Guest Post: Takt Time Competition

By Jon Miller | Post Date: April 16, 2008 3:51 PM | Comments: 23

Jon has blogged about takt time many times before. Most recently he offered a free Excel based takt time calculator.

The tricky part about calculating takt time is normally associated with what to use for the daily demand number.

Do you look at the past 12 months? Do you average things out or look at the median? Should we plan for peak demand or use buffer and safety stock to protect us from these swings? What about including sales forecasts, which we know are always wrong?

So, you see, this takt time calculation is never quite as easy as the lean books make it out to be.

With this said, I’d like to challenge the fine readers of Gemba Panta Rei to a little takt time competition.

The picture below represents some completely fabricated weekly sales data (in units).

You can click on the image to download a larger version if it is too small to read.

UCL stands for Upper Control Limit and LCL stands for Lower Control Limit. If you need a refresher on control charts, please check out this three part series (part 1, part 2, part 3).

Let’s assume the following:

• Company makes one product
• Company runs 1 shift
• Company works 5 days per week
• Company works 9 hours per day
• Company allows employees to take 30 minutes for lunch, 30 minutes for coffee, and 15 minutes for planned downtime.
• This is all the data they have available

What’s the takt time?

[Note from Jon: From the list of people who give serious and thoughtful answers to this question we will draw two names at random to be the lucky winners of the upcoming book The Illustrated Toyota Production by Ritsushi Tsukuda, to be published by Gemba Press very shortly.]

After thinking this a bit I think that good takt time would be 57.4 seconds. That is aiming for 85% efficiency on available working time. And that should give company about half on hour extra each day that could be used for implementing kaizen and catching up backlog if there is need for that.

In addition to that, I would start with some safety stock, say 100 pieces which would be decreased as company gains experience on lean. Or not, if sales volume can't be leveled more efficiently.

Naturally company would have to follow sales trend to see if figures need adjustment. Are those two spikes anomalies or are they signal of long term increase in sales.

Poster: Panu | Post Date: April 16, 2008 11:57 PM

Dear Jon & Ron
Gemba takt time competition is a great stimulus

Thank you for raising an interesting dilemma we all face everyday
Simple tact calculation is total seconds available per year divided by the total annual demand (7.75 x 60 x 60 x 5 x 51) = 71145005 /510000 (1000 x 51) = 139.5 sec
It fairly easy to calculate the tact time, the difficult bit is to establish customer’s true demand.
In the example you provided there is weekly fluctuation shown so on that simple tact time your customer demand wouldn’t be met on several occasions in the year.
It is very important to understand why the customer has demand fluctuation and it will be worth trying to smooth it out as a first step. Checking the customer’s order policy, lead-time, stocking and inventory requirements and their own customer or end user needs might highlight smoothing opportunities.
The graph shows an upper demand limit of 215 and a lower demand limit of 184
Managing the variations in demand efficiently and effectively at the least possible cost is the clever bit. Using Buffers, Inventory, Flexible working etc could answer that need if absolute smoothing cannot be achieved.
Taking account the annual working policy for both you and your customer is important
• 51 week working only closed at Christmas
• Closing for Holidays I week Spring, 2 weeks Summer, 1 week Christmas.
We have flexible working for each production here, line working & holiday patterns are decided based on what the customers requirement is.
I have made a PSI plan (Production, Sales, and Inventory) based on the demand graph you provided I’ve assumed this a future forecast requirement.
I have set my policy as always able to supply the customer. I can send a copy to you by e mail if you want it, just let me know. It shows a minimum end of week inventory of 0 and maximum of 280 the average is 119 units per week. I have set a starting inventory of 50
Actually there is no correct answer to your question because it depends on a number of issues specific to your business and the service level agreed with your customer.
Best Regards
Chris
PS I'd very much like the book but I've run out of budget

Poster: Chris Nicholls | Post Date: April 17, 2008 7:45 AM

The simplest answer to this question, based on the available information, is to base a takt time on running an average of 1000 parts per week (200/day) and plan on keeping ~40 pcs buffer stock to account for higher weekly demand. That would give a takt time of approx 140-144 sec/pc, depending on if the planned downtime could not be taken care of during the other breaks. The 1080 piece demand weeks (what appears to be week 45 & 47) could be covered with overtime.

I know the bullet points say this is all the data that they have available, but one of the biggest questions I have come across in calculating takt time is who dictates the demand. If you are an automotive supplier sending parts to an assembly plant, you don’t get to dictate how many vehicles they want to build. In that case, you have to keep the buffer stock and potentially make it larger to protect yourself. If you are a manufacturer of a high demand consumer good, you may be able to dictate that you are only going to make 1000/week and carry no buffer stock. (I’m reminded of something attributed to Ohno that said not to be afraid of a lost sale if it costs too much to fill…I can’t find the exact quote right now.) The other piece of information that would be critical to planning production would be to understand the cost of inventory and the cost and availability of raw material. If you are extremely tight on cash flow, but have an easily accessible supply of raw material you may want to plan to keep no buffer of finished stock and base your takt time on some of the lower demand points (approx 960/wk) and pay OT for when demand picks up.

Poster: Joe Wilson | Post Date: April 17, 2008 9:02 AM

Ideally, the takt time should be based on the demand for that day.

Given some of the variation we need to take into consideration, we could take the approach that:

A nine hour day would yield 32,400 available seconds. After breaks and planned downtime, this becomes 27,900 seconds of available work.

The average daily takt would be: 139.5 seconds.
At the upper end (UCL) we would have a takt of 129.7 seconds.

Between the average (mean) and the UCL there is less than a 10% difference. I would attempt to control this variance through a well designed process/layout, and adjusting staffing requirements accordingly. With the exception of the two anomalies going outside the UCL, the variation shouldn’t be a significant factor as none of the other data points are near the UCL. By constantly monitoring the process, as they have been doing, any trends or shifting of the mean should be quite visible. This gives manufacturing an opportunity to respond appropriately.

I would probably base the chart on a rolling six-month cycle though in order to maintain current information. There needs to be a balance as there may not be enough resolution changes if one uses long periods to plot data, and too short of a cycle may cause unnecessary reactions to anomalies.

Those peaks could be anomalies due to a particular marketing/advertising program that the marketing program failed to notify manufacturing about.

Poster: Wayne A. Marhelski | Post Date: April 17, 2008 10:19 AM

After breaks, there are 27,900 seconds per day.
Based on the average of 200 parts per day, this gives us a takt time of 139.5 s per part (same as above).

I wouldn't worry too much about the UCL, since this is a 1 shift, 5 day operation, any peaks can be accommodated by working extra overtime, either an hour each day or a weekend.

Poster: martin | Post Date: April 17, 2008 12:11 PM

Thank you Ron for posting this. First, it's nice that someone has acknowledged the fact that real life does not always follow book examples. Second, for challenging us to think!

Assumptions/Clarifications:
1)These calculations are of true takt, not planned or effective cycle time.
2)Employees don’t punch out for lunch or breaks so we should subtract 60min from the available working time. (This can be a good question to ask depending on different company’s definitions of “We work 9 hours a day.”) In other words, 稼動 does NOT equal time spent in the cafeteria.
3)Looking at the data, it appears that the average is stable, assuming this stability I have chosen to produce to average demand and buffer with inventory. However, Panu raised a very good question as to whether or not demand was increasing. We should find out if these were seasonal spikes. They occured the 2nd and 4th weeks of November (roughly). Was this holiday related? Or perhaps an end of the year promotion meant to generate revenue to help “make the quarter/year”? Both factors to research.

Based on these assumptions, I have calculated takt time to be 139.5 seconds or 2min 18 sec. Calculations were as follows:

Total day – lunch – coffee – planned downtime =
9 – 0.5 – 0.5 – 0.25 = 7.75 hrs available

7.75hrs * 60min * 60sec = 27,900 seconds available per day

27,900sec ÷ 200 units per day = one unit every 139.5 sec

Thanks again Ron & Jon and I look forward to whatever discussion stems from this.

Erik

Poster: Erik | Post Date: April 17, 2008 12:25 PM

You are very welcome, Erik. I am glad you liked the post.

Poster: Ron Pereira | Post Date: April 17, 2008 1:07 PM

Now that I read other replies I realized that I made mistake on my calculations. So that throws my suggestion out of the window.

After revising my numbers I'd suggest takt time of 130s. Again that gives around half an hour extra each to be used productively, either to catch up or for improvement activities.

What do you think about allocating time each day for kaizens? Worth it?

Poster: Panu | Post Date: April 21, 2008 1:06 AM

Jon or Ron,

Can you please explain if we can (or even should??) link our newly calculated takt time to the actual pull of the customer? That is to say, in your example above we've seen the average rate of demand graphed and we turned it into a drum beat of 139.5 seconds.

Does this mean we are not giving production instructions (in the form of production kanban) to the floor? If so, are those cards directly related to what the customer pulled from FGI? I know I am walking the line here between level & un-level pull, but that's why I am asking. I'm wondering about the boundary between the two.

If we produce strictly to takt and buffer with inventory, the law of averages should "protect" us and we'll produce 200 units/day even when the customer pulls more (or less). However, my sensei warned me against this saying "Very dangerous to start this way." and advised us to start with straight replenishment pull. But this of course is unlevel and each day places different demands on the cell/line. Not to mention what happens if multiple models run down the same line.

If you could shed some light on where to start, transitioning from one system to another it would be greatly appreciated.

Thank you,
Erik

Poster: Erik | Post Date: April 22, 2008 8:51 AM

It's certainly important to understand how to calculate takt time correctly, but most organizations should probably spend far more time in the beginning on pull to prevent overproduction and link material and information movement, and on creating connected flows in smaller lots.

Another question we should ask in regards to takt time is "what is the business model?" Is the product built to order, and essentially pre-sold? Would making it early and putting it on the shelf result in lower costs than attempting to flex resources up or down to match demand exactly to customer orders? Or is the product made to a forecast and shipped from stock? We don't know in this case.

We also need to consider the cost of goods sold. Is this a low cost country or a high cost country? Does the product consume expensive raw materials or components to produce, or is the raw material cost comparatively low, while the manufacturing value-added is high?

Takt time production doesn't work so well without averaging the mix and volume through heijunka. It's not clear in the scenario above whether changeover times are sufficiently short and processes are reliable enough to do heijunka to smooth out the demand and have a more even takt.

The answer to these questions would also change our production strategy and how closely we would model Toyota's takt, flow, pull of just in time production.

Poster: Jon Miller | Post Date: April 22, 2008 12:15 PM

Jon,

Sorry for the very general question. Thanks for the truly Socratic answer. I'm sure the answer I was looking for is in there somewhere but I'll only get it if I really think about it. (Reminds me of Mr. Ohno's "You're a bigger fool if you do exactly as I say..." comment).

Poster: Erik | Post Date: April 22, 2008 12:53 PM

I had to chuckle (actually I LOL) at Jon’s answer and your response Erik. If Socrates had asked his students the riddles that Jon just placed, his students would have given him the Hemlock and saved the leaders of Athens the trouble….
As to the problem at hand, I am a very old school industrial practitioner and I have found that if I start too complicated nobody gets it, or has the endurance to try for very long. So for this problem I am going to assume (I hope the ass in this statement forgives me) that we are in a beginning company, they make one part and have an old school mentality and workforce who are not going to jump through hoops and change their work schedule daily. I am also going to assume it is cheaper to hold some buffer than to pay the wages of workers only partially utilized as some may suggest to meet spikes. I always hate to teach workers that it OK not to have enough work, especially before we have the TPS type culture in place (this starts at upper management level). My ideal situation is to level production for the mid-long term, usually at least quarterly at first, and then watch projected changes being made at my customers. Since we supplied Ford in my history, I am used to unstable (in many ways) customers. I would level my production to the average, being 1000/wk. For this problem I agree with the previous answers that this would be 139 seconds the .5 seconds is rounded down (I am not going to expect my hourly people to worry about 10ths or 100ths of a second). Of course, since my customer is not level, I need a buffer store. I develop my store by setting my weekly-expected production at 1000 and the adding or subtracting the customer requirements for that week and developing a cumulative sum for the month:

Et cetera until I have the full year done. The maximum I go negative is the amount of buffer stock I will put in storage at the start of my changeover to level production. In this small example, I went 75 negative so I would start with 75 in the buffer. I will run my system using pull and flush the buffer on a standardized basis. Of course I would have a safety stock locked away for downtime and emergency issues, and would do a 5-why and countermeasure if any have to be used. I think that is enough detail for now, I don’t want to over explain this though there is much more that needs to be learned before we get to a level pull system….. Takt = 139 seconds.

Poster: lester | Post Date: April 24, 2008 7:07 AM

Jon,
I tried to include a little jpeg showing the cumulative sum matrix, but your site practices safe posting, so it would not take it.... I am not sure if you would want to include it, if you do drop me a note at lester@braynpool.com and I will send it to you.

Best,

Les

Poster: lester | Post Date: April 24, 2008 7:17 AM

So, would it always be bad idea to aim for having half an hour (or any other amount of time) of planned slack in work schedule?

And if not, what kind of situation would warrant use of planned slack?

Poster: Panu | Post Date: April 25, 2008 1:06 AM

Jon,
Thanks for posting this, the takt time is 140 seconds. I would caveat this that this is the ideal and to problem solve, going forward to achieve this type of leveling.

What would make this exercise useful is to let us know the machine & actual cycle times to figure the number people required to produce this takt time. To really spice it up, give us a pack-out quantity to figure our takt-pitch time.

Thanks also for the excel takt time calculator. I really enjoy these type of exercises.

Best wishes,
Justin

Poster: JWDT | Post Date: April 26, 2008 9:51 PM

Panu,
I am not sure what you mean by "planned slack" in term of manufacturing workers. If you mean should we always give workers an extra half hour daily of time that they can use to make up production because I do not have a level schedule then I would say no. If you mean should I give the workers time to do planned activities then I would say maybe, depending on the activity. I am a proponent of level schedules, so production should be planned to be a a pace that satisfies my customer requirements daily on a level schedule. Buffer absorbs the customer fluctuation and safety absorbs my process variability. As I move away from a straight manufacturing operation into a even more variable situation such as service, then my utilization of the people will decrease and I will have more slack. Not that I want it, but because I accept Spearman and Hoop's teaching from "Factory Physics" on variability and utilization. Tom Demarco has a very good book on utilization of people in variable positions title "Slack" that discusses the good points of slack. But remember, slack is a waste. It has cost and often adds no value unless it is planned and managed well. The planning and managing is where most companies loose the thread.

Poster: Lester | Post Date: April 28, 2008 7:47 AM

Lester,

I meant planned slack in a way that if there is no production to catch up to (which there ideally shouldn't be when we level load) the extra time would be used for kaizen activities, 5S etc.

My experience is that biggest obstacles for continuous improvement is that line workers don't have time to do it. Because they have more important things to do. Which they of course do, getting products to customers is most important thing to do.

But. if we take time for kaizen into account when calculating takt and leveling load then we can take that 'time' problem out of equation and work on next obstacle.

Poster: Panu | Post Date: April 30, 2008 12:49 AM

Dear All
There have been some comments made regarding operators' time allowance to do Kaizen. We had a very similar issue here, when do operators have time carryout kaizen ? when they are making products every minute of their working day. In the begining we found it necessary to make a one hour production stop every week to kick off kaizen in the Gemba. This is a big costly committment but essential for training and explaining Kaizen in the first place. Then using this planned time allowance for actually doing kaizen and making an improvement for themselves. This initial phase lasted 6 months but cost thousands in planned downtime and unrecovered labour charges. No business these days can plan to stop production every week for ever to do kaizen. Following the initial phase we arrange to do kaizen when the line stops for other reasons such as parts supply problems, equipment breakdown and model change-overs. Now we plan flexible kaizen activities for the operators if the line stops for these reasons. Also some keen operators do kaizen at breaktime or in overtime.
Nevertheless each individual organisation has to workout the best way for them to allow kaizen activities to take place.

Best Regards
Chris

Poster: chris nicholls | Post Date: May 1, 2008 1:27 AM

Interesting questions on the kaizen time.

As Chris said, this is an organization specific issue. For example, if you have a couple people working on the side and selling their goods at an online storefront like Etsy or something like that, you can plan whatever time you like. If you have 1000 people working side by side doing their part to assemble pickup trucks, I'm not sure how you would go about building that level of time in to your schedule, let alone manage that many people kaizen-ing at one time.

The other issue that I see with building kaizen time in to your takt, in general, is that you only allow time to improve when sales are down. Let's say that instead of a standard takt of 140 seconds, you decide to back out kaizen time and build around a cycle of 130 seconds. If orders go up, how would that get handled? Do you just run at 130 seconds for the full work week or keep the planned down time and run overtime? Sometimes the most interesting kaizen opportunities happen when you are pushed to maximum throughput and have the rest of your defenses forced away.

Poster: Joe W | Post Date: May 1, 2008 5:01 PM

Joe, Chris and Panu,
Time to do Kaizen and make improvements is best done on overtime as it is not a repeating activity that should be built into the level schedule. The kaizen is an out of cycle activity that often does not fit into a weekly schedule, so it is planned by those doing the kaizen and scheduled outside the normal working system (overtime). If I try to build it into my planned working time I will introduce a lot of waste since it is not a repeating activity that can be paced to takt.
Les

Poster: Lester | Post Date: May 6, 2008 6:53 AM

Thanks everybody who posted comments, questions and answers to the takt time competition. The competition is now closed, but please continue the conversation.

We are happy to say that you are all winners. Originally the idea was that only couple of people would win a copy of The Illustrated Toyota Production System, but I have changed my mind and we'll send each person a copy.

Please send us a note to info@gemba.com with "illustrated TPS" in the subject line if you are a winner, and let us know where you would like the book shipped.

Poster: Jon Miller | Post Date: May 9, 2008 10:32 PM

Hey!

I need some help, and I got to this forum in my search! I have to do a takt vs cycle time graph, but I have never done one. Is this done using any specific graph type or just a column graph? But how would I then have the takt time line? Can anyone help me with this? Thank you!

Poster: Cristina | Post Date: April 28, 2009 6:39 AM

Hi Cristina,

Sorry for the late reply. We missed some comments. For a cycle time vs. takt time graph a simple column chart with the cycle times on the X axis and the takt time on the Y axis is OK. Draw a horizontal line for the takt time, typically shown in red, and cycle times as bars for each process vertically across the bottom. Here is a takt vs. cycle time template you can use to make it easier.

Poster: Jon Miller | Post Date: June 20, 2009 10:42 AM
Comments are moderated to filter spam and inappropriate content. There may be a delay before your comment is published.