Lesson objective

Attach turnaround, callback and rework measures to the individual steps of a service line map, trace each callback to the step that caused it, and choose the step to redesign from that evidence.

Plain-language explanation

Why this matters now

Module 1 set a company-level baseline: how many callbacks, how much rework, how long things take. That number says whether the operation is getting better. It does not say where the problem is. For that, the measures have to sit on the map, step by step, so the company can see which step causes delay, which causes callbacks and which causes rework.

This matters because AI is usually proposed for the most visible step, not the costliest one. An owner who has watched an account manager spend twenty minutes typing a proposal wants AI to write proposals. The map may show that proposals wait four days for a site measurement, and that most callbacks come from quantities, not wording. Automating the visible step leaves the real problem untouched and adds a new source of error.

Professional guidance (voluntary framework): NIST's AI RMF 1.0 (January 2023) orders its functions as MAP, then MEASURE, then MANAGE, applied iteratively. It is voluntary and binds no landscaping company, and NIST says 1.0 is being revised. For an operation the order means: measure the steps you mapped before managing any of them.

A question to sit with

Enhancement callbacks rose last season. Is the cause in the estimate, the work order, the crew's execution or the customer's expectations? What would you need to have recorded to answer that from evidence rather than opinion?

What goes wrong

One number for the whole line. "Average proposal turnaround is six days" hides whether the time is spent waiting for a site visit, waiting for pricing or waiting for the customer.

Callbacks counted but not traced. A callback log that records only "return visit" cannot show whether the error began in sales, estimating, scheduling or the field.

Waiting time ignored. Most turnaround in a landscape office is waiting between steps, not working on them. Automating the working time of a step that mostly waits changes little.

Measuring people instead of steps. Time records and callbacks can quickly become a league table of crew leaders. That changes behaviour (people stop reporting callbacks) and it is not what the baseline is for.

Records that are not kept long enough. A company that discards last season's records cannot build a seasonal baseline. Law / rule (federal, binds FLSA-covered employers): under 29 CFR part 516, payroll records must be preserved for at least three years and time cards, and order, shipping and billing records, for at least two. Those are minimums for wage and hour purposes, and state law or a litigation hold can require longer, but a compliant company should have the last two seasons' time cards and billing records to hand.

The practical answer

Put three measures on every step. This is an AI Coalition Network recommendation, not a legal requirement.

Measure Definition Where to find it
Work time Time someone actively spends on the step Time records in aggregate; a short timed sample
Wait time Time between the previous step finishing and this one starting Dates and times in the CRM, email, work orders
Error share Callbacks or rework traced to this step, as a share of all traced callbacks or rework Coded callback log

Code every callback to its origin step. Add one field to the callback or rework record: the step on the map where the error began. The person who closes the callback fills it in, and a manager checks a sample each month so the coding stays consistent.

Work the enhancement example through. Suppose a fictional company recorded 450 enhancement jobs last season and 18 own-error callbacks, which is 18 ÷ 450 = 4 per 100 jobs. Traced to origin step, the callbacks fall like this:

Origin step Callbacks Share of callbacks
Estimate quantities 9 9 ÷ 18 = 50%
Work order scope 5 5 ÷ 18 ≈ 28%
Crew execution 3 3 ÷ 18 ≈ 17%
Scheduling 1 1 ÷ 18 ≈ 6%
Total 18 100% (rounded shares sum to 101%)

The estimate step is where half the callbacks begin. A proposal-wording tool would not touch it. Any AI proposal for this line should start with the estimate step, and it must be tested against the estimator's own takeoffs before it is trusted.

Separate wait from work. For the same line, suppose the median time from request to proposal sent is six working days, of which the site measurement wait is four, pricing is one and writing is one. Cutting writing time in half saves half a day. Scheduling site measurements sooner could save more, with no AI at all.

Choose the step to redesign from the evidence. Rank steps by their share of callbacks, rework and wait time, and by what a failure there would cost. The step at the top is the candidate for redesign, whether or not AI is the answer. Write down why you chose it.

Mark the steps that contain qualified decisions. A step where the certified applicator chooses a product and rate, or where a licensed irrigator (in a state that licenses the work) changes a system, can rank high for error cost even if it causes few callbacks. A low callback count there is not permission to automate it. Any redesign of such a step keeps the qualified person's decision exactly where Lesson 3 placed it, and the qualified person reviews the design before it is tested.

Keep the baseline and the map together. The baseline is only meaningful against the map it measured. If the map changes, record the date and re-baseline the changed steps.

Remember this

  • A company-level baseline says whether things improve; a step-level baseline says where.
  • Record work time, wait time and error share for each step.
  • Code every callback to the step where the error began, and check the coding monthly.
  • Most turnaround is waiting; automating work time on a waiting step changes little.
  • Measure steps, not people, or the data will stop being recorded.

Landscaping example

Thornhedge Grounds Care is a fictional company whose owner wants an AI tool to write enhancement proposals, because proposals "take forever". Branch manager Oksana Brandvold first builds a step-level baseline from last season's records. From dated CRM entries she finds the median request-to-proposal time was six working days: four waiting for a site measurement, one for pricing, one for writing. Writing itself took about 25 minutes of work time per proposal. She then reads the callback log. Last season had 450 enhancement jobs and 18 own-error callbacks, 4 per 100 jobs, but the log only said "return visit". With two crew leaders and the estimator, she traces each callback to its origin: 9 began in estimated quantities (short on mulch, plant counts wrong from an old site plan), 5 in the work order scope, 3 in crew execution, and 1 in scheduling. Half the callbacks began at the estimate step. Oksana presents three findings to the owner. First, AI proposal writing would save part of one working day out of six, while the four-day wait for site measurements is the larger delay. Second, half the callbacks begin in quantities, which a writing tool would not touch. Third, the callback log now has an origin-step field, and the estimator will review a sample of the coding each month. The owner agrees to test an AI takeoff-checking step on the estimate first, run beside the estimator's own takeoffs for six weeks, and to hire a part-time measurer for the spring. Oksana writes the baseline figures, their sources and the date into the map file so that the test can be compared against them.

Responsible-use warning

Do not choose the step to automate because it is the one you can see someone working on. Most turnaround in a landscape office is waiting, and most callbacks begin at a step other than the one that looks slow. Do not measure individuals: if crew leaders think callback coding will be used against them, callbacks will go unrecorded. Keep records for at least as long as the regulations require, and longer where state law or a hold applies.

Human-review requirement

The branch or operations manager owns the step-level baseline, the callback coding and the monthly check of a coding sample. Each step's owner confirms the figures for their step: the estimator for takeoffs and quantities, the account manager for scope, the crew leader for execution. Where a step contains a qualified decision, such as a pesticide application or irrigation work in a state that licenses it, the qualified person reviews any proposed redesign of that step before it is tested.

Practical exercise

Use a fictional company or anonymised material only. No real customer, property, worker or record.

  1. Take your Lesson 1 map and add three columns: work time, wait time and error share. Write a one-sentence definition of each at the top of the column.
  2. From fictional or anonymised records, estimate work and wait time for each step. Say which record each figure comes from and the months it covers.
  3. Add an origin-step field to a fictional callback log of at least 15 entries and code each one. Have a second person code five of them and compare.
  4. Calculate the callback rate per 100 jobs and each step's share of callbacks. Check that the shares sum to 100%, allowing for rounding.
  5. Identify the step with the largest wait time and the step with the largest error share. Are they the same step? What does that tell you?
  6. Rank the steps for redesign and write two sentences explaining your top choice, whether or not AI is the answer.
  7. Write how you will prevent the data from being used to rank individual workers, and how you will tell the crews what is being measured.
  8. Record where the baseline figures are kept, who produced them and on what date, and when the changed steps will be re-baselined.

Knowledge check

  1. What three measures should sit on each step of the map?

    Answer guidance: Work time, the time someone actively spends; wait time, between the previous step and this one; and error share, the callbacks or rework traced to this step.

  2. A line had 18 callbacks in 450 jobs, 9 of them from estimate quantities. What are the rate and the estimate step's share?

    Answer guidance: The rate is 18 ÷ 450 = 4 callbacks per 100 jobs, and the estimate step's share is 9 ÷ 18 = 50% of callbacks.

  3. Why might automating proposal writing change little when proposals take six days?

    Answer guidance: If four of the six days are spent waiting for a site measurement, the writing step is a small part of turnaround. Reducing wait time at the right step saves more than speeding up work time on the visible one.

  4. What do the 29 CFR part 516 retention periods mean for a seasonal baseline?

    Answer guidance: Payroll records are kept at least three years and time cards and billing records at least two, so a compliant company should hold the last two seasons' records. These are minimums, and state law or a litigation hold may require longer.

Sources and further reading