How to use Squint to close the continuous improvement loop

Continuous improvement is supposed to be a loop: watch the work, find the waste, fix it, update the standard, then measure again. Every one of those steps runs on data about how the work actually happens, and that data is hard to get. Timing cycles and processing footage by hand takes so long that most teams settle for a single cycle, a quick look at the video, or nothing at all. That leaves them improving from a partial picture.

Agents change that. When the slow, manual steps take minutes instead of months, teams can run the loop as often as the work changes, not once every few years. Squint’s agents turn work video into a time study, find waste, and recommend improvements. And what you learn can reach the floor the same day, as updated work instructions operators follow right at their stations.

Here’s how CI teams are running that loop today.

Start with a full shift, not one timed run

Upload video of a full shift, not just a few cycles timed by hand, and the agent finds every repeating cycle. Tell it where a cycle starts and ends if your team defines it differently, and footage of an empty station won’t skew your times.

Already have footage of the job? Bring all of it. Add as many process videos as you have to a single time study, from whatever devices you used to film them, and the agent works through every one. The result is a time study built on every cycle you recorded, not just the few you had the time to clock. Is more than one person working the job? The agent gives each person their own timeline and cycles, so shared work doesn’t blend into one.

At one industrial components maker, consultants spent a week timing a single changeover. The operations excellence director ran the same video through the agent and had a time study in 10 minutes. He now uses it to back capital requests with video proof.

Tell the agent what only your team knows

The agent sees everything in the video, but not what only your team knows: the fixture that can’t move this year, the cell with a spending cap, the part your team calls something nobody else would. Add that context, and its recommendations will account for it.

Attach what you’d hand a new engineer on day one. The documents that help most:

• Standard work or the work instruction for the station you filmed
• Photos of the station: fixtures, bins, gauges, posted SOPs
• Machine data, like cycle and dwell times
• The problem you’re solving: an A3, kaizen charter, or downtime log
• Your takt or target cycle time, and your last manual time study

You don’t need all of it. One video and a sentence about your goal are enough to get started.  Upload them directly or pull them from your Squint workspace. No documents? Write your instructions in plain language instead. The most useful ones cover:

• Goal: the decision the study feeds, like cutting station 3 from 68 seconds to a 52-second takt
• Cycle: what starts it and what ends it
• Required, not waste: steps that look idle but aren’t, like a 90-second leak-test dwell
• Ignore: warm-up cycles, partial cycles, people walking behind the cell

Context isn’t a one-time step, either. You can add a spec or change your instructions after the fact and regenerate the whole time study, or regenerate just the stretch of the timeline that needs another look. Either way, you see what changed before you accept it.

A time study you can check and change

No engineer acts on a number they can’t trace. That’s why every classification and recommendation links straight to the moment in the video it came from.

Every second of video sits on a gapless timeline, split into work elements marked value-added, necessary non-value-added, or waste. Findings are ranked by how much time they cost, each with a root cause and a recommendation. If the agent got a step wrong, or your team sees it differently, drag a boundary, add a work element, or reclassify it. Your edits are shared, so everyone on the team works from the same version. The agent does the first pass. Engineers have the final say.

At Nailor Industries, an HVAC equipment maker, the lean team checked the agent’s analysis against a line it had already improved by hand. “We compared notes, and it literally suggested the exact fixture we’d already built ourselves,” said Zak Rynn, who leads the team. With that proof, his team ran over 40 analyses in its first week, and now uses the agent to find throughput gains on the lines it’s improving.

Talk to your time study

Every time study is full of answers. Usually you have to go find them: scroll the timeline, dig through the findings, build a spreadsheet, and piece the conclusions together yourself. Now you can just ask.

Talk to your time study in plain English, and the agent does the digging. It pulls the data, finds the moments in the video that matter, and puts the answer together for you. Ask it things like:

• What are the three biggest sources of waste, and how much time does each one cost?
• Where is the operator waiting, and why?
• If we could only fix one thing this week, what should it be?
• What should I tell my plant manager about this job?

Then keep going. Every follow-up builds on the last, so it works like a sharp engineer who watched every second of the video and is ready to think it through with you.

Find where the variation lives

The same job can run minutes apart from one cycle to the next, and the average cycle time won’t tell you why.

Group time studies of the same job across operators, shifts, or days, and the agent runs a variance analysis that:

• Shows the full range of cycle times
• Highlights which steps vary the most
• Writes up which differences are worth pursuing

Click any step to watch its fastest and slowest cycles side by side.

TechnipFMC’s robotic arm team wanted to cut 30% of the labor time on its builds, but its standards had been set by stopwatch years ago and build times varied widely. The agent compared the same build across many recordings and showed which steps varied, and why. The fastest builds already beat the old standards by 35%. “We always kinda knew there was waste in this process, but the Lean Manufacturing agent picked it up immediately, classified it, and put a number to the hours and materials wasted,” said Axel Gunnarson, a senior manufacturing software engineer at TechnipFMC. The team has since reset its standards around its fastest builds and started on its first fix: keeping every cell fully stocked.

Balance the line before you touch the floor

Put your station time studies in sequence, and the agent models the whole line. A Yamazumi chart stacks each station’s value-added and non-value-added time against takt, and stations over takt are flagged as bottlenecks. Table and flow views show the same line in whatever format your team prefers.

From there, set which work has to happen before what, group work elements that belong together, and rebalance to a target cycle time or a target number of stations. Then run today's line and the rebalanced line side by side, and see which one holds up before anyone moves a station.

At one global energy equipment maker, the lean team spent over a month building time studies for a gas turbine repair line by hand, then another month rebalancing it. From the same footage, the agent found the bottleneck: welders, the line’s scarcest skilled workers, were doing tasks other stations could handle. It recommended a rebalanced line that same day, freeing up the line’s scarcest skilled workers to do only the work they can do.

Make the improvement stick

The hardest part of continuous improvement is getting it onto the floor. Too many findings end up in a slide deck while the floor keeps running the old way.

Turn a time study into a draft procedure. The agent writes each step from the video, with the time it takes and where the savings came from. Your team reviews and edits it, then publishes. Operators follow the updated work instructions step by step, right at their stations. And the next time you film the job, the loop starts again. That’s continuous improvement that’s actually continuous.

Try it on your line

Already using Squint? Log in, record one job, and upload the video to see your first time study the same day.

New to Squint? Book a demo, and we’ll walk you through it on your own footage.