolutely nothing wrong with it. The problem is there's a lot of charlatans out there who are basically saying everything is AI. Honestly, with all respect, it's just a load of crap. There are actually very few software solutions out there that you could, from a pure data scientist’s perspective, actually say this is AI…it's just nonsense. You do not have AI solutions that are pre-programmed to go and deliver. Yes, there may be some level of automated reporting or maybe some advanced BI suite that I know a lot of large consumer goods companies have invested in.?

Each problem is unique in its attributes. It's unique in terms of the organisation, the processes that exist within that organisation, how analytics feeds into functions and their requirements, whether it's inventory optimisation, network design, forecasting, planning, and then you also need the right skill sets. Every data scientist is not going to be solving every single problem. Very few data scientists will be able to be experts in forecasting and then switch over the next quarter and then go into an optimisation piece on transport. Just doesn't happen. Those people are few and far between and they’re paid a lot of money because they're the one in the million.?

One FTSE listed company I know of insisted on recruiting internally and put together a team of seven people and it quickly transpired that they'd hired all the wrong people. Two of the guys got sacked, one guy went on sick leave two months into the project because he just couldn't handle the work. Who hires these people is another interesting point. It's like me: if you asked me to go and hire a competent doctor now, I wouldn't have a clue because I've not got any expertise in medicine and I wouldn't know a good doctor from a bad doctor.?

I think putting the pressure on functional leaders, whether it's in supply chain or in transport operations or finance or revenue management, to put the pressure on them to bring in expert data scientists and to disseminate the good ones from the bad ones, it's extremely difficult. If you don't have a central analytics team doing that piece of work, it's just taking a plunge and because someone seems to talk a lot of sense to people who are relatively uninformed on advanced analytics and AI.


JP Doggett

Thanks, K. You were getting a lot of nodding as you were talking so I think that's resonated. Who would like to reflect on anything that N or K said??


T

Yeah, it definitely resonates. AI, machine learning…I'm just tired of it. I know where you can apply it sometimes maybe in forecasting but I think often it adds to the complexity of the results and the fact that you can't actually recognise why the forecasting engine or whatever AI engine has come up with that result. For me, it doesn't really add much and you have to be quite careful to make the right decision on what tool you go with. We are going more into an internal, 'build it ourselves' frame of mind and this is leading us more into using the tools that we have and building something on top of that in order to try to fix our broken processes. Whether that's necessarily the right thing I'm really not sure because you're also spending a lot of money on maintaining this solution that we built ourselves. Not quite sure which one is the right way really because the tools are there in the market.?

The solutions are best of breed that you would be able to buy. Whereas, if you create it yourself, then you’re sure it's tailored to your own processes. But it's not best of breed. Change my mind please because I think it's perpetuating the bad processes that we've constructed ourselves over the course of years.?


K?

I think there's been a propensity, not just in supply chain but across consumer goods and other industries, to look at solutions with an outside-in approach to get the solution to fix the internal problem. I think, T, you're probably right to look internally because your problems for your organisation are unique to yours. The data that you have, the objectives that you're looking to achieve, they're just totally different. Trying to find a solution from the outside to just magically be retrofitted into your organisation is unrealistic. I think everyone's got an ERP horror story that they could probably share round the table or over drinks and there are just so many of them.?


D

Can I add to that? A couple of things: you mentioned forecasting a couple of times. I did a project with some university students looking at some history of our demand patterns and getting them to see what they thought of the forecast and one of the students said he’d really like to take that away and do a machine learning project on it which I was happy for him to do. All it concluded for me was that, basically, with many of our products, you can't forecast them. You can come up with a result from machine learning that says this is the best fit but actually it doesn't add any value because it doesn't know what was driving the demand, it doesn't have enough information to actually give you any more value in the forecast it produces than you can do with some perhaps simple statistical model or something.?

That, for me, really challenged the value proposition of any of the analytics that we might do. If I'm going to go and ask to invest quite a lot of money then what am I going to get in return? What is it actually doing for the business that will add value to the business? I think the more obvious focus for us, I'm not sure whether you call that analytics as it's a very big topic, is more about the quality of our own internal data first, which we do have complete control over and making sure that's absolutely right first. There's no point in sticking an analytical tool on something where the data is incorrect. I would say that's where the business sees value at the moment and that's where we're focusing on making sure that our own internal data is right first before I think we'll get any value from going outside and trying to work out what impact the weather has on our demand and if a ship gets stuck, what's going to happen??All that stuff sounds really great, but the reality of it is we've got our own house to sort out before we'll get any real value from that, I think.?


W

For us it’s a very similar picture. We've had what you’d perhaps call ‘analytics mushrooms’ coming up. I look after the end-to-end supply chain so everybody has their own little dashboards and trying to then pull those together to get an end-to-end view has been a nightmare because actually it's about the data behind that is actually the main issue. We use some fairly traditional front ends to display that, just slice and dice. To be fair, I would say we're still in the place of trying to get that to be more KPI or static data type of responses rather than adding your AIs and MLs over the top. We've got a few of those in some places. I would still question some of those, as we've been saying, around what does that bring, how do we use it? The process has to be right to be able to use that information in the first place.?

I came into this job with a few things ‘in flight’ which we finished off but which have taken three times as long as they should have done because the data behind is either not correct, it's dirty, we don't know quite what the source is, it gets translated somewhere in between or some variant coming out of ERP is pulling out something it shouldn't do and all of a sudden you're trying to backtrack through these numbers. It may be the boring bit but, actually, what we're standing back for next year is to say, ‘what's the layer of data that we need for all of our reporting end-to-end?’ That is actually THE bit for me. It's not the sexy bit, it's not the bit that everybody wants to put the money behind because you don't immediately see that return on investment.?

What you see, then, is ‘I have wonderful data there, now I have to spend the money to actually do the reporting on the top’. We do have data scientists, but for me, actually the ones that really help us to translate that are our business translators. They sit above the data scientists, outside of the coding, and more trying to help to make sure that we are optimising the way that we go about it. Trying to say, ‘have you thought about this way of looking at it?’ with an analytics eye but also with a supply chain view. Exactly as K was talking about, having that expertise in the area or at least an understanding of it helps to really bring the business and the IT together in the right ways and looking at the right things at the right times.?That is the exciting bit that comes after the data is actually where you would need it to be and we have a long way to go for that.?


K?

Yeah, I think the data cleansing, there's no point doing analytics on crap data because you're just going to get rubbish outcomes. I think building that base layer in terms of solid data, then reporting is an accentuation of good data, or hopefully it is. Above and beyond that, then you do analytics as maybe your tertiary layers, I'd say. That's where you then start to get the really sexy outputs where you may be looking at optimising cost-to-serve or improving OTIF KPIs because you're optimising against a certain parameter.

I think one of the best things we can do is, forget about the SC analytics agenda and focus on specific, tangible challenges. For example, can we reduce our logistics spend and what are the levers that we can use to do that??Yes, analytics is a core component of what we're looking to get to in terms of value but how do we actually get over the line in terms of getting the PNL saving is about empowering supply chain and logistics category teams, for example, to say to their 3PL, ‘we're overpaying 25% based on our analysis so what can we do about?’ It's about finding these pockets of use cases and maybe as time goes by and an organisation becomes more mature and more aware acutely of all their underlying problems in supply chain, then you start to build the confidence to say, ‘hey, we’ve actually we've got a lot of work here over the next five years’ and it probably does warrant us recruiting an internal team because there's going to be enough work for them.?

The worst thing you could do is hire the analytics people and then spend the next six months just trying to see if they can justify their positions within the organisation. In the short term, organisations like N mentioned, they do extend their arm to external providers, but I think having a long term dependency on external providers is risky and obviously a lot of companies have been burned with a consultancy doing that.?


JP Doggett

J, go ahead.?


J

Yeah, I'm just trying to reflect on what I've heard. The conversation seems to have flipped on its head actually to some extent because we started with saying, ‘it's so specialised, that you can't do it yourself’ to ‘actually, you need to do it yourself’. I do think that there is a significant role for outside providers to play because the technology is moving so rapidly. I don't think you can keep up with it internally. My approach has been to make sure that my people are data literate and can converse and challenge and understand what's been said to them.?

There is a way to communicate with the IT team. There is a way to communicate and understand what's happening with the external providers, but use the external providers to do what they're best at and provide a solution to help. So I don't think it's either or. We didn't create our own data analyst team largely because of lack of scale, even though it was a global business, we didn’t want to put overhead into that, combined with the risks around retention. Back to the people piece, this is talent and my concern about developing internally is we would end up with a dead duck relatively quickly because the people that understood the system and put it in, have moved on and we've not developed the internal talent or don't understand actually what they're really doing.?

To some extent it was a risk mitigation process to make sure I was partnering with people of sufficient scale so there would be medium to long term resilience with those partners. I suppose my sense of what I've tried to do is to develop literacy internally and a degree of competence but rely on external providers to provide us with that ongoing help. And partner with people that are not coming in to do a job and then leave, but partner with people that are providing a service that means that the dialogue goes on. One of the supply chain reviews we've put in, we're still doing three years on with monthly and quarter reviews with the consultants that helped put the systems in because they're developing things and bringing in external insights. Making sure that there is that blend of both is what I've found to be successful in the organisations I've worked in.?


JP Doggett

Thanks, J. That's really interesting. N did you have your hand up earlier and wanted to come back in?


N

Not to get very specific here, but some of the comments that I just heard are maybe suggesting that analytics is not required: we have smart people who can do it, and so on and so forth are similar to comments I heard when I joined XXX. My opinion is, I think, just by analytics or just by having smart people or just by having machine learning, I don't think we can solve any of these problems. It's a combination of all of those things and, of course, with good process and data. My perspective on data is, over the past 20 years, it is the same situation in every company. People always blame the data just like everybody blames forecasting. If you just fix forecasting, everything will be good and better. But that's where it is. We have to do with what data we have and we have to make progress.?

I'm surprised because some of the use cases or some of the problems that got discussed here in terms of transportation…it was the same thing here in XXX. People believed that we have smart people and we can do this in Excel and I can take some pride in this now: I don't know whether you've heard about this tool called XXX? It's a very specialised skill set to work on network optimisation and transport optimisation and we have saved millions and millions of dollars which has been recognised by the operational teams. While it looks simple to say, ‘why are we sending half a truck to this location? Why are we not doing multi-drop, why are we not questioning the MOQs?’, I really don't believe that just by having smart people or smart Excel could solve it.?

You need expertise and tools to do that. Of course we have data challenges. Data is not all green and clean with us, but we are doing what we can with what we have. That would be my viewpoint that there is no correct answer, it's a mixture of all of this. We can see that many organisations are making progress using data and using analytics and those skill sets. That's been the challenge for us because hiring some of these smart people and making them interested, because we are a manufacturing company, we are not an analytics company and we have to serve our business where it matters and where they can save money and improve service.


K?

I think on that logistics piece, you're right. I always have this analogy that I think is with ten trucks or five trucks with ten individual delivery drops each, yields about 36 billion possible routing permutations. If you start looking at a fleet a size of 200-300, from a computational perspective, it's just enormous. Trying to even talk about logistics optimisation is actually futuristic.?


N

Absolutely. Some of our models run for, with good computing power, 60 hours to optimise and some of them even after running for 60 hours, it turns out to be infeasible. That is a challenge.


J

I've done a lot of network optimisation globally, regionally and in country. The question is do you want to have those XXX-type skills embedded in your business because there is scale there to do that or do you want to get people in who can do that programming? I did a massive network optimisation piece in Asia and initially we had trained some guys on the optimisation view with the intention that next year they'll run the processes again. The reality is they're not using the skills often enough and lose them so, for something like that, my view was don't bother. If you've got a Global Centre of Excellence that goes across 70 countries and whatever else and there is the scale to do that, absolutely fine.

Be wary about training people in very specialist skills if they're not going to be using them very frequently because they'll just lose it. I didn't hear a lot of disagreement, N, in what was being said earlier? I think analytics absolutely is one of the major tools that you've got to provide insight into what's going on. Your example of the multiple billion of routings is a classic example where it's not just smart people. What you need smart people to do is to analyse and get the insight from the data, the insights that are being brought to you, so you can actually work out how you're going to solve the problems for the insolvable or whatever else. But they've got to be literate enough to be able to know what they're talking about. And that's an issue.?


N

Absolutely...fully agree with you. We decided to insource this. This was done by external consultants before for a year and a half. The biggest challenge was exactly what you mentioned. They used to cost us a bomb, they used to build a model which nobody understood internally and when they walked away, we were not able to take over and refresh it. So we have now totally internalised for network and transport optimisation. We have built our own capability internally. That's been a good story, but I agree with you as our scale is massive. We are looking at 120 plants, roughly around 300 co-pack locations, 130 DCs. Our network is fairly complex with multiple products. We have just scratched the surface in the last year and that's the reason we took the decision to internalise it and run it. Same with forecasting: I think forecasting explainability is a big challenge because there are many tools which give you a forecast but doesn't give you the breakdown of why it is.?

Most demand planners find it easy to do their own stuff. That is where we are going next. These two areas have worked out pretty well for us. Inventory optimisation again is a tough topic with lots of data requirements, but these three we have internalised and we are looking forward to doing more. That is the other challenge for us: we don't consider any of this to be either AI or machine learning.?

There's another topic called ‘digital twin’ which pops up in many of the conversations and seems to be the favourite toy of many consultants trying to sell us stuff. That could be another topic which this group can throw some light on as to what is happening in your respective organisations?


J

Just on the digital twin piece: my experience has been very positive.


B

Okay. Where are you using it?


J

In end-to-end supply chain management.?


N

Okay, very interesting to hear your experience then. Maybe we can pick up offline?


M

For us, we're at the beginning of this journey. We're exiting from what we thought was a global 4PL arrangement which turned out to be just lots of little XXX operations just cobbled together into one service. It wasn't end-to-end, it wasn't strategic. As we've started the transition, one of the questions we asked them is 'can we have all our data for the last ten or twelve years?' A lot of it wasn't available, it wasn't in the right format and it wasn't in any way, shape or form, usable. We're global by design and by nature and by customer geography. One of the opportunities ahead of us is we haven't exploited our supply chain data very well. We're a production first business, which means that the central responsibilities for the supply chain is ‘where's my stuff?’?

There hasn't been a lot of thought on the fact we’re on four continents, we’ve got 5000 or 6000 suppliers globally. We've got customers in almost every conceivable destination that we need to support from a range of considerations both from finished product and also from an aftermarket. For most of our customers there's a high premium on runtime so you've got to be able to support globally and next day. A lot of that information that's been running around the supply chain has been understood from a service level in terms of ‘we've got inventory plans, we've got some interim profiles that we can use to plan against’ but we've got a very low level of automation in terms of how some of the data moves. What we want to do is get to that next level of optimisation that says, right, this is how I'm operating on a day to day basis, how can I do this better??

We're going to double in size over the next five years, which is the question that we have in the background. Right now our approach is more people, more sites, more factories but we need to stop and reflect. We've got a range of problems: we've been growing globally for the last ten years so are we in the right locations, are we set up and structured in the right places in terms of the networks, the plants, the factories and then are we set up correctly to support customers? We're seeing that as an opportunity for analytics and the way we've started to do it is we're putting it in the 4PL set up so that it's a service that we can consume and if it doesn't work, then you take it out. You plug it into your 4PL arrangement and it's a service that's provided within that portfolio.?

We know we don't have the internal capability. That’s probably a question that I can ask to everybody: when you say ‘data literate’ - which is a challenge and it's something that we want to develop - what's been your route to getting your people data literate to be in a position to support, translate and do the work of making an external provider useful for the business? We know the high level problems but usually there are local considerations in terms of how you make these people productive, useful and so on and so forth. It would be useful if anyone’s got any thoughts in terms of what's your roadmap to getting people data literate because that's a real challenge, in terms of being able to cleanse your data, which is problem number one. Global business, different ERPs, different production systems and what we're doing is we're using our global inbound manufacturing process which is effectively the main way we stitch the global business together in that we have responsibilities that cross business units and entities which means we're well placed to integrate data.?

We're using a control tower process to say, right, data clarity first, link excellence second and then we can start talking about what's the role of analytics. We don't even have a unified data platform where we can actually do centralised reporting that's end to end. It's been hugely enlightening to hear everybody's journey. It looks like it's something you should have as a capability certainly if you're global and in multiple locations. The 'how?' is probably the trial and error piece. It would be useful if we had better information as to how do you genuinely exploit that capability. We've got enough data now running around the business and enough need in terms of what's going to happen over the business in the next three to five years to know what we need someone.

We’ve got someone who is a business intelligence manager in our current 4PL who is supposed to pull data together from disparate elements of the business to give us intelligence in terms of what we need to do to improve. At the moment, we just need somebody who is comfortable with spreadsheets and not really that analytics capability. We're at that point where we say, ‘do we recruit knowing we've got a range we're probably going to have at least three to five years worth of roadmap for that person to be able to drive improvements and optimisations across the business?’ We're thinking person first and then the tools would be part of what we're doing in terms of the control tower and the end to end visibility, the digital twin. We're going aggressively after that because we think there's value in that: here's my physical supply chain, here's my commercial supply chain and here's my digital supply chain, which will drive optimisations commercially and in the physical supply chain.?

It would be useful to understand how other people who are at the early stage or who've got more experience in terms of going from nothing to something, what that journey should look like and potentially some potential partners. I know we've mentioned XXX and XXX. Those are specifically network design tools. Anyone with a supply chain focus that's more around looking at something holistically and trying to figure out how that visibility should best be driven from a big data pool?


D

I think we've got two independent activities going on. One is to build that common data layer. We've only got a few ERP systems, but even two is enough in terms of harmonising the data. There's a, I would say, very IT-led activity to build a common data layer. We're using XXX as our data platform and one of the things I sensed immediately was everyone's like, ‘oh, can I store this in there? Can I put that in there?’ We've got quite a strong gatekeeper in there who's saying, ‘no, this data in here has to be 100% reliable and controlled and it's not a free for all where everyone can start just opening up and storing things in there’ like they did in XXX or XXX or whatever. I think the philosophy there is to make sure that data set is completely reliable if you want to do reporting on top of it.?

However, from a supply chain visibility perspective, that doesn't help me. We're in the middle of implementing XXX to be used as our global supply chain planning system. We've chosen it for particular reasons in the nature of our business. You get into this thing, everyone's saying ‘we've got end to end visibility of the supply chain and some people are looking at shipments and freight and we can analyse all this’...so we ask ourselves, is XXX a digital twin of our supply chain? It might be to us, but to other people in the business that are looking at physical items that we make and tracking their performance out in the world, that's a different digital twin. Yeah, there's more than one type.?


M

There is.?


D

I know it's drifting off the topic of analytics but I see, ‘yes, we'll get a picture of our supply chain’, but when it comes to managing the transport or something, do we add that into XXX or do we then have that as a separate system? Every time you add on a system, you have to integrate into it and that is really expensive. The more systems you have, the more integration you have, the more it costs in overheads and, as you say, all the people that then manage those processes, do you have them internally or externally? So that's my experience where we're going.?


JP Doggett

Thanks D. N, you've got a great emoji going on there. I think it's a shocked emoji?!


N

It is a not shocked emoji! What you just said is the same experience especially in large companies which have a global plus a BU plus a regional structure, like we have. My opinion on this one is there has been tremendous efforts by external vendors to sell IT stuff. You call it the data lake, you call it big data or digital twin, whatever. There is massive investment and the common problem that everybody has in terms of getting the data right, putting it in the place…the big question is, how do you monetise it? That's where the story goes very weak because unless you have a very strong use case to say, okay, this is where it drives productivity, like network optimisation, transport, inventory or forecasting or whatever you have in mind.?

Unless that links very strongly with the data infrastructure and the investments that you're making, what happens is...I often call it the ‘Taj Mahal’. If you visit India, the Taj Mahal looks great, it's a splendid monument. What we really need is the kitchen and the garage to start with, which you can't find in the Taj Mahal, neither to cook nor to park your car. That is the current state of many of the organisations. So we have taken a different view. Of course, we have also made those investments about which many people are now scratching their head in terms of like how do we monetise this? We are saying, okay, even if not 100%, at some point we will use all of that data. What we have realised is the IT stream works on its own path, the business streams works on its own path and we come in between which we are the bridge between the business and the IT teams to find the right value.?

That's where we have also been struggling because there is a lot of data which has already migrated, but it's neither in the right format nor in the usable way that we need for our modelling and often we end up transforming all of that. This just makes me feel better because our data scientists and our functional consultants within my team often get frustrated because there is so much talk on all this data lake and all that stuff, but most of the things that we need are not usable.?


JP Doggett

Thanks N. We're coming towards the end of the 60 minutes that we've got, but we do have time for one more contribution, question or comment before we wrap up…anybody??


D

I was just going to say on the data scientist question, somebody said to me, do you have somebody that's a supply chain expert that adds data science to their toolkit or do you have somebody that started as a data scientist and tries to become an expert in supply chain?


K?

It's a good question, D. I think my experiences a couple of years ago, we went through an intensive recruitment exercise and we must have interviewed about 150 people over the course of five months that had PhDs from all the top universities. We only ended up recruiting one. Those are people that dedicated literally their entire life to studying analytics and data science. I think to try and get someone to come from supply chain and then become a data scientist, unless they come from some very rigorous mathematical background, it's going to be very difficult for them to pack a punch in analytics, that's just my experience.

I think you have to have that core discipline of data science behind you and then pick and choose where your ability lies, whether it's with pricing, analytics, consumer insights, whatever it is.


T

Though I've seen it also in the organisation because my background is in software. If you recruit analysts or software people to go into logistics and supply chain, it's quite a niche, it requires a lot of niche knowledge that might be very different to data analytics and marketing or finance and often supply chain doesn't pay the most. You just get the really clever or the analysts or the data scientists then just move away from supply chain and that's the flip side, then you just lose them again.?


K?

It's true still all the top data scientists generally end up in the financial industry.?


N

Fully agree to your comments. I think we have been more successful in getting the smart supply chain people who know the operations and the business who are interested in data science to make the investment and upgrade them on the tools and the technologies rather than the other way around. I fully agree that supply chain is low paying in most companies in most countries so we lose talent, but the supply chain people appreciate it. It's a skill upgrade for many of us, so we stick around. My closing comment, JP, is thank you for scheduling this. It just started out as a small two minute conversation between the both of us but I’ve really benefited from this conversation, Thanks to the group for making time for us. Thank you.?


JP Doggett

Thanks to everybody for joining. Hopefully, we've managed to cover at least some of the points on the questions that we outlined at the beginning. There are more areas to explore, but I'd be very happy to schedule follow ups, maybe deep diving into some of the particular areas. I'll drop you a line shortly just to get your thoughts, and if you do have any other areas or aspects of this discussion you'd like to explore further, do let me know.?

www.bestpractice.club

要查看或添加评论,请登录

JP Doggett的更多文章

社区洞察

其他会员也浏览了