First, some context, so the benchmark means something. Start the twelve questions
One of us reads every result and writes back with what we'd fix first. Your details are used only to reply: privacy notice .
1 of 12 · Documentation Is there a written description of how this process runs today?No: it lives in people's heads Yes, and a new hire could follow it Yes, and the rules in it are what the automation runs on Yes, including where the agent stops and asks a person Yes, and it's revised from what the system reports each month
2 of 12 · Documentation Do two people running it produce the same result?Usually not Yes, by design Yes, because rules execute the routine steps Yes, and the exceptions have named owners Yes, and variance is measured and reviewed
3 of 12 · Documentation When the process changes, where does the change get recorded?It doesn't, or in a chat thread In the document everyone follows In the document and in the automation's rules In the document, the rules, and the agent's thresholds In a maintained system of record the whole loop reads from
4 of 12 · Data and systems Where does the data this process needs come from?Rekeyed by hand between systems Exported and imported manually, but consistently Connected; the routine hops are automatic Connected, and messy inputs are handled with a person on the threshold Connected, validated, and the system reports its own data quality
5 of 12 · Data and systems Can the systems involved talk to each other without a person in between?No Not yet, but the steps are consistent enough to connect Yes, for the routine steps Yes, including variable inputs within set bounds Yes, end to end, with the loop monitored
6 of 12 · Data and systems If an input is wrong or missing, what happens?It's caught later, if at all Someone notices and follows the written fix The system halts loudly and a person fixes it The system flags it to a named owner with its reasoning The system flags it, and the pattern is reviewed to prevent the next one
7 of 12 · Measurement Do you know the cycle time of this process: start to finished?No Roughly, from memory or a spreadsheet Yes, from the automation's log Yes, including the 90th percentile Yes, live, and it drives what gets changed next
8 of 12 · Measurement Do you know the rework or error rate?No Anecdotally Yes, sampled from the automation's output Yes, measured continuously Yes, and it's trending down because of a review loop
9 of 12 · Measurement Could you show a before-and-after if something changed next month?No baseline exists We could reconstruct one Yes, we have a baseline from the system Yes, per decision type Yes, and we already do this monthly
10 of 12 · Ownership and exceptions Who owns this process end to end?Nobody, or it's contested One named person One named person, and the automation has an owner too A named owner with a deputy for escalations A named owner who reviews the system's patterns, not its instances
11 of 12 · Ownership and exceptions When something unusual comes up, how is it handled?Escalates to whoever knows Escalates to the manager every time, by the written rule The automation stops and a person handles it Named owners for each exception type, with thresholds Owners, thresholds, and a monthly review of the pattern
12 of 12 · Ownership and exceptions If an automated step made a wrong change, could you undo it?Nothing is automated yet With effort, manually Yes, there's a known rollback Yes, and the rollback is tested Yes, tested, and rollbacks are counted as a metric
See my result
Seeing your result sends your answers, these details and a PDF of your report to Cantera Works, so we can email you the written version. See our privacy notice .
Answer every question to see your level.
Your process is at
Below L1 Undocumented
L1 Predictable
L2 Automated
L3 Agentic
L4 Looped
Where the gaps are
Your lowest dimension sets your level. No averaging up: that's the rule we hold to on engagements too.
What to do first
Your quick-start plan four steps, nothing to buy.
Today
Write down how the process really runs, even where it differs from the SOP. One page.
Send the assessment result to whoever owns the process.
This week
Have someone who doesn't run the process follow your page. Note where they get stuck.
Book the thirty minutes and bring the page.
Next 30 days
Remove one manual hop between systems using capability you already license.
Take a cycle-time and rework baseline, same method you'll use afterwards.
Next 90 days
Pilot one workflow against criteria you set first. Go or no-go with the reasoning shown.
Two of your people in the build from day one.
Where each fix sits
Quick wins Do first, often no software
Big bets Pilot, then Build
Fill-ins When there's slack
Leave alone We'll say so
Effort →
Impact →
Quick wins first. The Survey puts every process in your company on this grid, with the effort and the hours attached.
Download your PDF report
Where you sit in your industry, and who's pulling away.
Benchmark: our working estimate from two decades of process work in these industries, refined as assessment results accumulate. It's a field estimate, not a survey.
The cost of staying here for another year.
The returned-hours range is what we typically see for one level of movement; the Survey replaces it with a measured number for your process.
What moving one level changed for companies that were where you are. Illustrative composites: details changed, patterns real
Getting to L1
A regional distributor: order entry The problem
Six people keyed orders from emails and PDFs into the ERP, each their own way. Orders took three days to enter, and the only person who knew the exceptions was on leave the week we arrived. If your process lives in people's heads, this is you.
Where we started
We didn't start with software. We sat with the six for two weeks, watched real orders, and wrote down what actually happened, including eleven rules nobody had ever documented.
What changed
One intake, one written process, one owner per exception type. A new hire followed it in her first week.
3 days → same day order-to-entry
1 in 3 errors gone before any software
0 new tools bought
L1 to L2
A mid-market manufacturer: purchase-order matching The problem
Matching was documented and consistent, but every step was done by hand across three systems. Two skilled buyers spent most of their week on it. If you have a good SOP and no automation behind it, this is you.
Where we started
We started with what they already owned. The ERP could match on rules; nobody had switched it on. We built the rules on their system, in their accounts, with their two buyers in the room.
What changed
Clean orders match themselves. Anything outside the rules halts loudly and a person handles it. Our access was revoked at week eight.
~70% fewer manual matching hours
2 buyers moved onto supplier negotiation
Week 8 our access revoked; their team runs it
L2 to L3
A healthcare group: claims and denial exceptions The problem
Automation handled the clean claims. Every denial with a real reason still went to one billing lead, and her backlog was four weeks long. If your automation works until the first exception, this is you.
Where we started
The agent started read-only: it read every denial and summarized why they were happening. Once it agreed with the lead most of the time, it began recommending a response for her to approve.
What changed
Routine denials resolve the day they arrive. The hard ones reach her with the reasoning attached. Every decision sits in a readable trail the auditor accepted.
8% → 41% same-day exception resolution
7 weeks to clear a four-week backlog
100% of decisions with a readable trail
L3 to L4
A pharmaceutical services firm: deviation reporting The problem
Agents were already drafting deviation reports within bounds. Nobody was looking at the pattern across them, so the same root cause was written up eleven times. If your system acts but never learns, this is you.
Where we started
We closed the loop. The system now surfaces recurring patterns monthly. The quality owner reviews them, adjusts thresholds, and her team logs the change.
What changed
Repeat deviations fell by half in two quarters. The monthly review takes three hours instead of two days, and it's run by their people.
-50% repeat deviations in two quarters
2 days → 3 hrs monthly review time
Their team makes every threshold change
Your next step if you want to move.
Want us to look at this with you?
Thirty minutes, no commitment. We keep a handful of slots open each week for people who've scored a process. Send us this result and we'll write back with two times that work. We'll say plainly whether a Survey is worth your money, including when the answer is no.
Received. One of us will read your result and write back within a business day with two thirty-minute slots. From info@canteraworks.com: worth adding to your contacts.
