We automate
The chore
Highly irritating, with little human value. Nobody would miss it if AI took it over.
Examples
Three completed canvases
Three organizations, three processes, three modes. Each example starts from the team’s friction log and walks through the nine blocks.
Fictional but realistic cases, built from situations observed in workshops.
How to read the examples
Each example is classified with two markers from the method. First the sorting matrix, which places a task according to two questions: does it weigh on the team? Does it call for real human value? Then the mode, which says what role AI is given.
The four quadrants of the matrix
HighPain level →Low
We augment
The noble burden
Heavy, but it takes professional judgment. AI prepares, the human decides.
We wait
The noise
Barely irritating and barely meaningful. It can wait.
We protect
The sanctuary
The heart of the work: relationships, empathy, judgment. Hands off.
LowHuman value →High
The three modes
- Automate
- AI runs on its own, within a clearly bounded scope.
- Augment
- AI prepares a first draft, the human decides and signs off.
- Assist
- The human does the work, AI helps as needed.
01 · Manufacturing SME
The machine shop and its quotes
The context
A 45-employee metal fabrication company serves the mining and forestry sectors. About 25 requests for quotes a month. A single estimator, 30 years of experience, three years from retirement. Each quote takes him 4 to 6 hours, and clients sometimes wait a week.
What the friction log revealed
The estimator spends more time digging than estimating: rereading 40-page specifications to find the requirements, searching the network for a similar past quote, copying lines into the Excel template.
- Process selected
- Preparing quotes
- Quadrant
- The noble burden
- Mode
- Augment
- Why this classification
- The noble burden, because a quote is heavy to prepare, but pricing and feasibility rest on the estimator’s judgment. Augment mode: AI digs and pre-fills, the estimator decides.
The canvas
AI Canvas
AI CanvasBeneficiaries
The estimator first. Clients (faster response). Management (less dependence on a single person).
AI value
Tasks
Pains
digging through specs, finding past quotes, re-entering data.
Gains
responding faster, winning more contracts, passing on the estimator’s know-how.
Meaning
judging technical feasibility, setting the price, talking to the client. We don’t touch that.
Point of integration
The existing network folder and Excel template. The estimator drops in the specs and gets a pre-filled template back. No new software to learn.
Human-AI relationship
AI extracts the requirements, suggests three comparable past quotes and pre-fills the template. The estimator validates every line and sets the price. No quote goes out without his signature.
Target process
Before: full read-through, manual search, re-entry, calculation, writing. After: review of a requirements summary, comparison with the suggested cases, calculation, validation.
Data and tools
Eight years of quotes in Excel, about half of them well organized. Plans and specs in PDF. Enterprise AI assistant, data not used for training.
Allies and dependencies
The estimator (champion or blocker, he decides). The IT provider. A junior estimator to be hired, who learns with the tool.
Costs and risks
Enterprise AI licences, about 40 hours of setup and data cleanup. Main risk: a pricing error that eats the margin. To check: confidentiality agreements with some mining clients.
Measurable gains
From 5 hours to 2.5 hours per quote. Response time under 3 business days. Time reinvested: training the junior and visiting clients.
- Desirable
- Feasible
- Viable
- Meaning
Filter verdict
Feasible, on one condition: review client confidentiality agreements before the test. The real gain isn’t just time. It’s the estimator’s know-how becoming searchable before he leaves.
Proposed test
Six weeks, ten quotes prepared with the tool, time tracked and prices compared with the usual method.
Open in the online canvas02 · Municipality
The small town and its citizen requests
The context
A small town of 4,000 residents. At the office: the general manager, an administrative assistant and a part-time inspector. About 60 requests a week come in by phone, email and Facebook.
What the friction log revealed
Most calls are about the same questions: collection schedule, tax deadlines, renovation permits, when the recycling depot opens. The real requests (a blocked culvert, a dangerous pothole) get lost in the inbox, and nobody knows which ones have been handled.
- Process selected
- Answering information questions and sorting requests
- Quadrant
- The chore
- Mode
- Automate (public information only)
- Why this classification
- The chore, because repeating the collection schedule dozens of times a week takes no judgment. Automate mode, but only for public information: decisions stay with staff.
The canvas
AI Canvas
Beneficiaries
Citizens (answers in the evening and on weekends). The assistant. The public works crew (clear, tracked requests).
AI value
Tasks
Pains
repeating the same answers, scattered requests, no follow-up.
Gains
instant answers, a request log.
Meaning
welcoming people at the counter, especially seniors, and handling disputes between neighbours. Those moments stay human.
Point of integration
The town website and Facebook page, where citizens already ask their questions. The phone line stays open: nobody has to talk to a robot.
Human-AI relationship
The assistant answers information questions on its own, with a link to the official source. It creates a ticket for each request and hands it to a human. It makes no decisions and has no access to any citizen file.
Target process
Before: a call, an answer from memory, a note on a scrap of paper or nothing at all. After: question answered online, or request logged, assigned and tracked until it’s closed.
Data and tools
Bylaws, collection calendar, fees, permit procedures. All public, but scattered across PDFs that are sometimes out of date. A cleanup is needed before plugging anything in.
Allies and dependencies
Town council (resolution and budget). The regional county, which could share the tool with neighbouring towns. The website provider.
Costs and risks
Subscription to a conversational assistant, monthly content updates. Risks: a wrong answer about a bylaw, personal information typed in by citizens. Framework: public-sector privacy rules.
Measurable gains
Repetitive calls per week, turnaround time for road requests, percentage of requests closed. Time reinvested: the grant applications that keep getting pushed back for lack of time.
- Desirable
- Feasible
- Viable
- Meaning
Filter verdict
Feasible if the scope stays narrow: public information and request creation, never tax accounts or personal files. The longest job isn’t the AI. It’s updating the town’s documents.
Proposed test
Eight weeks, limited to two topics (collections and permits), with call counts before and during.
Open in the online canvas03 · Community organization
The employment nonprofit and its paperwork
The context
A nonprofit that helps young people into jobs and back to school. Twelve caseworkers, several programs funded by public funders, each with its own reporting requirements. Files contain highly sensitive information.
What the friction log revealed
Caseworkers spend a big part of their week writing: follow-up notes after every meeting, then quarterly results compiled into three different templates for three funders. The most frequent sigh in the log: “the report, again.”
- Process selected
- Follow-up notes and reporting
- Quadrant
- The noble burden, bordering on the sanctuary
- Mode
- Assist
- Why this classification
- The noble burden, because the writing weighs heavily. But right next to the sanctuary: the meeting with the young person and the bond of trust cannot be delegated. Assist mode: the counsellor writes, AI helps structure and compile.
The canvas
AI Canvas
Beneficiaries
Caseworkers. Young people, indirectly (more time in meetings). The coordinator (reports delivered on time).
AI value
Tasks
Pains
writing late at night, rewording the same results for three funders.
Gains
more consistent notes, reports compiled in hours instead of days.
Meaning
the meeting itself, the bond of trust, clinical judgment. No AI in the room during the meeting.
Point of integration
The case management software already in use and the funders’ templates.
Human-AI relationship
Notes: the caseworker writes a summary in their own words, with no names or identifying details; AI structures it, they reread and sign. Reports: AI compiles aggregated, anonymous data and drafts a first version, the coordinator validates.
Target process
Before: notes written by hand, manual compilation, three reports written from scratch. After: a structured summary to review, indicators extracted, three reports adapted from the same base.
Data and tools
Indicators already entered in the case management software. Funders’ templates. Open question: can the software produce an anonymized export?
Allies and dependencies
The person in charge of protecting personal information. The software vendor. The funders, who might accept a common format. The board of directors.
Costs and risks
Major risk: sensitive information sent to a service hosted outside the country. A privacy impact assessment is required before any use on the notes. Secondary risk: notes that lose the caseworker’s voice.
Measurable gains
Writing hours per week per caseworker, turnaround time for quarterly reports. Time reinvested: more meetings, more follow-up in the community.
- Desirable
- Feasible
- Viable
- Meaning
Filter verdict
Green light for reporting, which relies on aggregated data. Yellow light for follow-up notes: not before the privacy assessment and a solution hosted in Canada or run locally. The exercise helps us better understand the risks and choose the steps to follow.
Proposed test
A full quarter of reporting produced with the tool, in parallel with the usual method. Follow-up notes wait until the assessment is done.
Open in the online canvasWhat the three examples show together
| Manufacturing SME | Municipality | Community organization | |
|---|---|---|---|
| Quadrant | The noble burden | The chore | The noble burden, near the sanctuary |
| Mode | Augment | Automate | Assist |
| What stays human | Pricing and the client relationship | Welcoming people and handling disputes | The meeting and clinical judgment |
| Main obstacle | Confidentiality of client specs | Outdated town documents | Sensitive personal information |
| Verdict | Green light, with a condition | Green light, narrow scope | Green for reports, yellow for notes |
In none of the three cases is technology the main obstacle. It’s the state of the data, trust or the legal framework. Not the tool. The work around the tool.
Want to do the exercise with your team?
KODRA Conseil runs AI Canvas workshops in person and online, across Quebec and beyond.