AI deployment simulation: roll out an AI assistant, safely
Five months, one AI assistant, forty agents and a board that wants hours saved with no headlines. Try your rollout plan here before you try it on real customers.
- Six use cases, three modes, five guardrails: you design the rollout
- A wrong answer reaches a customer and a data request lands
- Scored on return, risk, trust and judgement about AI
- A new scenario every play, a debrief every time
Not sure yet? Try the free 3-round Price War first: no sign-up, about five minutes.
What is a good simulation for learning how to roll out AI at work?
Deploy the Agent is a 20-minute Bodhih Arena simulation in which you roll out an AI assistant in an insurance service team. You choose use cases, guardrails, human review, training and measures, then handle a wrong answer and a data request. It scores return, risk, trust and judgement, and explains every result.
What is Deploy the Agent?
Deploy the Agent is a single-player business simulation from Bodhih Arena about putting generative AI to work responsibly. You run customer service at an invented insurer, Ostrella Home Insurance, and have five months to trial Sol, an AI assistant. Each decision changes hours saved, errors that reach customers, staff and customer mood, and a risk tally. The hidden error rate and adoption are only revealed in the debrief, which teaches human in the loop design.
Deploy the Agent at a glance
| Format | Micro-sim, single player |
|---|---|
| Length | About 20 minutes per play |
| Players | One person; every play is a new seeded scenario |
| Devices | Any modern browser on a laptop, tablet or phone |
| Scoring | Scored on the server out of 100, with sub-scores and a debrief |
| Credential | Open Badges 3.0 micro-sim credential at a score of 80 or more, with a public verification page |
| Competencies | Commercial Acumen (SPN.4.5), Information Security Awareness (SPN.6.5), Leading Change (SPN.3.5), AI Tool Fluency (SPN.6.4) |
| LMS | LTI 1.3 launch from Canvas, Moodle, Blackboard or D2L, with grade passback; usable as a Bodhih pathway stage |
| Price for one person | $12 |
| Price per seat | $12 per seat, volume pricing by seat count |
Every company is piloting AI. Few pilots are designed to show the truth.
The vendor demo is not your contact centre
A tool that looks flawless on stage meets messy policies, upset customers and half-trained staff. The error rate you were sold is rarely the one you get.
The dashboard says it works, the complaints say otherwise
Handling time and containment go up and to the right, while wrong answers quietly reach customers. Teams report what is easy to measure, not what matters.
Nobody has rehearsed the bad day
When the assistant tells a customer something false, or someone asks for every chat you hold on them, most teams improvise. That is where trust and regulators are lost.
Here is the situation you walk into
You run customer service at Ostrella Home Insurance: 40 agents and about 30,000 calls, chats and emails a month. The board has approved a five-month trial of Sol, an AI assistant built on a large language model. They want hours saved, no headlines and a clear recommendation at the end.
You decide what Sol does and whether it talks to customers directly or only drafts for agents. You choose the guardrails, the pilot size, the training and what you report. Then, during the pilot, a reply written by Sol tells a customer that flood damage from a burst river is covered. It is not, and the customer has already booked builders.
A month later a customer asks for every piece of data you hold on them, including their chats with Sol, and then asks you to delete it. They also have an open claim. What you built in month one decides how well you can answer.
5 decisions, one debrief
- STEP 1
Design the rollout
Pick two to four use cases (policy questions, claim status, draft replies, email routing, small-claim approvals, complaints), choose assist-only, mixed or direct answers, and buy any of five guardrails.
- STEP 2
Run the pilot
Choose one team of eight or all forty agents, set training hours and the share of direct answers a supervisor reviews, and pick up to three measures to report to leadership.
- STEP 3
Handle the incident
A wrong answer has reached a customer. Pause and fix the cause, patch it quietly, blame the AI or treat it as a one-off. Then decide whether to hold, expand or roll back.
- STEP 4
Answer the data request
Route it to the privacy process, let Sol handle it, delete the logs or wait. Your earlier guardrails, such as an audit log, change how this plays out.
- STEP 5
Advise the board
Scale what works, extend Sol to approving small claims automatically, hold for another quarter or switch it off. Then read the debrief and the hidden numbers.
Who Deploy the Agent is for
- Managers about to introduce AI. Contact centre, operations and back-office leads who will own an AI pilot and its consequences, and want to make the first mistakes somewhere safe.
- AI, data and digital leaders. People who set AI policy and need business colleagues to understand why guardrails, review rates and the right metrics are not optional.
- Risk, compliance and privacy teams. A shared, concrete case for discussing incidents, legal holds and data requests with the operational managers you advise.
- Business and management students. A short, scored exercise on technology adoption that turns a lecture on AI governance into decisions with visible trade-offs.
Who should skip it
It is not a technical course on building or prompting models; if you need to build AI systems, look at the AI/ML certification track instead.
What you can do after Deploy the Agent
Twenty minutes later you will have run a whole AI rollout, seen what it really cost, and learned why the same tool succeeds in one team and fails in another.
Choose AI use cases by the cost of an error
Start where errors are cheap and value is real, such as drafts and routing, and keep people in charge of money and complaints.
Design guardrails that change the error maths
Explain what grounding answers in approved articles, hand-off on low confidence and human review each catch, and what each costs in saved time.
Pick measures that show real quality
Report audited accuracy and customer outcomes alongside handling time, so leadership sees the whole picture, not only containment.
Run an AI incident properly
Pause the use case, fix the source, make the customer whole and tell your team, rather than patching quietly and hoping.
Handle a privacy request involving AI chats
Route it to the privacy process, find every record and respect a legal hold before anything is deleted.
Learning goals
- Choose AI use cases by value and by the cost of an error
- Design guardrails, human review and measures that show real quality
- Handle an AI incident and a privacy request through proper process
Skills practised: Decision making under uncertainty, Strategic thinking about AI adoption, Return on investment reasoning, Risk assessment, Change management and staff adoption, Choosing meaningful metrics, Incident response, Data privacy awareness.
Competencies scored
Sub-scores map to the open Bodhih Competency Spine, so results sit beside your other learning records.
What the debrief shows
- What happened: hours saved, the full cost including fixing errors, how many wrong answers reached customers and your final risk tally.
- Why: how adoption, guardrails and review rates changed the result, explained in plain words.
- What strong players do, and the reveal: Sol’s true error rate against the vendor’s claim.
- Your four sub-scores, each mapped to a Bodhih Competency Spine skill, and the key concept: human in the loop.
Human in the loop
Designing an AI system so that people review, approve or can override its outputs where errors are costly. The right amount of human checking depends on the cost of an error and on measured accuracy, and should change as evidence comes in.
Use Deploy the Agent with a team
Seats are $12 each. Buy up to 25 seats online and assign them from your company admin; above that, or for a facilitated programme or an institution licence, request a quote and we activate your seats after payment.
Before an AI pilot kicks off
Ask the pilot team to play it, then compare their choices with your actual plan. Gaps in measures and guardrails show up fast.
Workshop opener on responsible AI
Play in pairs for 20 minutes, then debate the incident and data-request choices. The debrief gives the facilitator the talking points.
Pre-work for an AI course
Set it before an AI for managers programme so people arrive with a shared story and their own numbers to discuss.
A stage in a learning pathway
Add it as a Bodhih Arena stage in a pathway; the stage closes itself when the learner finishes, and the score is recorded.
Volume pricing
| Seats | Per seat | How to buy |
|---|---|---|
| 1 | $12 | Online |
| 2 to 10 | $10.80 | Online |
| 11 to 50 | $9.60 | Online up to 25, then quote |
| 51+ | $8.40 | Quote |
Deploy the Agent compared with a case study or slide-based course
| Compared on | A case study or slide-based course | Deploy the Agent |
|---|---|---|
| What you do | Read about another company’s AI rollout | Run your own rollout and live with the results |
| Feedback | A discussion, if there is time | Instant scored results and a debrief on every choice |
| Hidden information | The ending is already known | The AI’s real error rate and adoption are hidden until the end |
| Practice | One reading, one answer | A new seeded scenario each play, so you can try another approach |
| Evidence | A completion tick | Sub-scores mapped to Competency Spine skills and a credential at 80 or more |
| Time | An hour or more of reading | About 20 minutes |
What to pair with Deploy the Agent
AI for Managers course
Takes the judgement you practised here into your own team’s week, with real work checked by a person.
See the course →Pro KitThe AI Use Policy and Governance Kit
Turns the guardrails, risk tiers and review rules from the simulation into a written policy your organisation can adopt.
See the kit →Pro KitThe Data Privacy Compliance Starter
Gives you the response letters and process behind the data request you handled in month three.
See the kit →Online courseLLM Engineering certification
For the engineers who will build the assistant: evaluation, guardrails, audit trails and an incident runbook.
See the course →AssessAll assessmentEveryday AI Judgment at Work (AssessAll)
Measures how well each person on the team judges when to trust, check or avoid generative AI.
Take the assessment →Related simulations
How much does Deploy the Agent cost for one person?
One person can buy Deploy the Agent online for $12. That gives your own Bodhih account access to this simulation. Every simulation is sold separately, so you pay only for the one you want. Try the free 3-round Price War first if you want to see how Bodhih Arena works.
How does pricing work for a team or company?
Companies buy seats at $12 per seat, with volume tiers by seat count. You can buy up to 25 seats online. For more seats, a facilitated workshop or an institution licence, request a quote: we invoice, you pay offline, and your seats are activated.
How long does it take?
About 20 minutes for one play, across five steps: design, pilot, incident, data request and board review. Reading the debrief carefully adds a few minutes, and it is worth it.
What exactly do I get?
Access to the simulation, a server-scored result with four sub-scores, a written debrief (what happened, why, what strong players do and one key concept), a reveal of the hidden numbers, and a credential if you score 80 or more.
Can I play it again?
Yes. Every play is a new seeded scenario: the AI’s true error rate, how fast staff adopt it and how much customers are harmed change each time, so a second play tests your judgement rather than your memory.
Do I get a certificate or credential?
A score of 80 or more earns a Bodhih Arena micro-sim credential issued as an Open Badge 3.0, with a public verification page and an Add to LinkedIn button.
Is this a technical AI course?
No. It needs no coding or data science. It is about the management decisions around an AI rollout: scope, guardrails, review, training, measures, incidents and privacy. It suits anyone who will sponsor, run or govern an AI pilot.
Can it run inside our LMS, such as Moodle or Canvas?
Yes. Bodhih Arena is an LTI 1.3 tool, so teachers and L&D teams can launch it from Canvas, Moodle, Blackboard or D2L Brightspace and send the score back to the gradebook.
What devices does it work on?
It runs in a web browser with nothing to install. A laptop, desktop or tablet gives the clearest view of the charts and dashboard.
How is it different from reading a case study about AI adoption?
A case study tells you what another company did. Here you make the calls yourself, the risks are hidden until they happen, and you get a score and a debrief that explain why your rollout worked or did not.
What happens to my data, and can I get a refund?
Your play is stored against your Bodhih account so we can score it and show your results. The privacy notice at bodhih.org/privacy explains what is held and why. Refund terms are set out at bodhih.org/refunds.
Make your first AI rollout mistakes here, not with customers
Twenty minutes, one board, one assistant and the two bad days most pilots never rehearse. Play it before your next AI steering meeting, then bring the debrief with you.
Want a feel for Arena first? Play the free 3-round Price War.
- Play it as often as you like for 12 months: every play is a new scenario
- A debrief after each play: what happened, why, and what strong players do
- A verifiable credential when you score 80 or more
- Scored on the server, mapped to the Bodhih Competency Spine
Not sure yet? Try the free 3-round Price War first, no sign-up.
More from the Bodhih family
Measure before and after, plan the next step and bring in a facilitator when you want one.