Leave a field blank only if you did not ask. If you asked and it does not apply, write n/a, because a blank and a deliberate n/a mean different things when someone reads this later.
Fields marked * are mandatory
A field is marked * if it is either impossible to recapture later or
blocks a decision. Do not close the engagement with one outstanding.
Everything unmarked is valuable but survivable.
Minutes of human effort per item, not elapsed time.
Task
Minutes
Who does it
Their loaded hourly cost
Source
* Read and triage one email
* Draft a routine reply
Produce a quote
Create a job in the system
Reschedule a job
Chase missing paperwork
Raise an invoice
Chase an unpaid invoice
Find and send a document
Re-key data between systems
Produce a report
Loaded hourly cost means salary plus on-costs, not the charge-out rate. Ask for a range if they will not give a figure. Without it, savings can only be expressed in hours rather than dollars.
Value at stake is the amount riding on that item, such as the average quote value for a quote never followed up. Cost when it happens is what it costs them once it goes wrong, such as the hours to redo it plus the parts.
This section is usually where the real value sits, and it is the section clients have never measured. Estimates are fine. Getting them to say a number out loud is the point.
This is the number every savings claim rests on. It must be theirs, and it must be recorded here. When it is challenged in month six, this is the answer.
Add a block per task if more than one multiplier is agreed.
Mark each email E for prompt example or V for eval case. Roughly eight and twelve. They cannot be the same emails: an email used as a prompt example will always be classified correctly, because the answer is in the prompt, so scoring against it measures nothing.
Fill "why" wherever the category was contested. That reasoning is what makes a disputed case reusable a year later, instead of someone re-litigating the same edge from scratch.
#
Category assigned
E / V
Why, if contested
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Template categories that did not fit
Categories to add for this client
Categories they would never automate
Keep this table. It is the seed for both the prompt examples and the eval set.
Choose the E eight for coverage, meaning the adjacent pairs, at least one
unclear, and at least one with two categories. Examples of the easy cases
teach nothing.
Do this by watching, not asking. "Show me where the last certificate you issued is, right now."
* Where it actually was
* In the job record, or filed elsewhere?
job record / folder / email / other
* If a folder: network or local?
network / local / unsure
* Path or location seen
* Does the job record have a document type field?
yes / no / unsure
* Actual file names, copied exactly
* Filed by hand, or automatically?
hand / automatic
* Typical attachment count on a finished job
Also lodged with a regulator or third party?
yes / no
Who would notice if the wrong document went out?
The four gates
* In a reachable system with an API
yes / no / unsure
* Identifiable among the other attachments
yes / no / unsure
* Reliably linked to the job
yes / no / unsure
* Mail connector can attach files
yes / no
* Retrieval feasible?
yes / no / needs a spike
* If no, the fallback and its price
The first two gates fail most often. A certificate saved by hand into a
network folder with whatever name the scanner produced is not retrievable,
however good the automation is.
document_request is the highest-volume automatable category and is nearly
worthless without the attachment. If retrieval is impossible, the fallback is a
reply saying it is ready and asking the office to send it. Price it as the
small saving it is.