On-shelf Availability - How On-Shelf Cameras Can Augment Your Systems

Shelf cameras are emerging as a practical way to improve on-shelf availability by adding a new signal to help identify and fix out-of-stocks.

Colin Peacock and Professor Daniel Corsten discuss what retailers are actually trialling today, what works (and what doesn’t), and why combining multiple signals matters more than relying on any single measure.

They also stress that technology only delivers value when processes and people are ready to act on the insights, with store rhythm and prioritisation playing a critical role.

Cleaned transcript

How shelf cameras help to improve on-shelf availability

COLIN:

Welcome again to the ECR Retail Loss podcast.

This week I’m joined by Professor Daniel Corsten and we’re looking at the use of shelf cameras and images to help improve the problem of shelf out-of-stocks.

We’re often talking about how do you measure and monitor and fix on-shelf availability.

We’ve been talking about that for 25 years.

But for the last five years or so, we’ve seen the emergence of potentially the use of shelf images from cameras, whether those cameras are on the shelf, or on the staff badge, or on drones flying around, or even robots.

Where do we think the industry’s at on the use of shelf images, and where do they fit in the whole world of how do we improve on-shelf availability, in your mind at least?

What retailers are actually trialling with shelf cameras

DANIEL:

Well, I’m not going to start with my mind.

I’m going to start with the mind of the experts in retailing because a couple of months ago we did a survey and we had 55 retailers answering our survey.

We asked them about the importance of different signals that are related to out-of-stock, because measurements are related to the signals.

We had five, six signals.

We looked at gap scans.

We looked at inventory records.

Gap scans are fairly reliable, depending on the routines.

Inventory records are consistently unreliable, but at least they’re available and provide some rough estimates.

Then we looked at some of the newer things regarding inventory, like in-store picks for online orders.

That’s a pretty good measure, but it only counts for a part of the assortment and not for every customer, so it’s not really a general measure.

Over the last decade, we’ve had a few new entrants, which are the cameras, the shelf images.

Our survey showed that out of those 55 that answered, there were maximum eight or nine who had actually trialled and tested those shelf cameras.

There were slightly more that tested these fixed cameras than these robots.

Perhaps more in the United States tested the robots than in Europe.

Those who have tested robots say they’ve tested them in some stores and on some categories, but not at the penetration you would imagine.

We also had some who had tested them and thrown them out again, including some very famous retailers.

So where do we stand on this?

Apparently these robots are getting a little bit more traction.

I have to be careful because our samples are small and there’s a lot of marketing money and venture capital money pushed into this, so we may be getting a little bit too much good news.

That’s why I wanted to mention that some retailers got rid of these cameras and robots again.

Overall, they seem to be finding some traction because they require a little bit less investment and fewer changes in fixtures.

The robot just roams around, so you don’t have to equip the shelves.

On the other side, typically a retailer has only one or two robots in a store, so they cannot take pictures all the time, which some retailers think is not good.

Others think, well, we can’t act upon all these signals anyway.

So I see something positive happening, as an emerging technology.

What was your impression?

Shelf cameras as an extra signal alongside lost sales and other measures

COLIN:

To your point, there are traditional means and there are limitations with every way you measure this.

It feels to me that shelf images are very helpful in augmenting the existing ways that you’re measuring on-shelf availability.

A lot of retailers are using the lost sales metric.

Maybe two-thirds of retailers in the sample were using it, looking at a period when there are some sales and then there’s no sales, and something’s going wrong.

That’s more reliable when you have higher volume.

But the cameras could potentially help you pick up on some of those gaps earlier.

So it’s going to augment what you currently have.

DANIEL:

Generally, the more different signals you have, the better it is.

There are signals that come without additional hardware.

Gap scans, you have the people.

Inventory records, you have the systems.

Velocity estimation, that’s something you can pull out of POS data, but you do have to have a good algorithm.

We find there is a positive correlation between retailers that use multiple signals, compared with those relying on just gap scans and inventory records.

These velocity-based signals do help.

The cameras provide an additional edge.

Cameras have an advantage because you can do different things with them and extract different information.

You know the picture and you can check when it was.

You can see when the shelf was replenished, when it wasn’t replenished.

You can also work more on compliance of certain actions you want store associates to do.

No matter what technology you implement, it is how people act upon this technology that really makes the difference.

We found that in our survey data, and we know from previous studies that the more compliant the store associates are with the processes, the better it is.

Shelf cameras for auditing, remote checks, and planogram compliance

COLIN:

Your point about what the cameras can give you is very valid.

We heard from some retailers about having counting in the store, using a third party to do counts, seeing where they counted zero, doing double checks, and remote auditing of the quality of the check.

What gets me quite excited, with my previous CPG experience, is planogram compliance.

You can start to understand if there’s a relationship between planogram compliance and shelf out-of-stocks, and between planogram compliance and store sales.

You can start to do A/B testing knowing the planogram you’ve got in the store.

You could do these things much faster and more effectively with cameras.

DANIEL:

Planogram compliance as a driver of out-of-stock is something that we looked at in 2007 in a major study in Slovenia, together with a big CPG, and we did find a positive correlation.

This brings us to an interesting point.

I see a world where all these signals coexist, but for different purposes and different use cases.

On fast movers, on the centre aisle, these velocity-based predictions work because of the velocity of those products.

Then maybe in areas with more slow movers, think about cosmetics, maybe that’s where you could use cameras better.

Think about rotisserie and delis, maybe that’s where the robots can do a better job.

Then you have the in-store picks, and if you take all this data and triangulate it, you get to a better position.

But I still have to stress: one retailer told me they have a model where they say it’s only 10% technology, 25% process, and 55% people.

At this moment, we should not get too much in love with the technology.

I’ve seen retailers implement technology and realise the culture and processes weren’t ready, and they ripped them out again.

Technology is not a panacea.

But every technology will find its home.

Many people think about these technologies as labour cutting, a threat.

I’m not convinced that’s what’s going to happen that much.

We are running short on labour in many areas, and reducing it more could lead to loss of customer service.

Making people do a better job and using the time they have is much more valuable, and should be a priority.

Automate the hate: using shelf cameras to remove the worst tasks

COLIN:

One of the phrases one retailer came up with, which I really liked, was: automate the hate.

The jobs that you hate.

And gap checking is not an exciting job when you’ve got 24 foot of cheese, all yellow in colour, to check.

DANIEL:

You have to watch out that she doesn’t turn it around and then hate the automate.

That’s the knee jerk reaction of people when they hear automation.

Another thing I’ve been thinking about is that maybe you don’t have to have cameras in every store, or a robot in every store.

Maybe you create a sample of stores in different areas and sizes, and use them as an experimental panel where you try out new things.

Once you know they work, you roll it out large scale.

Then you have a test and control environment and can still adjust, because maybe you don’t need all of this.

We have to get much more clever.

Right now it’s a bit of a brute force approach in many areas, and we have to fine tune this.

Rapid delivery platforms increase pressure for better on-shelf signals

COLIN:

One of the things that’s going to create tension is the emergence of rapid grocery delivery: the Uber Eats, Instacarts, Deliveroo, Everli’s.

They’re going to ask retailers to be much more demanding in terms of what’s on the shelf and is it there when they place the order.

Right now maybe a third of all orders have a missing item.

So there’s huge lost sales for both the delivery platform and the retailer for not getting this right.

I think there might be more tension growing, which will help create demand for more signals.

DANIEL:

These platforms provide service and a value proposition that some customers love.

But they do expose retailers to operational shortcomings.

The person who orders knows it was ordered from retailer XYZ, so they associate the retailer with bad performance.

We looked at some of these platforms and did serious research, which led to some PhD thesis.

On substitutions and picking efficiencies, there can be much more done.

This idea of sharing and collaborating, sharing information about location and inventory levels, there is something that could work.

I heard once there are platforms that put out penalties for retailers who are really bad, because they’re not making their sales numbers because of the retailer’s inefficiencies.

Maybe that’s a call to action, an external nudging to say: if we work together and improve this, we make more money together.

The “central brain”: turning shelf camera signals into the right actions

COLIN:

We talked about this becoming one signal of many signals.

Gap scan could be another one, online grocery could be another one.

There’s got to be some sort of central brain in the retailer to bring all this together.

Where do you think that fits?

DANIEL:

From a technological point of view, this will always be distributed because it’s too much information to be handled.

If you put all that into some centralised science fiction thing, that won’t work.

It will create too much risk and people will be averse to that.

You need redundancies.

There are different companies trying to get more out of that brain, whatever it is.

Whether it will be a reinvented data company, an older company reinventing itself in the age of cloud, somebody from inventory management, or somebody from images.

There are lots of images: integrating images for shrink, for on-shelf availability, for general staff health and sanity.

Maybe we should be talking more to chief information officers, because we as an ECR group look at different buckets and maybe think too much in silos.

We should think about how all of this comes together.

For example, if we think about all that data on shelf availability, it tells us so much about substitution and complementarity that it should be fed back into forecasting and demand planning.

Sales is what you sell, and demand is what you could have sold had you been in stock.

We have to create more circular routes.

The same thing with self-checkouts: all that information, what do we do with it, and how does it influence other buckets?

In this new world of generative AI, we need to map out which functions will continue to exist, and how the pieces connect.

Store rhythm: why shelf camera alerts must match how teams work

COLIN:

Self-checkout is the gift that keeps on giving to the problem of inventory record inaccuracy.

The point about the brain was brought to life by one retailer sharing their journey on shelf images.

They said only 27% of the signals could be acted upon by the store.

They got 100% of the signals, but 73% had no action required because the product was being delivered, or was en route, or the supplier’s warehouse had burned down, or they had phantom inventory, ghost inventory.

So you absolutely need that brain, otherwise you’re sending people to do stuff that is very unproductive.

DANIEL:

Exactly.

I talked to two retailers.

In the early days, people were saying we want real-time pictures and to act in real time, but what’s the point?

You’ll just create chaos in the store.

There has to be a rhythm of the work that matches the rhythm of the signals.

If every 50 minutes there’s a new signal, you cannot have people running around chasing holes every 15 minutes.

These images have to be collected, and then at a certain moment you send out a wave of people at an opportune time, and then they fix these shelves and these gaps.

Otherwise, a lot of signals will be wasted.

I’ve talked to retailers that experimented with shifting the times they send people into departments, creating tasks and a rhythm where these things work together in a concerted way.

That leads to more efficiency and more productivity.

Just getting all these signals creates chaos.

It’s not good.

Closing thoughts and next steps

COLIN:

Every good podcast must come to an end.

We’ve got lots more work ahead of us in this working group, and it’s all very good stuff.

Hopefully we get more and more retailers engaged.

DANIEL:

The next meeting is coming up in a few days, and then we’re going to have one in July.

It’ll continue.

COLIN:

Exactly.

Well, thank you very much, Daniel.