Industries
Ten industries.The work we would test first.
Each has its own work, systems and rules. Choose yours to see what Butterfly can change there, the tasks we would test first and the dates that matter.Choose yours to see the tasks we would test first and the dates that matter.
Placing in its industry…
The first task we would testBanking and credit
ExampleYoursKYC and AML alert disposal
Screen, stage 2 of 5 in the work
The first task we would testInsurance
ExampleYoursFirst notice of loss triage and cover
Notice, stage 3 of 7 in the work
The first task we would testHealthcare administration
ExampleYoursPrior-authorisation packets
Authorise, stage 3 of 7 in the work
The first task we would testAccounting and finance operations
ExampleYoursBank and intercompany reconciliations
Reconcile, stage 3 of 6 in the work
The first task we would testLegal
ExampleYoursPlaybook contract review
Review, stage 2 of 5 in the work
The first task we would testPharma and life-sciences regulatory
ExampleYoursAdverse-event case processing
Monitor, stage 5 of 5 in the work
The first task we would testAsset and wealth management
ExampleYoursNAV reconciliation
Value, stage 3 of 5 in the work
The first task we would testTelecom and customer operations
ExampleYoursGuided fault diagnosis
Diagnose, stage 2 of 6 in the work
The first task we would testPublic-sector casework and permitting
ExampleYoursPlanning-application validation
Validate, stage 2 of 6 in the work
The first task we would testManagement consulting
ExampleYoursMarket sizing
Size, stage 2 of 4 in the work
The first task we would testInsurance
ExampleYoursDeciding a motor claim against the policy wording
Assess, stage 4 of 7 in the work
The first task we would testPharma and life-sciences regulatory
ExampleYoursChecking a regulatory document for claims the data don’t support
Submit, stage 2 of 5 in the work
The first task we would testInsurance
ExampleYoursSettling a home insurance claim after the excess and limits
Assess, stage 4 of 7 in the work
The first task we would testBanking and credit
ExampleYoursDeciding a card dispute under the card scheme’s rules
Resolve, stage 5 of 5 in the work
The first task we would testPublic-sector casework and permitting
ExampleYoursChecking a planning application for anything missing
Validate, stage 2 of 6 in the work
The first task we would testTelecom and customer operations
ExampleYoursDeciding whether a broadband fault needs an engineer’s visit
Diagnose, stage 2 of 6 in the work
The first task we would testPharma and life-sciences regulatory
ExampleYoursReviewing a batch record before the batch goes to release
Manufacturing quality, before a batch’s release
The first task we would testHealthcare administration
ExampleYoursCoding a hospital stay from its clinical notes
Code, stage 6 of 6 in the work
The first task we would testLegal
ExampleYoursTriaging an incoming NDA: sign it, amend it, or escalate it
Intake, stage 1 of 5 in the work
The first task we would testAccounting and finance operations
ExampleYoursReconciling a client’s bank and ledger accounts
Reconcile, stage 3 of 6 in the work
The first task we would testBanking and credit
ExampleYoursKYC and AML alert disposal
Screen, stage 2 of 5 in the work
The first task we would testInsurance
ExampleYoursFirst notice of loss triage and cover
Notice, stage 3 of 7 in the work
The first task we would testHealthcare administration
ExampleYoursPrior-authorisation packets
Authorise, stage 3 of 7 in the work
The first task we would testAccounting and finance operations
ExampleYoursBank and intercompany reconciliations
Reconcile, stage 3 of 6 in the work
The first task we would testLegal
ExampleYoursPlaybook contract review
Review, stage 2 of 5 in the work
The first task we would testPharma and life-sciences regulatory
ExampleYoursAdverse-event case processing
Monitor, stage 5 of 5 in the work
The first task we would testTelecom and customer operations
ExampleYoursGuided fault diagnosis
Diagnose, stage 2 of 6 in the work
The first task we would testPublic-sector casework and permitting
ExampleYoursPlanning-application validation
Validate, stage 2 of 6 in the work
The first task we would testHealthcare administration
ExampleYoursUK clinical coding
Code, stage 6 of 6 in the work
About a hundred held-back cases, each checked against
The investigator’s final decision on the same alert
- Every alert, every day
- Not licensed: stays in your cloud
About a hundred held-back cases, each checked against
The handler’s route and the claim’s final outcome
- Every claim
- Can be licensed
About a hundred held-back cases, each checked against
The payer’s decision on the same request
- Every procedure that needs approval
- Synthetic copy
About a hundred held-back cases, each checked against
The reconciliation your controller signed
- Every month-end, every entity
- Can be licensed
About a hundred held-back cases, each checked against
The senior lawyer’s mark-up
- Every contract that comes in
- Synthetic copy
About a hundred held-back cases, each checked against
The case as your safety team finalised it
- Every report received
- Synthetic copy
About a hundred held-back cases, each checked against
The NAV as signed off
- Every valuation point
- Can be licensed
About a hundred held-back cases, each checked against
Whether the fault was fixed first time
- Every fault contact
- Can be licensed
About a hundred held-back cases, each checked against
The officer’s validation
- Every application
- Can be licensed
About a hundred held-back cases, each checked against
The model the team delivered, assumption by assumption
- The first week of a project
- Synthetic copy
About a hundred held-back cases, each checked against
The handler’s decision and the claim’s final outcome
- Thousands of times a week
- Can be licensed
About a hundred held-back cases, each checked against
Your regulatory team’s review of the same document
- Dozens of times a week
- Synthetic copy
About a hundred held-back cases, each checked against
What the handler paid on the same claim
- Thousands of times a week
- Can be licensed
About a hundred held-back cases, each checked against
The analyst’s outcome on the same dispute
- Thousands of times a week
- Synthetic copy
About a hundred held-back cases, each checked against
The officer’s validation
- Hundreds of times a week
- Can be licensed
About a hundred held-back cases, each checked against
Whether the fault was fixed first time
- Thousands of times a week
- Can be licensed
About a hundred held-back cases, each checked against
Your quality team’s review of the same record
- Hundreds of times a week
- Synthetic copy
About a hundred held-back cases, each checked against
Your coding team’s codes and audit
- Thousands of times a week
- Synthetic copy
About a hundred held-back cases, each checked against
The lawyer’s decision
- Hundreds of times a week
- Synthetic copy
About a hundred held-back cases, each checked against
The reconciliation your reviewer signed
- Thousands of times a week
- Can be licensed
About a hundred held-back cases, each checked against
The investigator’s final decision on the same alert
- Every alert, every day
- Not licensed: stays in your cloud
About a hundred held-back cases, each checked against
The handler’s route and the claim’s final outcome
- Every claim
- Can be licensed
About a hundred held-back cases, each checked against
The payer’s decision on the same request
- Every procedure that needs approval
- Synthetic copy
About a hundred held-back cases, each checked against
The reconciliation your controller signed
- Every month-end, every entity
- Can be licensed
About a hundred held-back cases, each checked against
The senior lawyer’s mark-up
- Every contract that comes in
- Synthetic copy
About a hundred held-back cases, each checked against
The case as your safety team finalised it
- Every report received
- Synthetic copy
About a hundred held-back cases, each checked against
Whether the fault was fixed first time
- Every fault contact
- Can be licensed
About a hundred held-back cases, each checked against
The officer’s validation
- Every application
- Can be licensed
About a hundred held-back cases, each checked against
Your coding team’s codes and audit
- Every finished episode of care
- Synthetic copy
01 · The ten
What we would test first,industry by industry.
Choose an industry: the work from start to finish, the tasks we would test first, and what each one is checked against.
Banking and credit, task 01: KYC and AML alert disposal
01 of 10
Banking and credit
Analysts review an agent’s work on each alert, spread and credit memo, instead of starting every one from a blank page.
The work is high-volume and can be checked step by step, and UK model-risk rules already expect banks to validate the models they take from vendors, AI included.56
Nearly half of UK financial firms report a partial understanding of the AI they use, and a third of their use cases are third-party.
Source: Bank of England / FCA, 21 Nov 2024 7
The work, start to finish
A task we would test first
- 01Onboard
- 02Screen
- 03Lend
- 04Pay
- 05Resolve
The tasks we would test first
Task 01 · ScreenEvery alert, every day
KYC and AML alert disposal
- The agent gets
- An alert, the customer’s file and the transactions behind it
- It hands back
- Close or escalate, with the reasons written up for an investigator
- Checked against
- The investigator’s final decision on the same alert
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
- Can be licensed
- Synthetic copy
- Stays in
Spreading and payment repairs can be licensed. Anti-money-laundering files never leave.
Task 01 · ScreenEvery alert, every day
KYC and AML alert disposal
- The agent gets
- An alert, the customer’s file and the transactions behind it
- It hands back
- Close or escalate, with the reasons written up for an investigator
- Checked against
- The investigator’s final decision on the same alert
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
Task 02 · LendEvery application and annual review
Loan spreading and covenant extraction
- The agent gets
- A borrower’s accounts and the facility agreement
- It hands back
- The spread in your template, and each covenant with its threshold
- Checked against
- The analyst’s own spread, line by line
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · LendEvery credit decision
Credit memos
- The agent gets
- The spread, the application and your credit policy
- It hands back
- A memo ready for a credit officer to review and sign
- Checked against
- Your policy, and the memo the officer signed
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · PayEvery business day
Payment exceptions
- The agent gets
- A payment that failed a check, its message and the rule it broke
- It hands back
- Repair, return or refer, with the fix
- Checked against
- What operations chose, and whether the payment then settled
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 05 · ResolveEvery complaint
Complaint handling
- The agent gets
- The complaint, the account history and your redress rules
- It hands back
- An outcome and a final response in your tone
- Checked against
- The handler’s outcome and the response that was sent
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
KYC analysts, credit analysts, payment operations and complaints teams.
Where the work lives
- Alert and case-management platforms
- Lending platforms
- Core banking and payment hubs
- Complaints and CRM
Rules with dates
- 17 May 2024PRA model risk, SS1/23, in Research
- 17 Jan 2025DORA, in Research
- 2 Aug 2026EU AI Act transparency, in Research
- 7 Dec 2026EU T+1, first deadline, in Research
- 1 Jan 2027California automated decisions, in Research
- 11 Oct 2027EU T+1 settlement, in Research
- 2 Dec 2027EU AI Act high-risk duties, in Research
02 of 10
Insurance
Claims handlers check an agent’s triage and cover decision, and underwriters start from a risk summary instead of a raw submission.
A claim ends paid, declined or recovered, so an agent’s work can be scored against what really happened. The London market makes the UK a natural place to start.
The Lloyd’s market in London wrote £57.9bn of gross premium in 2025, up 4.2% on the year before.
Source: Lloyd’s, full-year results 2025, 19 Mar 2026 31
The work, start to finish
A task we would test first
- 01Submission
- 02Underwrite
- 03Notice
- 04Assess
- 05Reserve
- 06Recover
- 07Account
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Claims steps without health data suit lab licensing well.
Task 01 · NoticeEvery claim
First notice of loss triage and cover
- The agent gets
- The first notice, the policy wording and any photos or reports
- It hands back
- The route (fast track, adjuster or specialist) and whether the loss is covered, with the clause
- Checked against
- The handler’s route and the claim’s final outcome
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 02 · UnderwriteEvery submission
Submission to risk summary
- The agent gets
- A submission: schedules, loss history and surveys
- It hands back
- A one-page risk summary in your underwriting template
- Checked against
- The underwriter’s own summary, field by field
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · ReserveEvery open claim, at every review
Reserving
- The agent gets
- The claim file and the payments so far
- It hands back
- A case reserve and the reasoning behind it
- Checked against
- What the claim finally cost
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 04 · RecoverEvery settled claim
Subrogation
- The agent gets
- A settled claim and the facts of the loss
- It hands back
- Whether someone else should pay, and the case for recovery
- Checked against
- The recoveries actually made
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 05 · AccountEvery month, every binder
Bordereaux reconciliation
- The agent gets
- A coverholder’s monthly bordereau and the binder’s terms
- It hands back
- Every row matched, every break flagged with its reason
- Checked against
- The reconciled bordereau your team signed off
Readiness Tested on held-back cases
Lab licensing Can be licensed
Who does it today
Claims handlers, underwriting assistants, reserving teams and delegated-authority teams.
Where the work lives
- Claims systems
- Policy administration
- Email and document intake
- Bordereaux from coverholders
03 of 10
Healthcare administration
Staff check an agent’s prior-authorisation packet, appeal or codes before they go out, instead of assembling each one by hand.
The work follows payers’ written criteria and the coding rules, so every packet, appeal and code can be checked against an answer.
US physicians handle about 40 prior authorisations a week, which take them and their staff 13 hours.
Source: American Medical Association, 2026 28
The work, start to finish
A task we would test first
- 01Book
- 02Check cover
- 03Authorise
- 04Treat
- 05Code
- 06Bill
- 07Appeal
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Patient data stays in. Labs get synthetic copies, never the records.
Task 01 · AuthoriseEvery procedure that needs approval
Prior-authorisation packets
- The agent gets
- The order, the patient’s notes and the payer’s criteria
- It hands back
- A complete packet that answers each criterion, with the evidence cited
- Checked against
- The payer’s decision on the same request
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 02 · AppealEvery denial
Denials and appeals
- The agent gets
- A denial, its reason code and the record
- It hands back
- An appeal that answers the stated reason
- Checked against
- The appeals that were won
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 03 · CodeEvery visit
ICD-10 and CPT coding
- The agent gets
- The visit notes
- It hands back
- The codes, each traced to a line in the notes
- Checked against
- Your coders’ codes and your coding audits
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · Check coverEvery booking
Eligibility checks
- The agent gets
- The patient’s cover and the planned service
- It hands back
- Covered or not, and what the patient will owe
- Checked against
- The payer’s own response
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 05 · CodeEvery finished episode of care
UK clinical coding
- The agent gets
- A discharge summary
- It hands back
- ICD-10 and OPCS-4 codes for the episode
- Checked against
- Your coding team’s codes and audit
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Prior-authorisation teams, clinical coders, denials and appeals staff and patient access teams.
Where the work lives
- Electronic health records
- Billing and practice management
- Payer portals
- Coding software
04 of 10 · Where every company would start
Accounting and finance operations
Controllers review the reconciliations, accruals and matches an agent has prepared, so month-end closes on evidence, not late nights.
Every company closes its books, and every step has a right answer. It is also where a good average can hide costly errors.
On APEX-Accounting the top average score is 56.4%, yet the top rate of passing a task on all 8 tries is 2.6%, from a different model.
Source: APEX-Accounting, arXiv, 29 Jul 2026 11
The work, start to finish
A task we would test first
- 01Record
- 02Match
- 03Reconcile
- 04Accrue
- 05Close
- 06File
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Hard, multi-entity work suits labs; simplified close tasks are already solved.
Task 01 · ReconcileEvery month-end, every entity
Bank and intercompany reconciliations
- The agent gets
- Statements, the ledger and the intercompany balances
- It hands back
- Items matched, breaks explained, journals proposed
- Checked against
- The reconciliation your controller signed
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 02 · AccrueEvery month-end
Accruals and close flux analysis
- The agent gets
- This period’s and last period’s trial balances, and the open purchase orders
- It hands back
- Accruals with workings, and a reason for every material movement
- Checked against
- The posted accruals and the controller’s commentary
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · MatchEvery invoice
Three-way match in accounts payable
- The agent gets
- An invoice, its purchase order and the goods receipt
- It hands back
- Pay, hold or query, with the line that does not match
- Checked against
- The accounts payable team’s decision
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 04 · CloseEvery new kind of contract
Revenue-recognition memos
- The agent gets
- A contract and your revenue policy
- It hands back
- A memo on how and when the revenue is recognised under IFRS 15
- Checked against
- The memo your technical team approved
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 05 · FileEvery quarter
VAT returns
- The agent gets
- The period’s transactions and their tax codes
- It hands back
- The return’s boxes, every figure traced to its transactions
- Checked against
- The return as filed, and any later adjustments
Readiness Tested on held-back cases
Lab licensing Can be licensed
Who does it today
Financial controllers, accounts payable, revenue accountants and tax teams.
Where the work lives
- ERP systems
- Close and reconciliation tools
- Bank feeds
- Spreadsheets
05 of 10
Legal
Lawyers review an agent’s first pass on contracts, diligence and citations, and keep the judgement calls.
A wrong answer here becomes public, and tested tools split by task: ahead of lawyers on document questions, behind them on redlining. So each task is tested on its own.20
Court cases involving AI-hallucinated content keep mounting.
Source: Damien Charlotin database, 25 Oct 2026 22
The work, start to finish
A task we would test first
- 01Intake
- 02Review
- 03Diligence
- 04Dispute
- 05Advise
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Privileged work stays in. Labs get synthetic copies, never the files.
Task 01 · ReviewEvery contract that comes in
Playbook contract review
- The agent gets
- The other side’s contract and your playbook
- It hands back
- Each clause marked against the playbook, with your fallback position
- Checked against
- The senior lawyer’s mark-up
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 02 · DiligenceEvery transaction
Due-diligence red flags
- The agent gets
- The contracts in a data room
- It hands back
- Change of control, exclusivity and unusual liabilities, each with its clause
- Checked against
- The diligence report the team issued
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 03 · DisputeEvery disclosure exercise
Privilege review
- The agent gets
- Documents gathered for disclosure
- It hands back
- Privileged, not privileged or partly, with the reason
- Checked against
- The review team’s calls and quality checks
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
Task 04 · DisputeEvery filing
Citation checks
- The agent gets
- A brief or a skeleton argument
- It hands back
- Every authority checked: it exists, it says that, it still stands
- Checked against
- The law reports themselves
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 05 · IntakeEvery NDA that arrives
NDA triage
- The agent gets
- An incoming non-disclosure agreement
- It hands back
- Sign, amend or escalate, with the clauses that decide it
- Checked against
- The lawyer’s decision
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Associates, paralegals, knowledge lawyers and in-house legal teams.
Where the work lives
- Document management
- Review platforms
- Virtual data rooms
- Contract lifecycle tools
06 of 10
Pharma and life-sciences regulatory
Safety and regulatory teams check an agent’s case processing and submission checks, and keep their time for the medical judgement.
Safety reporting runs at scale and to fixed deadlines, and every submission is validated before a regulator reads it.
The EU’s database of suspected side effects took in about 1.77 million individual case safety reports in 2025.
Source: European Medicines Agency, 2025 annual report on EudraVigilance, 12 Mar 2026 30
The work, start to finish
A task we would test first
- 01Trial
- 02Submit
- 03Review
- 04Label
- 05Monitor
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Label and submission checks can be licensed; patient case data cannot.
Task 01 · MonitorEvery report received
Adverse-event case processing
- The agent gets
- A report of a suspected side effect, from a doctor, a patient or the literature
- It hands back
- A coded case: the product, the event in MedDRA terms, its seriousness and the narrative
- Checked against
- The case as your safety team finalised it
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 02 · SubmitEvery submission and variation
eCTD quality checks
- The agent gets
- A submission’s modules before publishing
- It hands back
- Every broken link, missing document and inconsistency, listed
- Checked against
- The health authority’s validation result
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · ReviewEvery review cycle
Health-authority query responses
- The agent gets
- A regulator’s questions and the dossier
- It hands back
- An answer to each question, citing the dossier
- Checked against
- The response your team sent, and what the regulator accepted
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · LabelEvery label change
Label comparison
- The agent gets
- Your label and the reference label, or two country versions
- It hands back
- Every difference that matters, with its section
- Checked against
- The labelling team’s own comparison
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 05 · TrialEvery monitoring visit
Protocol deviations
- The agent gets
- Site visit notes and the protocol
- It hands back
- Each deviation classed as major or minor, with its effect
- Checked against
- The deviations your clinical team logged
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Case processors, regulatory affairs, labelling teams and clinical operations.
Where the work lives
- Safety databases
- Regulatory document systems
- eCTD publishing
- Trial management
07 of 10
Asset and wealth management
Operations clear the breaks and reconciliations an agent has already worked, and advisers check suitability before advice goes out.
The operations are highly checkable, and settlement is speeding up: from 11 October 2027 EU trades settle one business day after they are made.60
The EU moves to settling securities trades one business day after trading on 11 October 2027; the first deadline is 7 December 2026.
Source: ESMA, Statement on T+1 preparations, 20 Jul 2026 60
The work, start to finish
A task we would test first
- 01Trade
- 02Settle
- 03Value
- 04Advise
- 05Report
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Operations can be licensed; client data cannot.
Task 01 · ValueEvery valuation point
NAV reconciliation
- The agent gets
- The fund’s positions and prices, and the administrator’s NAV
- It hands back
- Every difference explained before the NAV is struck
- Checked against
- The NAV as signed off
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 02 · SettleEvery trading day
Trade breaks
- The agent gets
- Your trade record and the counterparty’s or custodian’s
- It hands back
- The cause of each break and the fix
- Checked against
- How operations resolved it
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · AdviseEvery recommendation
Suitability checks
- The agent gets
- A client’s fact-find and the recommendation
- It hands back
- Suitable or not, with the reasons, against your rules
- Checked against
- Your own file reviews
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
Task 04 · ReportEvery quarter, every portfolio
Client-report commentary
- The agent gets
- A portfolio’s performance and holdings for the quarter
- It hands back
- Commentary in your house style, every figure from the data
- Checked against
- The commentary that was approved and sent
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 05 · ReportEvery investor request
Due-diligence questionnaires
- The agent gets
- An investor’s questionnaire and your approved past answers
- It hands back
- Every question answered from an approved source
- Checked against
- The answers your team approved
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Fund accountants, middle-office operations, paraplanners and client reporting teams.
Where the work lives
- Portfolio and fund accounting systems
- Custodian and counterparty feeds
- Order management
- Adviser CRM
Rules with dates
- 17 May 2024PRA model risk, SS1/23, in Research
- 17 Jan 2025DORA, in Research
- 2 Aug 2026EU AI Act transparency, in Research
- 7 Dec 2026EU T+1, first deadline, in Research
- 1 Jan 2027California automated decisions, in Research
- 11 Oct 2027EU T+1 settlement, in Research
- 2 Dec 2027EU AI Act high-risk duties, in Research
08 of 10
Telecom and customer operations
Agents take fault diagnosis and billing disputes with a person on every escalation, tested for quality before they meet a customer.
Customer operations is the largest workforce of the ten, and the lesson is public: test quality and escalation, not cost alone.
The work, start to finish
A task we would test first
- 01Contact
- 02Diagnose
- 03Fix
- 04Bill
- 05Retain
- 06Resolve
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Troubleshooting suits lab licensing well.
Task 01 · DiagnoseEvery fault contact
Guided fault diagnosis
- The agent gets
- The account, line tests and the fault in the customer’s words
- It hands back
- The next step: a fix the customer can make, a remote reset or an engineer visit
- Checked against
- Whether the fault was fixed first time
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 02 · BillEvery day
Billing disputes
- The agent gets
- The bill, the usage record and the tariff
- It hands back
- Whether the charge is right, the credit if it is not, and the reply
- Checked against
- The adviser’s outcome and any repeat contact
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · RetainEvery request to leave
Retention conversations within policy
- The agent gets
- A customer who wants to leave, their contract and your retention rules
- It hands back
- A discount or plan inside the rules, or a clean exit
- Checked against
- Your policy limits, and whether the customer stayed
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 04 · ResolveEvery complaint
Complaint classification
- The agent gets
- A complaint in the customer’s words
- It hands back
- The category, the root cause and the deadline it runs to
- Checked against
- Your complaints team’s classification
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 05 · ResolveEvery day
Vulnerable-customer handling
- The agent gets
- A conversation with signs that someone needs extra care
- It hands back
- The flag, the right support, and a person brought in
- Checked against
- Your vulnerability specialists’ review
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
Who does it today
Contact-centre advisers, technical support, billing teams and vulnerability specialists.
Where the work lives
- Contact-centre platforms
- Billing systems
- Line and network diagnostics
- Knowledge bases
09 of 10
Public-sector casework and permitting
Officers check an agent’s validation and code checks, so applicants hear sooner and officers keep their time for the decisions.
The checks are clear-cut against national and local lists, and the volumes are large. Benefit decisions are high-risk under the EU AI Act, so they are tested before any use.54
District planning authorities in England received 314,200 planning applications in the year to June 2026.
Source: MHCLG, Planning applications in England: April to June 2026, 24 Sep 2026 32
The work, start to finish
A task we would test first
- 01Apply
- 02Validate
- 03Check
- 04Inspect
- 05Decide
- 06Renew
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Permitting checks can be licensed as rewrites; benefit decisions stay in.
Task 01 · ValidateEvery application
Planning-application validation
- The agent gets
- An application with its plans and documents
- It hands back
- Valid, or the exact items missing against the national and local lists
- Checked against
- The officer’s validation
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 02 · CheckEvery building-regulations application
Building-code checks
- The agent gets
- Drawings and a specification
- It hands back
- Each requirement of the building regulations met, missed or unclear, with its reference
- Checked against
- The surveyor’s plan check
Readiness Tested on held-back cases
Lab licensing Can be licensed
Task 03 · InspectEvery inspection
Building-control reviews
- The agent gets
- Inspection notes and photos from site
- It hands back
- What is complete, and what needs another visit
- Checked against
- The surveyor’s sign-off
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · DecideEvery claim
Benefit casework
- The agent gets
- A claim and its evidence
- It hands back
- A recommendation for an assessor, with the rule behind each step
- Checked against
- The assessor’s decision and any appeal
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
Task 05 · RenewEvery renewal
Licence renewals
- The agent gets
- A renewal and the licence’s conditions
- It hands back
- Renew, renew with conditions, or refer
- Checked against
- The licensing team’s decision
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Validation officers, building-control surveyors, benefits assessors and licensing teams.
Where the work lives
- Planning and building-control systems
- The national planning portal
- Casework and document management
- Maps and GIS
10 of 10
Management consulting
Consultants start from an agent’s market model, cost analysis or bid, and spend their time with the client.
The work repeats from project to project, and consulting firms reach many of the other nine: as much a partner as a client.
UK management consultancies earned about £21.8bn in fees in 2025.
Source: Oxford Economics, MCA Annual Industry Report 2026, 10 Jul 2026 33
The work, start to finish
A task we would test first
- 01Win
- 02Size
- 03Diagnose
- 04Recommend
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Client material is confidential, so it stays in.
Task 01 · SizeThe first week of a project
Market sizing
- The agent gets
- A question, a market and the data you subscribe to
- It hands back
- A sized market, every assumption and source shown
- Checked against
- The model the team delivered, assumption by assumption
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 02 · DiagnoseEvery cost programme
Cost-out from a data room
- The agent gets
- A client’s cost base in a data room
- It hands back
- The savings levers, each sized and sourced
- Checked against
- The team’s findings, and what the client went on to save
Readiness Tested on held-back cases
Lab licensing Not licensed: stays in your cloud
Task 03 · DiagnoseEvery deal
Commercial due diligence
- The agent gets
- A target’s market, customers and plan
- It hands back
- The diligence questions answered, with the evidence
- Checked against
- The report the team issued
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · WinEvery bid
RFP responses
- The agent gets
- A request for proposal and your approved past answers
- It hands back
- A complete response from approved material
- Checked against
- The response the bid team sent, and whether it won
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Analysts, associates, bid teams and knowledge managers.
Where the work lives
- Spreadsheets and slide decks
- Virtual data rooms
- Market-data subscriptions
- Proposal libraries
03 of 10
Healthcare administration
Staff check an agent’s prior-authorisation packet, appeal or codes before they go out, instead of assembling each one by hand.
The work follows payers’ written criteria and the coding rules, so every packet, appeal and code can be checked against an answer.
US physicians handle about 40 prior authorisations a week, which take them and their staff 13 hours.
Source: American Medical Association, 2026 28
The work, start to finish
A task we would test first
- 01Book
- 02Check cover
- 03Authorise
- 04Treat
- 05Code
- 06Bill
- 07Appeal
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Patient data stays in. Labs get synthetic copies, never the records.
Task 01 · AuthoriseEvery procedure that needs approval
Prior-authorisation packets
- The agent gets
- The order, the patient’s notes and the payer’s criteria
- It hands back
- A complete packet that answers each criterion, with the evidence cited
- Checked against
- The payer’s decision on the same request
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 02 · AppealEvery denial
Denials and appeals
- The agent gets
- A denial, its reason code and the record
- It hands back
- An appeal that answers the stated reason
- Checked against
- The appeals that were won
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 03 · CodeEvery visit
ICD-10 and CPT coding
- The agent gets
- The visit notes
- It hands back
- The codes, each traced to a line in the notes
- Checked against
- Your coders’ codes and your coding audits
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · Check coverEvery booking
Eligibility checks
- The agent gets
- The patient’s cover and the planned service
- It hands back
- Covered or not, and what the patient will owe
- Checked against
- The payer’s own response
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 05 · CodeEvery finished episode of care
UK clinical coding
- The agent gets
- A discharge summary
- It hands back
- ICD-10 and OPCS-4 codes for the episode
- Checked against
- Your coding team’s codes and audit
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Prior-authorisation teams, clinical coders, denials and appeals staff and patient access teams.
Where the work lives
- Electronic health records
- Billing and practice management
- Payer portals
- Coding software
03 of 10
Healthcare administration
Coders and booking teams check the codes, referral routes and pathway corrections an agent has prepared, instead of starting each one by hand.
The work follows the national coding standards and the referral-to-treatment rules, so every code, route and pathway can be checked against an answer.
US physicians handle about 40 prior authorisations a week, which take them and their staff 13 hours.
Source: American Medical Association, 2026 28
The work, start to finish
A task we would test first
- 01Refer
- 02Triage
- 03Wait
- 04Treat
- 05Discharge
- 06Code
The tasks we would test first
- Can be licensed
- Synthetic copy
- Stays in
Patient data stays in. Labs get synthetic copies, never the records.
Task 01 · CodeEvery finished episode of care
UK clinical coding
- The agent gets
- A discharge summary
- It hands back
- ICD-10 and OPCS-4 codes for the episode
- Checked against
- Your coding team’s codes and audit
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 02 · TriageEvery referral
Referral triage
- The agent gets
- A referral and the clinic’s referral criteria
- It hands back
- The right clinic and urgency, with the criterion that decides it
- Checked against
- The triage clinician’s decision on the same referral
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 03 · WaitEvery open pathway, at every validation
Waiting-list validation
- The agent gets
- A patient’s pathway record and the referral-to-treatment rules
- It hands back
- The pathway confirmed or corrected, with the rule behind each change
- Checked against
- Your validation team’s own corrections
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 04 · TreatEvery clinic appointment
Outpatient letters
- The agent gets
- The consultant’s notes of a clinic appointment
- It hands back
- The clinic letter to the GP, in your template
- Checked against
- The letter the consultant signed
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Task 05 · DischargeEvery discharge
Discharge summaries
- The agent gets
- The notes of a hospital stay
- It hands back
- A summary for the GP, every line traced to the notes
- Checked against
- The summary the clinician signed
Readiness Tested on held-back cases
Lab licensing Synthetic copy
Who does it today
Clinical coders, booking staff and triage clinicians, waiting-list validation teams and medical secretaries.
Where the work lives
- The patient administration system
- Electronic patient records
- Referral and booking records
- Coding software
02 · Readiness and licensing
Two questionsfor every task.
Readiness asks whether an AI can be trusted with the work. Licensing asks whether AI labs can learn from it. Every task gets the first; some get the second.
1Readiness
Can an AI be trusted with it?
Agents will be tested on about a hundred of your own held-back cases, in your own cloud, against the bar your task owner sets. The tests and the results stay yours.
Inside your cloud
- Your work
- Tested on held‑back cases
- A score you can trust
Every task, in all ten industries
2Lab licensing
Can an AI lab learn from it?
When the work has a clear right answer and is allowed to leave, a privacy-safe rewrite of the task could be licensed to labs, with your approval. The records never leave your cloud.
What leaves: the rewrite
- Your work
- Rewritten, privacy‑safe
- Lab 1 and Lab 2
$1,000–2,000 a licensed taskIllustrative
The ten, by what a lab could license
- License wellClear right answers, and the work can leave as a rewrite
- Parts can be licensedSome tasks can be licensed; others stay in
- Stay inLabs can have synthetic copies, never the records
Banking and credit: 2 of 5 tasks can be licensed
Banking and credit5 of 5 tested for readiness · 2 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| KYC and AML alert disposal | Test it | Not licensed: stays in your cloud |
| Loan spreading and covenant extraction | Test it | Can be licensed |
| Credit memos | Test it | Synthetic copy |
| Payment exceptions | Test it | Can be licensed |
| Complaint handling | Test it | Synthetic copy |
Spreading and payment repairs can be licensed. Anti-money-laundering files never leave.
Insurance5 of 5 tested for readiness · 5 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| First notice of loss triage and cover | Test it | Can be licensed |
| Submission to risk summary | Test it | Can be licensed |
| Reserving | Test it | Can be licensed |
| Subrogation | Test it | Can be licensed |
| Bordereaux reconciliation | Test it | Can be licensed |
Claims steps without health data suit lab licensing well.
Healthcare administration5 of 5 tested for readiness · 0 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Prior-authorisation packets | Test it | Synthetic copy |
| Denials and appeals | Test it | Synthetic copy |
| ICD-10 and CPT coding | Test it | Synthetic copy |
| Eligibility checks | Test it | Synthetic copy |
| UK clinical coding | Test it | Synthetic copy |
Patient data stays in. Labs get synthetic copies, never the records.
Accounting and finance operations5 of 5 tested for readiness · 5 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Bank and intercompany reconciliations | Test it | Can be licensed |
| Accruals and close flux analysis | Test it | Can be licensed |
| Three-way match in accounts payable | Test it | Can be licensed |
| Revenue-recognition memos | Test it | Can be licensed |
| VAT returns | Test it | Can be licensed |
Hard, multi-entity work suits labs; simplified close tasks are already solved.
Legal5 of 5 tested for readiness · 0 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Playbook contract review | Test it | Synthetic copy |
| Due-diligence red flags | Test it | Synthetic copy |
| Privilege review | Test it | Not licensed: stays in your cloud |
| Citation checks | Test it | Synthetic copy |
| NDA triage | Test it | Synthetic copy |
Privileged work stays in. Labs get synthetic copies, never the files.
Pharma and life-sciences regulatory5 of 5 tested for readiness · 2 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Adverse-event case processing | Test it | Synthetic copy |
| eCTD quality checks | Test it | Can be licensed |
| Health-authority query responses | Test it | Synthetic copy |
| Label comparison | Test it | Can be licensed |
| Protocol deviations | Test it | Synthetic copy |
Label and submission checks can be licensed; patient case data cannot.
Asset and wealth management5 of 5 tested for readiness · 2 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| NAV reconciliation | Test it | Can be licensed |
| Trade breaks | Test it | Can be licensed |
| Suitability checks | Test it | Not licensed: stays in your cloud |
| Client-report commentary | Test it | Synthetic copy |
| Due-diligence questionnaires | Test it | Synthetic copy |
Operations can be licensed; client data cannot.
Telecom and customer operations5 of 5 tested for readiness · 4 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Guided fault diagnosis | Test it | Can be licensed |
| Billing disputes | Test it | Can be licensed |
| Retention conversations within policy | Test it | Can be licensed |
| Complaint classification | Test it | Can be licensed |
| Vulnerable-customer handling | Test it | Not licensed: stays in your cloud |
Troubleshooting suits lab licensing well.
Public-sector casework and permitting5 of 5 tested for readiness · 2 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Planning-application validation | Test it | Can be licensed |
| Building-code checks | Test it | Can be licensed |
| Building-control reviews | Test it | Synthetic copy |
| Benefit casework | Test it | Not licensed: stays in your cloud |
| Licence renewals | Test it | Synthetic copy |
Permitting checks can be licensed as rewrites; benefit decisions stay in.
Management consulting4 of 4 tested for readiness · 0 of 4 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Market sizing | Test it | Synthetic copy |
| Cost-out from a data room | Test it | Not licensed: stays in your cloud |
| Commercial due diligence | Test it | Synthetic copy |
| RFP responses | Test it | Synthetic copy |
Client material is confidential, so it stays in.
Healthcare administration5 of 5 tested for readiness · 0 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| Prior-authorisation packets | Test it | Synthetic copy |
| Denials and appeals | Test it | Synthetic copy |
| ICD-10 and CPT coding | Test it | Synthetic copy |
| Eligibility checks | Test it | Synthetic copy |
| UK clinical coding | Test it | Synthetic copy |
Patient data stays in. Labs get synthetic copies, never the records.
Healthcare administration5 of 5 tested for readiness · 0 of 5 can be licensed
| Task | Readiness | Lab licensing |
|---|---|---|
| UK clinical coding | Test it | Synthetic copy |
| Referral triage | Test it | Synthetic copy |
| Waiting-list validation | Test it | Synthetic copy |
| Outpatient letters | Test it | Synthetic copy |
| Discharge summaries | Test it | Synthetic copy |
Patient data stays in. Labs get synthetic copies, never the records.
The ten, by what a lab could license
- License well: Insurance · Finance operations · Customer operations
- Parts can be licensed: Banking · Pharma regulatory · Asset and wealth · Public sector
- Stay in: Healthcare admin · Legal · Consulting
Licensing would come two ways: your tasks, and the synthetic data made from your records, each made inside your cloud and approved by you.
How licensing would pay: the Data program03 · Beyond the ten
Industries we’re open to exploring.
The ten are where we expect to find strong sets of tasks. Others we take case by case.
Among them: energy and utilities, logistics, retail and e‑commerce, manufacturing, travel and hospitality, software and IT services, real estate, construction and education.
Energy and utilities: where we would look first, meter‑read exceptions, billing disputes and outage updates.
Energy and utilities: where we would look first, meter‑read exceptions, billing disputes and outage updates.
Logistics: where we would look first, freight‑invoice audit and customs classification.
Retail and e‑commerce: where we would look first, supplier invoices, returns and customer contact.
Manufacturing: where we would look first, supplier quality, warranty claims and purchase‑to‑pay.
Travel and hospitality: where we would look first, refunds, disruption rebooking and group bookings.
Software and IT services: where we would look first, support tickets, renewals and security questionnaires.
Real estate: where we would look first, lease abstraction and service‑charge reconciliation.
Construction: where we would look first, tender review, variations and payment applications.
Education: where we would look first, admissions checks and student casework.
Your proposal
Get your proposal.Then register your interest.
A readiness report made from your company’s public website: what your tasks cost, and the first 25 to test. If it reads right, register your interest: a person from Butterfly will call you to talk it through.
Or read on:Agent benchmarksHow it works