How to create a value stream map
A current-state value stream map is drawn in eight steps, and seven of them happen on the floor rather than at a desk. What follows is the sequence, what to collect at each step, and the mistake each step is most often got wrong by.
The sequence below is the one set out in Rother & Shook's Learning to See, with the arithmetic at each step linked to the figure it produces. Do it with a pencil, a clipboard and a stopwatch the first time; the walking is the part that teaches you something, and it cannot be delegated to a data extract.
-
Choose one product family
A value stream map covers one family of products that share most of their processing steps and equipment, from the customer end backwards. Not the whole plant. List your products against the process steps they pass through, and a family is the group whose rows look alike.
Pick the family that matters commercially — highest volume, worst delivery performance, biggest customer complaint — because you will be asking people for three days of attention on the strength of it.
Most common mistake: Mapping "the plant" rather than a family. The resulting map averages products with different routings and describes something that does not exist, so no finding from it can be acted on.
-
Establish demand and the working calendar
Get the customer's actual requirement over a stated period, and the calendar that serves it: shifts per day, hours per shift, breaks, planned maintenance, working days per week.
These two produce takt time, which every later finding is measured against. Weekly and monthly quantities are spread over operating days rather than calendar days — 1,500 a week from a five-day plant is 300 a day, not 214, and the difference is an entire capacity argument.
Most common mistake: Reducing available time by uptime before computing takt. Takt is a statement about the customer, not about how well your equipment runs; folding in capability produces a rhythm the plant is already comfortable with and calls it demand.
-
Walk the flow backwards, from shipping to receiving
Start at the shipping dock and walk upstream. This is deliberate: starting at the customer end keeps you attached to what the customer actually receives, and it stops the walk turning into a tour of the plant's favourite machines.
Draw the process boxes as you go — one per flow step, where a flow step is a place material stops and something is done to it. Two machines a part moves between without stopping are one box.
Most common mistake: Drawing the flow from the routing file instead of walking it. Routings describe the intended path; the map is supposed to capture the actual one, and the gap between them is often the finding.
-
Fill the data box at each process
Under each process box: cycle time, changeover time, uptime, number of operators, shift pattern, batch size, and scrap or rework rate. Time the cycle yourself, several times, on the floor.
Cycle time feeds processing time; scrap and rework feed first pass yield; uptime, performance and quality together feed OEE — and only together. A figure computed from uptime alone is not OEE, whatever the spreadsheet calls it.
Most common mistake: Recording standard times rather than observed times. The standard is what the process is supposed to take; if it matched what it takes, there would be nothing to map.
-
Count the inventory between processes
Every place material waits gets a triangle with a count in it: raw material, work in process between each pair of operations, finished goods, and anything sitting in transit. Count it. Do not query it.
Inventory converts to waiting days by dividing by daily demand, and those days are almost always where the lead time actually is.
Most common mistake: Trusting the inventory system. The count on the floor and the count in the system differ routinely, and the map is a record of the floor.
-
Draw the information flow
Above the material flow, draw what tells each process to produce: the forecast from the customer, the schedule from production control, the daily list, the kanban, the phone call. Mark each material arrow as push or pull — a push arrow means the upstream station produces to a schedule regardless of what the downstream one took.
This half of the map is what explains the other half. A pile of inventory is rarely a storage decision; it is usually a scheduling one.
Most common mistake: Leaving it out because it is "the same for every station". If it were, the inventory would be evenly distributed — and it never is.
-
Build the timeline
Along the bottom, a sawtooth: waiting time on the upper step (from each inventory triangle), processing time on the lower. Sum each separately.
The lower sum is processing time; the two together are production lead time; the ratio is process cycle efficiency. This is the number the map exists to produce, and in most first maps it is under 5%.
Most common mistake: Mixing units. Cycle times are in seconds and waiting is in days; adding them without converting produces a lead time that is wrong by orders of magnitude and looks plausible on a slide.
-
Find the constraint
Compare every process's cycle time against takt. Any process whose cycle time exceeds takt cannot meet demand, and that is the finding the whole exercise exists to produce.
Then look at line balance: how the work is distributed against takt across the stations. A stream can have no single process over takt and still fail, because the work is bunched.
Most common mistake: Stopping at the map. A current state with no future state and no implementation plan is the artefact this method most often produces and the least useful one — see when not to map.
After the current state
The current-state map is half the method. The future state asks a fixed set of questions against it — what is the takt, will you build to a supermarket or to ship, where can continuous flow be introduced, at which single point will you schedule, and how will you level the mix — and the answers become a design with an implementation plan attached. Creating Continuous Flow is the reference that treats that half in detail.
Common questions
Where do I start if I have never drawn one?
Choose one product family, borrow a stopwatch, and walk the flow from shipping backwards with a pencil. The first map is not supposed to be pretty or complete — it is supposed to be observed. Redrawing it neatly afterwards takes an hour; the walk cannot be shortened.
How many processes should be on the map?
Typically five to fifteen boxes for a door-to-door plant map. If you have forty, you are mapping operations rather than flow steps — combine machines a part moves between without stopping. If you have three, the scope is probably too narrow to show where the lead time goes.
Do I need cycle time or takt time first?
Takt first, and it comes from demand and the calendar rather than from the floor — so you can compute it before the walk. It is the yardstick every cycle time you then collect gets compared against.
What if a process runs several products?
Record the changeover time and the batch size, and carry them into EPEI — how often the process can cycle through its whole mix. A shared process is not a reason to leave a step off the map; it is a reason the map has an EPEI on it.
How often should a map be redrawn?
When the plant changes materially — a new product family, a routing change, a demand shift large enough to move takt — or when the future state has been implemented and has become the new current state. Redrawing on a calendar rather than on a change produces maps nobody reads.
Sources
- Mike Rother & John Shook, Learning to See: Value Stream Mapping to Create Value and Eliminate Muda, Lean Enterprise Institute, 1999.
- Mike Rother & Rick Harris, Creating Continuous Flow, Lean Enterprise Institute, 2001 — the future-state half in detail.
Everything above can be done on paper, and for a first map it probably should be — the walking and the counting are the exercise, not the drawing. What paper does not do is the arithmetic, and the arithmetic is where a map quietly stops agreeing with the report made from it.
VSM Works is a value stream mapping tool built around that problem: one calculation engine produces the on-screen metrics, the A3 print, the PDF, the workbook and the image, so a figure in a presentation cannot disagree with the map it came from. It runs in the browser, needs no account to try, and works in English and Turkish. Its methodology is published in full, and its arithmetic is checked against the totals Rother & Shook published.