Sample report
AI opportunity assessment
Claims operations
A 420-person insurance claims administrator
AI opportunity assessment | Claims operationsContents
- Executive summary3
- Assessment approach4
- Current allocation of time5
- Workflow ranking6
- Value and difficulty7
- Selection of the first build8
- Workflow 1: Medical record and demand review9
- Workflow 2: File notes and client status reports10
- Workflow 3: Large-loss and reserve reports11
- Workflow 4: New claim intake12
- Workflow 5: Recovery spotting on closed files13
- First build: record review14
- Measurement15
- Manager readiness16
- Redeployment of returned hours17
- Responsibilities that remain with staff18
- 90-day plan19
- Next steps20
AI opportunity assessment | Claims operationsExecutive summary
Claims operations can transfer a large share of its reading, drafting and keying to AI systems. This report sizes four workflows. Together they return between $1.2M and $2.0M a year in expert and staff hours at loaded cost, with a most likely figure of $1.7M.
$1.7M
The four sized workflows return this much a year, which is 35,877 hours.
$524K
The first build returns this much a year. That is 8,450 of the 11,700 hours senior adjusters spend reading records.
Week 6
The first six adjusters use it on live files in week six of a 90-day sprint.
- 1Medical record and demand reviewBuild first$524K a year
- 2File notes and client status reportsBuild second$816K a year
- 3Large-loss and reserve reportsBuild third$116K a year
- 4New claim intakeBuy, do not build$278K a year
- 5Recovery spotting on closed filesAudit firstNot sized
Recommendation
Goal Boss recommends building record review first. It only reads documents, so it does not depend on the claim system vendor, and the senior adjusters asked for it. File notes and status reports are worth more, and that build should start once the vendor confirms write access. Large-loss reports come third because they reuse what record review produces. Intake should be bought from the vendor after a trial. Recovery spotting needs a 200-file audit before it can be valued.
Decisions for the leadership team
- Whether to proceed with the 90-day sprint described on page 20.
- Where the returned hours will go, so supervisors can answer adjusters who ask about their jobs. Our recommendation is on page 17.
- Who owns the result. We suggest the VP of Claims.
AI opportunity assessment | Claims operationsAssessment approach
Goal Boss spent five days on site. The assessment drew on 12 interviews, 9 file reviews observed from start to finish, readiness results from all 18 supervisors, and twelve months of the company's claim data.
The company provided a twelve-month export from the claim system, scheduled the interviews, and had all 18 supervisors complete the readiness assessment.
A three-hour working session with the five leaders listed below. The group listed every recurring task in claims operations and identified the ones that consume expert time.
Five interviews with senior adjusters and supervisors. We observed four file reviews from the first page to the evaluation.
Four interviews with adjusters and claims assistants. We observed five more file reviews and two hours of new claim intake.
Three interviews, with IT, compliance and client services. We sized each workflow against the export and drafted this report.
We reviewed the figures with the VP of Claims in the morning and presented this report to the leadership team in the afternoon.
Monday session attendees
- The COO
- The VP of Claims
- The director of workers' compensation claims
- The director of liability claims
- The senior adjuster who trains new hires
Interviews
- Senior adjusters 3
- Supervisors 2
- Adjusters 2
- Claims assistants 2
- IT 1
- Compliance 1
- Client services 1
Basis of estimates
Hours a year are a count from the company's claim system multiplied by a time per task from the interviews and the observed file reviews. Hours returned assume a person still reviews everything a system produces. Loaded cost is salary plus 30% for benefits and payroll taxes, over 2,000 hours a year. One week supports an estimate, so every value in this report is a range with a most likely figure. The first two weeks of a sprint replace the estimate for the first build with a measured baseline.
AI opportunity assessment | Claims operationsCurrent allocation of time
Senior adjusters spend about 27% of their year reading medical records and demand letters, or 532 hours each. Red marks the work this report proposes transferring to a system.
Senior adjusters 22 people
- Reading medical records and demands 27%
- Evaluation, negotiation and settlement 28%
- Calls and email with claimants, counsel and clients 20%
- File notes and status reports 13%
- Large-loss and reserve reports 5%
- Everything else 7%
Adjusters 118 people
- Investigation and contacts 34%
- Evaluation, reserves and payments 29%
- File notes and status reports 13%
- Compliance letters and forms 9%
- Everything else 15%
Claims assistants 16 people
- New claim intake 40%
- Mail and document indexing 30%
- Payment support 18%
- Everything else 12%
Hours by workflow
Shares are estimates from the interviews and the Monday session, checked against the counts in the claim system. They are rounded to the nearest percent.
AI opportunity assessment | Claims operationsWorkflow ranking
The ranking weighs annual value against how hard the work is to change. The largest opportunity is ranked second because it depends on the claim system vendor, and the first build should not depend on a third party.
| Rank | Workflow | Performed by | Hours a year | Hours returned | Annual value | Recommendation |
|---|---|---|---|---|---|---|
| 1 | Medical record and demand reviewIt only reads documents, so nothing waits on the claim system vendor. | 22 senior adjusters | 11,700 | 8,4506,435 to 9,360 | $523,900at $62 an hour | Build first |
| 2 | File notes and client status reportsIt is the largest opportunity, and it has to write into the claim system of record. | 118 adjusters | 28,320 | 16,99211,328 to 19,824 | $815,616at $48 an hour | Build second |
| 3 | Large-loss and reserve reportsIt reuses most of what record review produces. | Supervisors and senior adjusters | 2,700 | 1,7551,350 to 2,025 | $115,830at $66 an hour | Build third |
| 4 | New claim intakeThe claim system vendor already sells a module for this. | 16 claims assistants | 12,400 | 8,6806,200 to 9,920 | $277,760at $32 an hour | Buy, do not build |
| 5 | Recovery spotting on closed filesNobody knows how many recoveries are being missed. | No one is assigned to it | Not sized. A 200-file audit is needed first. | Audit first | ||
| Four sized workflows | 55,120 | 35,877 | $1,733,106 | |||
Annual value is the cost of the hours returned. None of it counts as savings until the hours are redeployed (page 17). Page 4 explains the basis of the estimates.
AI opportunity assessment | Claims operationsValue and difficulty
Each circle is a workflow. Its height is the most likely annual value, its position from left to right is how hard the work is to change, and its size is the hours spent on it today.
Range of estimates
Each bar runs from the low estimate to the high estimate. The red dot marks the most likely value.
AI opportunity assessment | Claims operationsSelection of the first build
A first build should meet three criteria. It should be worth at least $100,000 a year, ship without depending on another company, and have the support of the experts who will use it. Record review is the only workflow that meets all three.
| Workflow | Worth $100,000 or more a year | Can ship without the vendor | Asked for by the experts |
|---|---|---|---|
| 1 Medical record and demand review | |||
| 2 File notes and client status reports | |||
| 3 Large-loss and reserve reports | |||
| 4 New claim intake | |||
| 5 Recovery spotting on closed files |
AI opportunity assessment | Claims operationsWorkflow 1 of 5
Medical record and demand review
Recommendation: build first. It only reads documents, so nothing waits on the claim system vendor.
Current process
When a bodily injury demand or a lost-time file comes in, a senior adjuster reads the medical records, the bills and the demand letter from end to end. Files run from 80 pages to more than 1,000. The adjuster builds the treatment chronology by hand and totals the bills by provider before forming a view of the claim.
Proposed system
The system reads the file and drafts the chronology, the bill totals by provider and a list of flags: gaps in treatment, prior conditions, and places where the demand letter and the records disagree. Every line carries the page it came from.
Retained by staff
The adjuster checks the draft against the pages, corrects it and forms the evaluation. Reserves, coverage and settlement value stay with the adjuster.
Observations
Senior adjusters estimated three to six hours of reading per file. The nine file reviews we observed averaged 4.4 hours. Each adjuster keeps a personal checklist, and the three most experienced adjusters read in nearly the same order.
Estimated value
$524K
This is the most likely annual value. The range is $399K to $580K.
2,600 files a year at 4.5 hours each, at a loaded cost of $62 an hour.
Requirements
- Read access to the document store
- Three senior adjusters for two hours a week during the build
Next step
In the first two weeks of a sprint, senior adjusters log their time on every file, and Goal Boss documents the review order used by the three most experienced adjusters.
Risks
- A missed prior condition is costly. The 50-file comparison on page 15 has to pass before the whole team uses the system.
- Handwritten and badly scanned pages read poorly. About one file in eight has some, and the system marks those pages for the adjuster to read.
AI opportunity assessment | Claims operationsWorkflow 2 of 5
File notes and client status reports
Recommendation: build second. It is the largest opportunity, and it has to write into the claim system of record.
Current process
Adjusters write a file note after every contact and every decision. They also send each client a status report on its open files every 30, 60 or 90 days, depending on the account. Most of a status report restates notes that are already in the file.
Proposed system
The system drafts the note from the call, email or document that prompted it, and drafts each status report from the notes written since the last one. The adjuster edits the draft and sends it.
Retained by staff
The adjuster decides what the note concludes and what the client is told.
Observations
Adjusters estimated four to six hours a week on notes and status reports. We compared four status reports with their files, and most sentences restated an earlier note. Supervisors reported that late status reports are the most frequent client complaint.
Estimated value
$816K
This is the most likely annual value. The range is $544K to $952K.
5 hours a week each, 48 weeks, at a loaded cost of $48 an hour.
Requirements
- Write access through the claim system vendor
- Reporting templates for the 20 largest client accounts
Next step
Ask the claim system vendor to confirm write access and its cost in writing. The build can start when record review reaches all 22 senior adjusters.
Risks
- The system has to write into the claim system. The vendor has an interface for that, and access should be confirmed in writing before any build starts.
- File notes are evidence when a claim is litigated. The system records how long each draft was open before it was saved, so supervisors can see drafts that went in unread.
AI opportunity assessment | Claims operationsWorkflow 3 of 5
Large-loss and reserve reports
Recommendation: build third. It reuses most of what record review produces.
Current process
When a reserve crosses a reporting threshold set by a client or an excess carrier, a supervisor or senior adjuster writes a large-loss report. It covers the facts, the injuries, treatment to date, the exposure and the plan for the file.
Proposed system
The facts, injuries and treatment sections restate what record review has already produced, so the system drafts those three and leaves the exposure and the plan to the writer.
Retained by staff
The writer does the exposure analysis, recommends the reserve and sets the plan for the file.
Observations
The export shows 900 reports in twelve months. Supervisors put each one at two to four hours and said the facts and treatment sections take more than half of that time.
Estimated value
$116K
This is the most likely annual value. The range is $89K to $134K.
900 reports a year at 3 hours each, at a loaded cost of $66 an hour.
Requirements
- Record review live and trusted
- Ten recent reports in each carrier format
Next step
Collect ten recent reports in each of the three main carrier formats during the sprint, so this build can start in the following quarter.
Risks
- Each excess carrier wants its own format. The first version covers the three formats that account for most of the reports.
- The draft is only as good as the record review behind it, so this waits until record review has passed its quality check.
AI opportunity assessment | Claims operationsWorkflow 4 of 5
New claim intake
Recommendation: buy, do not build. The claim system vendor already sells a module for this.
Current process
New claims arrive by email, web form, fax and phone, in about a dozen formats. Sixteen claims assistants key each one into the claim system, which takes about 24 minutes a claim.
Proposed system
Intake automation is a product. It reads the incoming notice, fills in the claim record and routes the file to an adjuster. Several vendors sell it, including the company's own claim system vendor.
Retained by staff
Claims assistants handle the exceptions: notices with missing or conflicting information, and anything the module is unsure about.
Observations
We observed two hours of intake. A clean web form took under ten minutes, and a faxed notice with handwriting took more than forty. The assistants estimated 20 to 30 minutes for an average claim.
Estimated value
$278K
This is the most likely annual value. The range is $198K to $317K.
31,000 new claims a year at 24 minutes each, at a loaded cost of $32 an hour.
Requirements
- A four-week trial of the vendor module on last quarter's claims
- A per-claim price in writing
Next step
Request the trial and the price from the vendor this month. The COO owns this workflow, and the trial can run alongside the sprint.
Risks
- A custom build would mean maintaining a dozen input formats and a write path into the system of record, for a result the vendor already supports.
- Vendor pricing determines the outcome. At 31,000 claims a year, a fee above about $9 a claim leaves no saving.
AI opportunity assessment | Claims operationsWorkflow 5 of 5
Recovery spotting on closed files
Recommendation: audit first. Nobody knows how many recoveries are being missed.
Current process
When someone else caused a loss, the company can recover what was paid from that party or from their insurer. Adjusters are supposed to flag those files. Under caseload pressure some get missed, and no one reviews closed files to catch them.
Proposed system
A system can read every file closed in the last two years for the facts that point to a recovery: another vehicle, a defective product, a contractor on site.
Retained by staff
A recovery specialist decides whether to pursue each one and makes the demand on the other party.
Observations
Three supervisors raised missed recoveries without being asked, and none could estimate how many there are. The claim system has a recovery flag, and it was set on 3% of the files closed last year.
Estimated value
Not sized
The audit comes first. Its result determines whether this workflow is worth building.
Requirements
- A 200-file audit by a recovery specialist
- The recovery terms in the ten largest client contracts
Next step
Commission the 200-file audit this quarter. If the audit finds enough, Goal Boss will size the workflow and rank it against the second build.
Risks
- The miss rate is unknown, so the value is unknown. A reviewer with recovery experience can settle it by reading 200 closed files, which takes about two weeks.
- The company keeps a share of each recovery on most accounts, so any value here arrives as revenue and depends on those contract terms.
AI opportunity assessment | Claims operationsFirst build: record review
A bodily injury or lost-time file can carry several hundred pages of medical records, bills and a demand letter. Reading them is the most expensive routine work in claims operations.
Sample draft, from an invented file
| Date | Entry | Pages |
|---|---|---|
| March 3 | Emergency department. Seen after a fall from a loading dock. X-ray of the left wrist, no fracture. | p. 14 |
| March 10 | Orthopedist. Wrist sprain. Splint, and light duty for four weeks. | p. 31 |
| April 21 | Physical therapy. First of 12 visits. | p. 58 |
| July 8 | Orthopedist. Released to full duty. | p. 112 |
| Flag | Gap in treatment. No visits between April 30 and June 24. | pp. 71 to 96 |
| Flag | Prior condition. A 2019 injury to the same wrist appears in the intake history and is not mentioned in the demand. | p. 16 |
| Flag | Demand and records disagree. The demand letter claims 22 therapy visits. The records show 12. | pp. 58 to 104 |
Design basis
The system follows the way the company's three most experienced adjusters review a file: the order they read in, what they look for, and what makes them stop. They work with Goal Boss to document that process, and they test every version before anyone else sees it.
Hosting and scope limits
The system runs inside the company's own cloud account, and claim documents are not used to train any model. Model and hosting fees at 2,600 files a year should stay under $20,000. The first version does not write to the claim system, value the claim or suggest a reserve.
AI opportunity assessment | Claims operationsMeasurement
The sprint is judged on hours per file and on whether the system misses anything a person would have caught.
Hours to review one file
Speed
In weeks one and two, the senior adjusters log their time on every file they review, which replaces the estimate with a measured baseline. From week six, the system logs time per file for each adjuster using it. At the target, each file takes 3.25 fewer hours, which is 8,450 hours a year.
Quality
In weeks ten to twelve, 50 files are reviewed both ways. A senior adjuster compares what each review missed without knowing which was which. The system passes if it misses no more than the manual review does.
Weekly scorecard for the CEO and the sponsor
- Files reviewed with the system, and by whom
- Average hours per file against the baseline
- Corrections adjusters made to the drafts
- Anything the system missed, and what was done about it
Stop condition
If the comparison shows the system missing items that the manual review caught, the full team does not receive it until that is fixed. The six adjusters from week six keep using it under the same check.
AI opportunity assessment | Claims operationsManager readiness
All 18 supervisors completed the Goal Boss readiness assessment before the week began. The bars show the average score in each of its five areas, out of 10.
Implications for the first build
- No supervisor has been told what to say when an adjuster asks whether this costs jobs. The company needs that answer before the first adjuster sees the system. Our recommendation is on page 17.
- 11 of 18 supervisors could not say which adjuster tasks the company expects AI to change. Adjusters will ask them in the first week of the build.
- Only 5 of 18 teams hold a weekly review. Adoption is tracked in that review, so the other 13 teams need one before week six.
Supervisor preparation, complete before week six
- The company's answer on jobs, in writing, with the suggested statement on the next page
- A one-page list of the adjuster tasks this build changes and the ones it leaves alone
- A 20-minute agenda for the weekly review: usage, hours per file, corrections, problems
Nothing in the assessment reads sentiment or infers how anyone feels. Supervisors answered direct questions about their own teams.
AI opportunity assessment | Claims operationsRedeployment of returned hours
Record review returns about 8,450 senior adjuster hours a year. The company should decide how those hours will be used before the first adjuster sees the system, because adjusters will ask.
Recommended allocation of the 8,450 hours
- Two client accounts arriving next year (4,200 hours) 50%
- Lower caseloads on the complex desk (2,700 hours) 32%
- Quality review and training new adjusters (1,550 hours) 18%
Staffing impact
Attrition ran at 19% last year, so about 27 of the 140 adjusters leave in a normal year. Two client accounts arrive next year and would otherwise need new hires. The company can absorb that work with its current staff, without layoffs, and can tell them so.
Path to EBITDA
The saving appears when the new accounts are staffed without hiring and when departures on the complex desk are not all replaced. The CEO and the VP of Claims should agree on that plan in week one and report against it each quarter.
Suggested statement for supervisors
This system reads the records and drafts the chronology. You check it, and you make the call on the claim. Nobody is losing a job because of it. The hours it gives back are going to the two new accounts and to bringing the complex caseloads down. If that plan changes, you will hear it from me before you hear it from anyone else.
AI opportunity assessment | Claims operationsResponsibilities that remain with staff
Some work should remain with people at this company regardless of the hours involved.
- Reserves, coverage positions and settlement authority stay with licensed adjusters and their supervisors. The system prepares the file and never values the claim.
- Any letter that denies or reduces a benefit is written and signed by a person.
- Conversations with injured workers and claimants stay with adjusters.
Governance
State insurance regulators have issued guidance on the use of AI in insurance, and carrier clients will ask how the system is controlled. A one-page description ships with the first build. It states what the system does, what it never does, who reviews its output, and where the documents are processed.
AI opportunity assessment | Claims operations90-day plan
The sprint covers record review for the 22 senior adjusters. The VP of Claims owns the result. Goal Boss builds the system and runs the weekly review with the supervisors.
Weeks 1 to 2
The CEO decides where the returned hours go and supervisors receive the language for it. Senior adjusters log time per file so the baseline is measured.
Weeks 3 to 8
Goal Boss builds record review with the three most experienced adjusters. Six senior adjusters use it on live files from week six.
Weeks 9 to 12
All 22 senior adjusters use the system. Each supervisor reviews usage and time per file weekly, and the 50-file quality check runs.
Week 13
Time per file is measured against the baseline and reported to the CEO and the sponsor, with a recommendation on the second build.
AI opportunity assessment | Claims operationsNext steps
The findings support starting with record review. The steps below would put it in the hands of all 22 senior adjusters within 90 days and prepare the two builds that follow.
For the leadership team
- Decide where the returned hours will go, and give supervisors the answer before the build starts.
- Name the leader who owns the result.
- Ask the claim system vendor to confirm write access in writing and to quote its intake module.
- Commission the 200-file recovery audit.
How Goal Boss can help: a 90-day sprint
Deliverables
- Record review, built and running in the company's cloud account
- A measured baseline and a measured result
- Supervisors who can run it and explain it
- A recommendation on the second build
Client commitments
- The VP of Claims as the accountable leader
- Three senior adjusters for two hours a week
- Read access to the document store
- Thirty minutes a week with the CEO
Fixed, and stated in the company's copy of this report. The fee for this assessment comes off it when the sprint is signed within 30 days of this report. After 30 days the credit lapses.
Goal Boss can begin on the second Monday after approval.
A weekly scorecard, and a final report in week thirteen.
About Goal Boss
Goal Boss is led by Will Pemble, who built and sold Web.com, one of the largest web hosts on earth. He has been a full-stack engineer since 1995 and has worked with AI and machine learning since 2018, and he has spent more than twenty years helping owners and leadership teams get people to do the work that moves the number. He is the author of Goal Boss: The Art and Science of Getting Stuff Done and Judgment Day: How to Lead When AI Breaks Everything.
goalboss.com
Want this for one of your functions?
An assessment takes one week on site at a fixed fee. It starts on Monday with your senior experts in a room, and you get the report in person on Friday afternoon.
