What Changed in Machine Vision and What to Ask Your Supplier Now

Part two of two. Part one covered what I learned building and running a machine vision distribution business. This article looks at how the technology and the market changed over the same twenty five years, and what that means for the way you buy today.

The short version

• Going direct to a manufacturer rarely saves money, and it costs you impartial advice and independent fault validation.
• If there is no fundamental technical barrier, a genuinely new vision technology takes seven to ten years to move from the show floor to a routine purchase order.
• I called hyperspectral wrong. It solves real problems, but it is waiting on a cost curve rather than on the technology.
• The last five years have changed how machine vision is bought and sold more than the previous twenty put together.
• Product lifecycle is the least discussed subject in vision procurement and one of the most expensive. Vision components get designed in, not bought.
• The twelve questions at the end are the ones I would put to any prospective supplier.

Why write this part separately?

Because the two halves fail differently.

The first part of this article was about how a distribution business is built, and where it goes wrong. This part is about the ground it is built on, and that ground has moved. Some of it moved slowly and predictably, in the seven to ten year cycles this industry has always run on. Some of it moved very suddenly indeed, between 2020 and 2023, in ways we are all still absorbing.

If you buy vision, the second kind of change is the one that will cost you money, and it is the one least likely to be discussed in a sales meeting.

What happens when a manufacturer decides to go direct?

We had a supplier that had made an interesting journey of its own, from distributor to low end acquisition card manufacturer to low end camera manufacturer. They had a differentiated, lower cost strategy, and we were growing that line faster than any other brand we carried.

Then they approached our sales director and opened a UK office. We remained a distributor for a while on the stated basis that they were targeting a different type of customer directly. Within a couple of years they terminated us and went fully direct. They have since done the same in around ninety percent of the countries they operate in. We were doing about a million pounds a year on that line at the time.

Once they went direct they ended up with three staff in the UK.

That number is the whole lesson. Going direct looks attractive because the distributor margin appears to fall straight to the bottom line. It does not. An office and three people consume a large part of it before you count management time. What a manufacturer gains is control and ownership of the customer relationship, which are real. What they lose is breadth of coverage, cross selling into applications they do not make products for, and consultative selling by people who carry a wide enough range to recommend the right answer rather than the available one.

Many manufacturers now run a dual channel strategy, and it can work. It needs genuine trust and a clearly defined split, because without those it just creates channel conflict that the customer ends up paying for.

There is a myth on the buying side worth dealing with here too. It seems logical that dealing with the manufacturer gets you a better price and better support, because you have cut out a layer. In practice neither reliably follows.

On price, the manufacturer's cost of talking to you is the same as anyone else's. Someone still has to answer the phone, visit the site, and hold the stock, and that cost has to be recovered whichever badge is on the business card. The margin has not disappeared. It has moved.

On support, the difference is more important than the money. If your system underperforms, a manufacturer has an obvious incentive to conclude that their product is behaving as specified and the problem lies elsewhere. A good distributor will validate the fault independently, and if the product genuinely is at fault, they carry considerably more weight with that manufacturer than you do on your own, because they represent a great deal more business than your single order.

The same asymmetry applies before you buy. A distributor carrying several competing lines will tell you when a particular product is not the right fit for your application. A manufacturer will not, because it is the only product they have. And since no manufacturer makes everything, you end up running separate relationships for the camera, the lens, the lighting and the software, each of which will point at the others when the system underperforms. That is not bad faith. It is what happens when nobody owns the result.

Then there is the local support you actually wanted. Either those people are not there, or the margin you thought you were saving is what pays for them. Direct suppliers are also far less likely to offer the surrounding services that make your life easier: cable assemblies, pre-configuration, calibration, stock holding against your build schedule. Cheaper is rarely what going direct turns out to be.

I have set out the full trade-off, including when going direct does make sense, in Buying Machine Vision Direct vs Through a Distributor.

If you are buying vision: when a manufacturer goes direct in your country, ask what happened to the local applications knowledge. Sometimes it moves across. Often it does not.

How long does new technology really take to arrive?

Longer than the show floor suggests, and the interval is remarkably consistent.

Looking back over twenty five years, these are the shifts that genuinely changed the industry:

Standards. Analogue to digital, then Camera Link, GigE Vision and USB3 Vision. When I started, every system combination was effectively custom. Today it is close to plug and play. The standards work behind that, much of it unpaid committee effort, is the most under-appreciated contribution anyone has made to this industry.

CCD to CMOS. Driven by the consumer electronics industry's appetite for image sensors rather than by us. It delivered a large cost reduction at the same time as improvements in resolution, frame rate, sensitivity and dynamic range. Getting all of those at once is unusual, and it is probably the single biggest change of my career.

Frame grabbers. Once compulsory for every application, now needed only by the most demanding. And yet more frame grabbers are sold today than when every system required one. That single fact tells you more about how much this market has grown than any published statistic.

Smart cameras. Repeatedly predicted to kill the component market, and they never have. The pattern is the same as with frame grabbers. When an application becomes commoditised it migrates down to a smart camera, and a new high end application appears that needs something bespoke. Commoditisation drives the migration. It does not shrink the market.

Embedded vision. ARM based systems opened up higher volume, lower cost applications that were simply not viable before.

The one I called wrong was hyperspectral imaging. I expected it to scale into a significant segment and it has not, and the reason is instructive. Hyperspectral solves genuinely valuable problems, problems that nothing else solves. It just solves them at a price. What happens in practice is that you do all the feasibility work, you prove the application, and then the project dies at budget approval because the return does not justify the capital. That is a technology waiting on a cost curve, not a technology that does not work. Once the components evolve far enough for the price point to come down, it will fly. I still believe that.

Being wrong about hyperspectral taught me the rule I now use. If there is no fundamental technical barrier, a genuinely new vision technology takes around seven to ten years to go from being interesting on a stand to being a routine line on a purchase order. I have rarely been wrong about the timescale. I have been wrong about which technologies.

“If there is no technical barrier, a new vision technology takes seven to ten years to reach a routine purchase order. I have been wrong about which ones. Rarely about the clock.”

What I think is overhyped right now: the idea that deep learning will replace rule based tools. AI is solving problems that were previously unsolvable, and that is genuinely valuable. But AI only platforms will not address the full scope of machine vision software, and they carry a heavier cost of deployment than the marketing suggests. I would also put event based imaging in this category. It is useful, but it requires such a fundamental shift in how you think about a problem that I expect it to remain a niche technology for specific applications.

What I think is underrated: system design. Matching the right lighting and the right lens to the application. It has always been the foundation, and it is where most systems are won or lost long before anyone writes any software. If AI assistance arrives properly in the system design phase, that could be a genuine game changer.

Have the last five years changed more than the previous twenty?

Yes, and not because of one event. Because of a sequence of them.

Covid closed factories and delayed products. Home working and a spike in consumer electronics purchasing then triggered the chip shortage. At the supplier end, components were discontinued overnight and manufacturing capacity was reallocated to higher volume markets. Any company with the buying power went out and bought machine vision stock wherever they could find it. We saw a sharp dip followed by a significant artificial peak. Lead times on some components went from a couple of weeks to over a year, which was catastrophic for businesses that could not build without them.

Then came global tensions and a spike in interest rates, which put capital expenditure on hold. Six years later, capital spend and machinery volumes are still low in parts of the market, and companies that stockpiled are in some cases still working through what they bought.

The second order effects will outlast the causes. Component manufacturers had to divert engineering resource into redesigning products around obsolete chips, which cost the industry real innovation. Customers second sourced in order to keep manufacturing, and that gave Chinese brands the foothold in Europe they had been struggling to establish for years. Business development moved online, and online meetings have stayed.

2026 is seeing growth return to the industry, but it is cautious growth. Taken together, this is the biggest shift in how machine vision business is done that I have seen in my career.

If you are buying vision: supply chain resilience is now a supplier selection criterion, not a procurement afterthought. There is little point asking what happened to a supplier's lead times in the shortage, because nearly every supplier suffered. The revealing question is how they handled it: what they told customers, how early they told them, and what they put in place afterwards.

How long will your supplier support the product you designed in?

This is the least discussed subject in machine vision procurement and one of the most expensive.

Vision components do not get bought, they get designed in. An OEM selects a camera, builds the mechanics and the software around it, validates the machine, and then sells that machine for years. If the camera goes obsolete two years later, the cost is not the price of a replacement camera. It is a redesign: new mechanics possibly, new optical path, revalidation, updated documentation, and engineering resource pulled off new product development to redo work that was already finished. It eats resource, and it earns nothing.

So a supplier's lifecycle strategy matters as much as their price. What you want is a clear, published position on how long a product will remain available, early warning when something is heading towards end of life, and enough notice to plan a redesign into your own roadmap rather than react to it. Last time buy notices that arrive with a few weeks on the clock are not a lifecycle strategy.

This is also somewhere a good distributor earns their margin. They should have a process for tracking products through the lifecycle and actively notifying the customers who have designed those products in, rather than leaving you to notice a discontinuation on a website. Some do this well. Many do not do it at all.

If you are buying vision: before you design a component in, ask what the manufacturer's lifecycle commitment is, and ask your distributor what their notification process is. Get the answer to the second question in writing, because it is the one that will actually reach you.

How has the buyer changed, and what should you ask?

When I started there were no PCs on desks, no mobile phones, and in many European countries no working assumption that anyone would speak English. Distribution was the only sensible route to market unless you were large enough to put people in every country, and leads came from magazine adverts with reply cards and from exhibitions. It was field sales, all the way through. The internet then gave search and catalogues worldwide visibility, translation tools made cross border communication straightforward, and since Covid online meetings have made it far easier to operate internationally without local support on the ground.

What has not changed is that seeing an application in person is how you build real understanding and real trust. Trade shows have changed role rather than declined. They are now a meeting point rather than a lead generation exercise.

Younger engineers increasingly want to self serve, and there is a genuine market for that. But a web shop on its own will not work in machine vision yet, because the product remains complex and the sale remains consultative. Self service needs a knowledgeable team behind it who will help, and who will come to site when that is what the problem needs. Vision is still engineer led, and the best people selling it are commercially minded engineers. They are rare.

So if you are choosing a supplier, here is what I would ask.

1. Who will actually be my technical contact, and how long have they been doing this?
2. Can you tell me why this product is right for my application, not just that you stock it?
3. What will you tell me if vision is the wrong answer to my problem?
4. How will we approach de-risking this project, and who is doing what to define it?
5. Who owns the risk at the interface between your part and everyone else's?
6. What are the acceptance criteria, and what sample set will we validate against, in the lab and on the machine?
7. What is your support model, and what does it cost, in writing, before anything goes wrong?
8. What is the lifecycle commitment on this product, and how will you tell me when it changes?
9. Where is your stock held, and how did you handle the component shortage?
10. If I need something you do not list, will you go and find it?
11. What happens to my project if the person I am dealing with leaves?
12. Can you explain in thirty seconds why I should buy from you rather than anyone else?

Question four deserves a note, because it cuts both ways. A great many customers ask a supplier how they will de-risk a project while expecting the supplier to do all of the work, for free, with minimal effort on their own side. That is not a de-risking exercise. It is a transfer of responsibility, and it usually ends up back with the customer anyway. The useful version of the question is a shared plan with named owners on both sides.

The last one sounds glib. It is still the most revealing question on the list.

A closing note on Clearview

I have deliberately kept this article free of company promotion, because the lessons stand on their own and you can judge any supplier against them, including the ones I used to run.

It is fair to say why I chose to work with Clearview, though, because it follows directly from the sections above. When I look at a machine vision business now I look for the same combination: a genuine care ethic alongside real growth ambition, engineering depth that is funded rather than claimed, and people willing to tell a customer when vision is the wrong answer. That combination is less common than it should be. Where I find it, I am happy to be associated with it.

Measure them against those twelve questions. Measure everyone against those twelve questions.

About the Author

Mark Williamson has worked in machine vision since 1989. He founded Pinnacle Vision in 1997, which merged with Vortex Vision in 2000 to form Firstsight Vision, later acquired by Stemmer Imaging, where he served as Managing Director of the UK business and as a member of the group senior leadership team until 2024. He was Chairman of the UK Industrial Vision Association from 2006 to 2016 and sat on the VDMA Machine Vision Board from 2015 to 2024, chairing it for the final three years. He now consults independently with companies across the UK and European vision market, including Clearview. He can be reached via LinkedIn.