AI Playbooks
Your best process, run the same way every time
Write a job down once in plain language. Fusion runs it step by step, checks its own work, and asks before anything changes.
The Process Challenge
The Challenge
Every energy and commodity data team has jobs it repeats again and again:
- Setting up monitoring and quality checks for a new dataset
- Checking this morning's forward curves before traders rely on them
- Working out why a delivery was late, and telling the right people
- Triaging failed processes and deciding what to rerun
- Building the month-end summary of the markets you follow
Each job has a right way to do it, but that way usually lives in one person's head or in a wiki page nobody opens. When that person is busy or away, the job is done differently, done late, or not done at all.
General purpose AI does not solve this on its own. Ask it twice and you get two different approaches, with no record of what it did and no point where a person signs off.
Our Solution
Fusion AI Playbooks turn your team's processes into repeatable, governed AI workflows on the OpenDataDSL platform.
A playbook is a markdown document that describes one job: what the person running it fills in, the steps in order, the checks each step must pass, where someone approves, and what the run hands back.
Fusion turns the playbook into a simple form. Someone fills it in and presses Run, and Fusion works through the steps on your platform data, exactly as written, with a full record of every step.
AI Playbooks: Your Process, Run by AI, Governed by You
Key Benefits
Capture how your best people work, and let the whole team run it with the same quality every time.
The Same Quality, Whoever Runs It
Your process, not the AI's guess.
The playbook sets the order of the steps and the tools each step may use. Fusion does the work inside each step, so a new analyst gets the same result as your most experienced one, and the process survives holidays and staff changes.
AI That Checks Its Own Work
Every step is verified before the run moves on.
Checks such as "every curve was found", "the script validates" or "every number in the commentary matches the figures" run after each step. If a check fails, Fusion tries again and is told why, or the run stops so a person can look.
People Stay in Control
Approvals before anything changes.
A run waits for approval before it saves, creates or sends anything. Approvers can be named people or user groups, they are emailed when a run needs them, and your tenant decides whether people may approve their own runs.
Runs While You Are Away
Background runs and schedules.
Run a playbook in the background and close the page, or give it a schedule so it runs every weekday morning. Long runs are carried out one step at a time, and a run that stops can be resumed from the step where it stopped.
A Complete Audit Trail
See exactly what happened, and what it cost.
Every run records each step's result, every check, every approval and the tokens used. Each playbook can be shown as a flow diagram coloured as the run progresses, and the Playbook Runs report shows success rates and costs across all your playbooks.
Built in Minutes, Not Weeks
Describe the job and Fusion drafts it.
Describe a job in a sentence and Fusion writes the playbook, checking it as it goes, or start from the OpenDataDSL library. Fusion can also suggest the values for a task, and you can start a playbook straight from a Fusion chat.
How AI Playbooks Work
Write It Once
A playbook is plain markdown with a few marked blocks that Fusion understands:
Inputs: the fields on the form, such as the curves, the dataset, the month or a schedule, with validation so a run starts with good values.
Steps: the work, in order. Each step has instructions in plain language, the assistant best suited to it, and the tools it may use.
Checks and transitions: what each step's result must satisfy, and rules that choose the next step, such as "if nothing is wrong, end here".
Approvals and outputs: where a person signs off, and what the run hands back, such as a report, a script or a summary.
Every save is versioned, so you can see what changed and every task records the version it ran.
Run It Safely
When a playbook runs, Fusion follows it one step at a time:
Focused Steps: Fusion works only on the current step, with only that step's tools plus a few read-only ones. A step that reads curves cannot save a script.
Verified Results: each step records its result in a structured way, and its checks decide whether the run carries on, tries again or stops.
Human Sign-off: the run pauses at each approval, emails the approvers, and continues only when one of them approves.
Within Budget: each run has a token cap, and runs respect your tenant's Fusion AI budget and limit.
Ready-Made Playbooks
Start from the OpenDataDSL library. Run any playbook as it is, or copy it to make it your own; your copy tells you when the library version improves.
Curve Quality Check
The Challenge: A bad price or a stale curve reaches traders and risk before anyone notices.
The Playbook
- Reads each of your curves after it builds
- Flags big day-on-day moves, missing tenors, stale curves, bad prices and shape breaks
- Uses each curve's own calendar and holidays
- Ends quietly when everything is fine
- Raises one alert per curve with problems
Result: Curve problems found minutes after the build, on a schedule, with no one watching.
Dataset Onboarding
The Challenge: Setting up completeness and quality checks for a new dataset takes specialist knowledge and is easy to get wrong.
The Playbook
- Reads the dataset, its fields and tenors
- Turns quality rules written in plain words into dataset checks
- Writes one setup script you can rerun
- Waits for approval, then runs the setup
- Verifies that the monitoring is in place
Result: A dataset monitored properly from day one, set up the same way every time.
Late Dataset Investigation
The Challenge: When a delivery is late, someone has to dig through processes, logs and history to find out why.
The Playbook
- Checks the delivery and its last 30 days
- Tells a one-off apart from a habit
- Reads the loading processes and their logs
- Checks whether the provider itself is late
- Writes a short root-cause summary with evidence, and emails it after approval
Result: A clear answer in minutes, with the evidence to back it up.
Process Failure Triage
The Challenge: One broken script or late source can fail many processes, and working out which to rerun takes time.
The Playbook
- Finds failures not already fixed by a rerun
- Groups them by root cause, with log evidence
- Proposes a fix for each cause
- Lists the processes a rerun can fix
- Reruns them once someone approves
Result: Fewer failures chased one by one, and no reruns that were bound to fail again.
Scheduled Curve Report
The Challenge: A daily report on a set of curves needs a script, a template and a schedule, and usually a developer.
The Playbook
- Checks the curves you choose
- Designs the report data
- Writes and validates the ODSL script
- Writes and validates the report template
- Waits for approval, then saves both and creates the scheduled report
Result: A daily curve report with day-on-day changes, spreads and calendar spreads, without writing code.
Month-End Market Summary
The Challenge: The month-end summary means gathering figures from many markets and writing commentary by hand.
The Playbook
- Works out averages, ranges and moves for each market over the month
- Writes commentary for your audience
- Checks every number against the figures
- Emails the summary after approval
Result: A consistent, accurate month-end summary on the first business day of every month.
Integration with Platform Capabilities
AI Playbooks bring the whole platform together
AI Assistants
Each step runs with the assistant best suited to it, from Curve to Code.
Smart Curves
Playbooks check, report on and build forward curves.
Data Management
Playbooks set up dataset monitoring and investigate deliveries.
ODSL Language
Playbooks write, validate and save ODSL scripts and report templates.
AI Agents
Agents handle open-ended monitoring; playbooks handle jobs with a defined process.
Custom Tools
Steps can use your own tools to reach your internal systems.
From Tribal Knowledge to Team Capability
AI Playbooks turn the way your best people work into something the whole team can run. The process is written down once, improved over time, and followed every time, by Fusion or by a person.
Your team spends less time repeating routine jobs and more time on the decisions that need them, with every run checked, approved where it matters, and recorded.
See a Playbook Run on Your Data
Book a demo and we will run a playbook against your own curves or datasets, and help you sketch one for a job your team does every week.
Get Started TodayCustom Tools
How about a demo?
Our team is here to find the right solution for you, contact us to see this in action.
Fill out your details below and somebody will be in contact with you very shortly.