What should AI take on, who checks the result, and how will you measure its value? These three templates put the key decisions in one place. Free, with no signup.
Copy a template into your own document and replace the square brackets. Keep open questions visible instead of treating assumptions as commitments. You can also select the text directly or print the template.
AI project brief
For a first pilot with a defined task, a measurable result and someone responsible for the decision. Agree on the brief with the people doing the work before requesting proposals.
Example
A service team tests draft replies for 30 approved historical enquiries. The team lead reviews every draft. The aim is shorter handling time with at least the same subject-matter quality, measured including review and corrections.
AI project brief
Project / version: [name / date / version]
Owner / decision-maker: [name / role]
1. Task and scope
Today: [trigger → steps → finished result]
Problem and people affected: [specific bottleneck / team]
The pilot will: [take on one bounded task]
Out of scope: [cases, data and actions]
2. Baseline and success
Volume: [cases per week] across [working weeks per year]
Current time: [minutes per case, including review and corrections]
Evidence / period: [sample, log or clearly labelled estimate]
Test scope: [number and selection of representative cases]
Success criteria: [time, subject-matter quality, acceptable errors]
Acceptance: [who checks what, by when, using which evidence?]
3. Data and human review
Approved sources / maintenance: [systems, documents / owner]
Access and limits: [who may use which data for what purpose?]
Human checkpoint: [who checks the result before it is used or sent?]
Unclear or incorrect cases: [escalation owner / manual fallback]
4. Effort and commitments
Currency / cost basis: [EUR, SEK, … / including or excluding tax]
One-off costs: [setup, integration, training / amount]
Recurring costs: [licence price/user/month × user count, usage/month, operations, support]
Internal work: [hours for preparation, review and upkeep × hourly cost]
Potential capacity released: [hours/year / assumption and measurement]
Assess net value only after recurring and one-off costs; time released is not a commitment to lower payroll costs.
Our contribution: [data, people, testing time / deadline]
Supplier commitment: [deliverables, limits, support / deadline]
5. Decision and exit
Start / evaluation: [date / date]
Continue, change or stop: [criterion / decision-maker]
At the end: [data export and format, access removal, deletion / owner and evidence]
Open questions: [question / evidence needed / owner / deadline]
Next agreed step: [action / person / deadline]
AI workshop agenda
A 90-minute internal working session to define a possible AI pilot. Adjust the timing and participants to your team. This is an example agenda, not the duration of a booked AI Værk workshop.
Example
Example: A service lead, a case handler and someone from IT examine recurring status enquiries. They bring three approved example cases and finish by recording which source is missing, who can approve it and when they will decide on a limited test.
AI workshop agenda · 90-minute example
Topic / meeting: [one process / date and time]
Facilitator / notes: [name / name]
Participants: [people doing the work, process owner, IT; other roles as needed]
Decision to reach: [which bounded pilot should we investigate further?]
Preparation
[ ] Bring the current workflow and approximate case volume.
[ ] Record handling time and typical errors; label estimates.
[ ] Prepare two or three examples approved for use.
[ ] Agree on permitted tools and data; do not enter confidential data without checking permission.
00–10 min · Goal and limits
Output: [bottleneck, desired improvement, explicitly excluded tasks]
10–25 min · Walk through the actual work
Output: [trigger, steps, responsibilities, current time per case]
25–40 min · Check sources and risks
Output: [required data, access rights, maintenance, unresolved prerequisites]
40–50 min · Break
50–65 min · Define a small pilot
Output: [task, test cases, human checkpoint, manual fallback]
65–80 min · Record value, costs and responsibility
Output: [measures, internal hours, one-off/recurring costs with units and currency]
80–90 min · Decide and assign actions
Output: [investigate / change / stop; next decision and date]
Decision record
Pilot in one sentence: [task / users / result]
Measurement and acceptance: [baseline / target / evidence / owner]
Commitments: [who delivers what by when; required team contribution]
At the end of the pilot: [data export, access removal, deletion / owner]
Open questions: [question / evidence / owner / deadline]
Next step: [action / person / deadline]
AI supplier and proposal checklist
Give every supplier the same project brief. Complete one checklist per proposal and compare the answers side by side. A documented commitment is more useful than a polished demo.
Example
Example: One proposal lists a monthly platform fee; another also includes setup and internal project work. Compare both over twelve months using the same user count and usage. If data export is unspecified, mark it as open and ask for concrete evidence.
AI supplier comparison · One sheet per proposal
Supplier / proposal / date: [name / version / date]
Shared project brief: [version and test task]
Comparison basis: [user count, usage/month, term, currency, tax included/excluded]
Must-haves before comparing: [which requirements are essential?]
For each item, record: met / partly met / open / not met.
Alongside it: specific commitment + evidence/proposal reference + open question.
[ ] Task fit: Does the proposal solve our defined test task?
[ ] Result quality: Which representative cases were tested, and how?
[ ] Sources: Are citations visible; how are missing or conflicting data handled?
[ ] Human review: Who approves outputs and actions before execution?
[ ] System access: Which integrations and base licences are included or needed separately?
[ ] Data access: Who gets access; how are existing permissions preserved?
[ ] Data use: Storage locations, subprocessors, retention, training and deletion clarified?
[ ] Operations: Who monitors errors, updates sources and handles outages?
[ ] Service limits: Volumes, response times, support hours and excluded work specified?
[ ] Our contribution: Which data, decisions and internal hours must we supply?
[ ] Delivery: Deliverables, dates, acceptance, fixes and changes agreed?
[ ] Contract: Minimum term, renewal, cancellation and price changes clarified?
[ ] Exit: Which data, configurations and content can we export, in which formats?
[ ] Handover: Who handles access removal, deletion confirmation and a supplier switch; at what cost?
Costs on the same comparison basis
One-off: [setup + integration + training = amount]
Recurring: [licence price/user/month × user count + monthly usage, operations and support = amount/month]
Internal: [hours within the comparison period for setup and review/upkeep × hourly cost]
Period: [e.g. 12 months; allow for ramp-up and contract commitments]
Total for the period: [one-off + recurring over the period + internal costs]
Excluded / usage risk: [item / trigger / price per unit]
Decision
Unmet must-haves: [requirement / evidence needed by when]
Open risks and questions: [item / owner / deadline]
Assessment: [suitable / further evidence needed / unsuitable; reason]
Next step and approval: [action / decision-maker / deadline]
