AI Innovation Workshop: A Three-Day Sprint That Ends in Ranked Pilots

Will an AI innovation workshop give your organisation anything more than a wall of sticky notes? Our straight answer: yes, if it is built to end in a ranked list of two or three pilot candidates. Each one needs a named sponsor, the data it would use, a reason it might fail and a date for a go/no-go decision. If the agenda cannot promise that by the last afternoon, we recommend not running the workshop yet.

What should an AI innovation workshop produce?

Judge the workshop by what leaves the room, not by the energy in it. Fix four outputs before anyone books a venue.

• A ranked pilot list. Two or three ideas, scored against the same criteria, in order.
• A one-page card for each pilot. Problem, workflow, data, sponsor, success and stop criteria, and a test date.
• A parked list. Good ideas that fail today on data, risk or ownership. The reason is written down so they can return later.
• A stop list. Ideas the group agreed not to pursue, so they do not come back as "new" next quarter.

The parked and stop lists matter as much as the winners. Without them, rejected ideas can resurface in side conversations, and the ranking loses its weight.

Who to invite, and what to prepare beforehand

We recommend choosing the problems before the workshop, not on day one. Ask three or four department heads to name one workflow each that costs their team time or causes errors. Each brings a real, redacted example of it. The trade-off is less open exploration. In return, the group spends its days on work that exists, and the scoring has evidence to use.

We suggest a room of 8 to 12 people, with four kinds of participant:

• The decision owner who will approve or stop pilots, present at least for the final afternoon.
• People who do the work every day, not only their managers.
• Someone who knows the data: where it lives, who is responsible for it and how clean it is.
• Someone from risk, legal or information security, who can say early what data an idea may touch.

Prepare a safe data pack too. Prototypes in the workshop should use synthetic or redacted examples, and only AI tools your organisation has approved. A workshop is a poor place to discover that a team pasted customer records into a personal chatbot account.

One, three or five days: choosing the format

The five-day format follows the design sprint published by GV. Its days run in a fixed order: map the problem, sketch competing solutions, decide, build a realistic prototype, then test it with customers. That suits one big question. For a first AI innovation workshop we recommend three days instead. It is long enough to test ideas with the people who would use them, and short enough to keep senior people in the room.

What about an internal AI challenge?

An internal AI challenge, where staff teams propose and build ideas over a few weeks, reaches more people than a workshop. Give the challenge its own judging rubric, published on day one, and use the same pilot card for the winners.

How to run a three-day AI innovation workshop, step by step

The agenda borrows the shape of the Double Diamond from the UK Design Council: first understand the problem, then develop and test answers. Its first phase asks teams to "understand, rather than simply assume, what the problem is". Design thinking, in plain terms, means starting from the people who do the work and testing ideas with them early.

1. Day 1, morning: walk through each workflow. The people who do the work show the real example, step by step. Mark where time goes and where errors start.
2. Day 1, afternoon: frame the problem. Write each problem as one sentence: who, which task, what goes wrong, and how you would know it improved.
3. Day 2, morning: generate options. Sketch at least three answers for each problem, including one that uses no AI at all. A simpler process fix sometimes wins.
4. Day 2, afternoon: build quick prototypes. A written prompt and a few redacted documents in an approved assistant can be enough to test the idea.
5. Day 3, morning: test with users. Put each prototype in front of two or three people who do the task. Record where it fails, especially on Arabic input and mixed Arabic and English text.
6. Day 3, afternoon: score, rank and decide. The group scores each idea against the table below. The decision owner confirms the ranked list and signs each pilot card.

How to rank workshop ideas into pilots

Dot-voting is quick, but it tends to reward the most exciting demo. We recommend scoring every idea from 1 to 5 on five criteria, with the evidence for each score written next to it.

Two rules sit above the totals. An idea that scores 1 on data readiness or on cost of a wrong answer goes to the parked list, whatever its total. And no more than three ideas become pilots, because each needs a team, a sponsor and regular reviews.

This follows the logic of the MAP function in the NIST AI Risk Management Framework (AI RMF 1.0, 2023), a voluntary framework. NIST says that mapping an AI system's context should give enough knowledge for an initial go/no-go decision. It also asks organisations to document the potential costs of AI errors. A workshop ranking is a first, light version of that decision. The pilot then tests it on evidence.

A pilot card to copy

• Pilot name and workflow: the task, the team and where the work happens today.
• Problem: one sentence, taken from day one.
• Users: who will use it during the pilot, and how many people.
• Data: sources, classification and the approval still needed.
• Success criteria: the number you expect to change, how you will measure it and the starting point.
• Stop criteria: the result that ends the pilot early.
• Sponsor and pilot lead: two names, not departments.
• Dates: the review dates and the go/no-go meeting.

For the success and stop lines, our guide to an AI proof of concept charter with kill criteria has wording you can reuse.

Two illustrative workshops in Jordan and the Gulf

Illustrative case 1. A mid-sized microfinance company in Amman brings three workflows: loan application intake, branch emails and complaint replies in Arabic. In testing, the intake prototype misreads scanned application forms. It scores 1 on data readiness and goes to the parked list until the scanning problem is checked. Branch email sorting ranks first: the data sits in one mailbox, and a person reviews every draft reply.

Illustrative case 2. A logistics company in Dubai runs a one-day session for its executive team, then a three-day workshop for customer service. The one-day session produces fourteen ideas but no tests. The three-day workshop narrows them to two pilots: shipment status replies in Arabic and English, and a daily summary of delayed orders for supervisors. A third idea, automatic refund approval, scores 1 on cost of a wrong answer, and the group agrees to put it on the stop list rather than park it, because its errors would reach customers' money directly.

In both cases the winning ideas are not the most ambitious ones. They are the ones with data in reach, a person checking the output and a sponsor willing to own the result.

Mistakes that turn a workshop into theatre

• Starting with the tools. A vendor demo on day one narrows thinking to what that product does.
• No decision owner in the final session. The ranking becomes a recommendation that waits for another meeting.
• Too many pilots. Five half-staffed pilots tend to teach less than two properly staffed ones.
• No follow-up date. Book the first pilot review before people leave the room.

How to start planning your workshop

If you are deciding whether to run an AI innovation workshop, start with the problems, not the agenda. List three workflows, name the person who would decide on pilots, and check which data could be used safely. Our Innovation Sprints & Design Thinking service in the Academy Lab runs design-thinking workshops, innovation sprints and internal AI challenges that surface pilot candidates.

If you first need a scored view of readiness and use cases across the organisation, see our AI strategy page and our guide to running an honest AI readiness assessment. Feasibility tests for the ranked pilots can follow in our Prototype & PoC Studio. Not sure where to begin? The free AI Exposure Scan is a 45-minute conversation that ends in a written next step.

Related lab: Academy Lab ↗