Loss adjuster software for a small firm should cover six jobs: instruction intake, a claim file with a diary, site-visit capture, estimating where you cost repairs, report drafting and fee billing. AI helps most with reading documents, tidying notes and drafting report sections. It should never decide cover, set reserves or sign the report.
No single product does all six well. Most firms end up with a claims file system at the centre, an estimating tool if they price repairs, a capture app for photos and an AI assistant, so the real buying question is how much of the AI sits inside the first of those. Software built for public adjusters, who act for the policyholder, is a different shape; AI for public adjusters covers that side.
Figures below are USD list prices from the vendors' own pages, rechecked on 4 November 2026. Several products publish no price and are marked quote-only.
One rule runs through everything that follows: AI drafts, the adjuster decides and signs. Software that shows what the AI summarised, what it drafted and what a person approved is worth more than software that merely has an AI button.
The six jobs a small firm's software must cover
Use this table as a demo script. Each row says what the job needs and a test you can run with one of your own closed files, so the vendor demonstrates your work rather than theirs.
| Job | What it has to do | What to test in a demo |
|---|---|---|
| Instruction intake | Capture insurer, policy number, insured and contact details, loss date and type, authority and reporting deadlines, and open the file without re-keying | Forward a real instruction email and see which fields fill |
| Claim file and diary | One record per claim holding documents, emails, diary dates, status and an audit trail | Set three diary dates, close one, then find out who changed what |
| Site-visit capture | Appointment booking, notes, labelled photos, sketches and measurements, offline use | Run a visit on a phone with no signal and see whether it syncs |
| Estimating and scope | Line-item estimates priced for the loss location, or a place to store and compare the repairer's figures | Price three items and ask where the price data comes from |
| Report drafting | Templates for first, interim and final reports, photo placement, versions and sign-off | Produce a report from a closed file and time it |
| Fee billing and reporting | Time and expenses, the fee schedule each insurer client sets, invoices, receivables and a file-ageing view | Invoice one claim from its time entries without a spreadsheet |
Estimating is the conditional one. Many loss adjusters don't price repairs themselves; they review a contractor's or surveyor's figures. If that is you, job four shrinks to storing and comparing quotes, and the largest line in a typical software budget disappears.
Instruction intake deserves extra scrutiny because it is where re-keying hurts most. If your instructions arrive through insurer portals, ask each vendor which portals it connects to and which it handles by email forwarding or manual download, since a tool can't read a portal it can't reach. AI submission intake for insurance brokers shows the same re-keying problem from the broker side.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowFour kinds of tool, with prices where vendors publish them
The table covers one tool from each kind that I could verify on its own page. It is a starting list, not a league table, and the "fits when" column is my judgement.
| Tool | Covers | List price | AI on the vendor's page | Fits when |
|---|---|---|---|---|
| FileRidge | Claim files, receivables, invoices generated from carrier fee schedules, real-time reporting, field logging of time and notes | $40 per user a month, all features, month-to-month | AI voice field logging, listed without detail | A small firm wants a light claims and billing system (the page says it is built for firm owners and claims managers, not field estimating); check references and its security statement as you would with any small supplier |
| VCA ClaimsCore | Intake, assignment, handling, payments, reporting and an audit trail, with add-on modules | Quote only; no per-user pricing | ClaimsAI, an add-on: summarises uploaded documents, helps at claim creation, answers questions about a claim, flags what is due | You want configurable workflows and a vendor-led setup |
| Xactimate Pro | Sketching, scoping, line-item estimates with localised pricing and reports, on desktop, online and mobile | $2,690 a year on a 12-month term ($350 a month on a one-month term); XactAI add-on at an introductory $29 a user a month billed annually ($348 a year) or $39 month to month | XactAI: describes photos automatically and gives plain-language line-item recommendations | Your adjusters write repair estimates |
| CompanyCam | Photo capture and organisation for site visits | Core about $63 a month on annual billing | The plan table lists 10 AI credits on Core, though its FAQ says AI features come with Crew and above; confirm | Visits are photo-heavy; it is not a claim file system |
| ChatGPT Business, Claude Team or Microsoft 365 Copilot Business | General drafting, summarising and extraction from your own prompts | ChatGPT Business and Claude Team: $20 a seat on annual billing, $25 monthly. Copilot Business: $21 annual, $25.20 monthly | Not built for claims, so no diary or file | You want drafting help beside any of the above, on a plan that doesn't train on your content by default |
Four details from the vendor pages matter more than the prices. VCA says ClaimsAI hands every judgement back to a person, and its guide never decides a claim, sets a reserve, approves an action or authorises a payment. Verisk lists independent adjusters and adjusting firms among Xactimate's users, but says its pricing research covers specific markets, so check that your loss locations are covered before you pay. VCA publishes no price list, so ask for ClaimsAI as its own line in the quote. And Xactware's product page says Standard covers one platform of your choice while Pro covers all three; Standard's price isn't printed there, so a single-platform plan may cost less than $2,690.
When each option wins, and five buying mistakes
A claims file system wins when billing and diary compliance are the pain: invoices going out late, diary dates living in someone's head, nobody able to say how many files are open. An estimating tool wins only if your adjusters write estimates. A capture app wins when visits produce more photos than anyone can sort. A general assistant wins on its own only for a firm of one or two adjusters, where a spreadsheet diary and a shared folder still work.
Take a two-adjuster firm with about 12 new instructions a month. A shared spreadsheet for the diary, two assistant seats on annual billing ($40 a month) and a report template could serve for a year. A file system becomes worth buying when invoices start leaving late, or a missed diary date costs you an insurer client, and not before.
- Buying estimating software for adjusters who review figures rather than write them.
- Adding an AI feature before the file system exists, so it has no clean diary or documents to work from.
- Judging by the vendor's demo file instead of one of your own closed files.
- Pasting claim documents into a free personal assistant account because it is quicker.
- Skipping the exit terms: you need to leave with every document, note and diary entry intact.
Where AI fits at each stage of a file
The table gives the division of labour. The three examples after it show what the first, third and fifth rows look like in practice, with the mistakes a careful adjuster would catch.
| Stage | AI can | You check | Stays with the adjuster |
|---|---|---|---|
| Instruction intake | Pull fields from the instruction email or PDF into a draft record | Policy number, loss date, contacts and deadlines against the original | Accepting the instruction, conflict check, authority limits |
| Document reading | Summarise policy wording, schedules, repair quotes and earlier reports | Every quoted clause and figure against the source page | Any opinion on cover |
| Site visit | Turn a voice note into structured notes and label photos | Room names, readings, and what each photo actually shows | What was seen, what was said, who attended |
| Chronology | Order emails and notes into a dated timeline | Dates and who said what | Deciding which events matter |
| Report draft | Draft circumstances and inspection findings from your notes | Facts, hedged wording and anything it added | Cause, cover observations, quantum, recommendations, signature |
| Fee billing | Turn time entries into a narrative for the invoice | Hours and descriptions match the file | Fees agreed with the insurer |
| Status letters | Draft update emails from the file | Tone, dates and promises | What and when to tell the insured |
Reading an instruction email into a draft record
Give the assistant a strict extraction job and forbid guessing. The prompt below suits any of the tools above, or a business assistant you paste into.
Read the instruction email below and return a table of: insurer, claim or policy number, insured name, site address, date of loss, reported cause, authority or reserve (if stated), first-contact deadline, first-report deadline, special instructions.
Copy values exactly as written. If a field is missing, write NOT STATED. Do not guess or convert relative dates.
[paste instruction email]A realistic reply, illustrative, fills eight fields correctly and writes "Date of loss: 14 October". The email actually said "last Tuesday", and the model worked out a date without knowing when the email was sent. The fix is in the prompt and in the habit: forbid relative-date conversion, and check every date against the original before the file opens, because the loss date drives the diary and sometimes the cover discussion.
Site notes: a voice note before and after
Dictating on site is faster than typing, and AI can sort the result. Here is a raw dictation, then the structured version it produced (both illustrative).
Raw: "kitchen ceiling stain about a metre square near the extractor, damp to touch, meter reads 38 near the stain and 12 at the far wall, insured says it started after the boiler repair last week, bathroom above has a new silicone seal, couldn't get into the loft."
- Location: kitchen ceiling, beside the extractor.
- Observed: stain roughly 1 m by 1 m, damp to touch.
- Readings: 38 at the stain, 12 at the far wall (scale as shown on the meter; confirm).
- Insured's account: began after a boiler repair the previous week (reported, not verified).
- Also noted: new silicone seal on the bathroom above. Loft not accessed.
- To follow up: bathroom seal and plumbing, loft access, boiler repair invoice.
What a careful adjuster fixes: the AI left the meter units open, which is right, but it also drifted slightly on "roughly 1 m by 1 m", which was dictated as "about a metre square". Keep the dictated wording, hold the insured's account apart from what you saw, and never let the tool fill gaps such as the cause. Roof damage reports for insurance claims takes the same "document, don't adjust" approach for photos.
A report draft that marks its own gaps
The riskiest use is also the most tempting: drafting report sections. Limit it to sections that restate your own notes, and make the assistant flag what it doesn't have.
Using only the file notes below, draft two report sections: 1) Circumstances reported and 2) Inspection findings.
Use neutral wording such as "the insured states" and "I observed". Do not state a cause, a coverage view or any figure that is not in the notes.
Wherever information is missing, insert [CHECK: what is missing] instead of filling the gap.
[paste notes]The illustrative reply says: "The insured states that staining appeared after a boiler repair the previous week. I observed a damp stain of about one metre square on the kitchen ceiling beside the extractor [CHECK: moisture meter scale and make]. The loft was not accessed [CHECK: reason and whether access can be arranged]." Two things to fix: tighten any verb that implies more certainty than your notes support, and replace each CHECK with the fact or a decision to leave it open. The sections on cause, cover observations, quantum and recommendations stay yours, and so does the signature.
Fee narratives from time entries
This is a safer first use than report drafting, because the input is a list of hours rather than a judgement. Give the assistant the entries (0.4 hours reading the instruction, 0.3 hours first contact with the insured, 2.5 hours attending site and photographing damage, 1.8 hours drafting the first report) and ask for one invoice paragraph using only those entries. A good result reads: "Reviewed the instruction and made first contact with the insured (0.7 hours), attended site and recorded the damage (2.5 hours) and prepared the first report (1.8 hours)." Check that the hours still total 5.0 and that each description matches the file. A model that rounds or merges entries has quietly changed your bill.
For ongoing correspondence, AI for insurance adjusters covers claim-file summaries and status letters in more detail than this comparison can.
What the adjuster checks before signing
A signature tells the insurer that a competent person stands behind every line. Make it mean that, with a short check you do every time, whether or not AI touched the draft. AI home inspection software: what it writes, who signs and what it costs reaches the same conclusion for another profession that signs reports.
- Every date, name, policy number and figure matches its source document.
- Every sentence about cause is yours; delete anything the draft asserts that you didn't observe or conclude.
- Facts you saw, the insured's account and your opinion sit in separate, clearly labelled passages.
- Each photo matches its caption and room.
- Quoted policy wording matches the policy, word for word.
- Quantum figures reconcile to the estimate or the repairer's quotes.
- Nothing goes beyond your instruction or authority, such as a cover position the insurer reserved for itself.
- You could explain each paragraph to the insurer on the phone without looking at the draft.
Two practical points sit alongside the checklist. Claim files hold personal and sometimes health data, and insurer instructions or panel terms may restrict third-party tools or require disclosure, so read them before anything leaves your systems. And use business plans or the AI inside your claims software rather than free personal accounts: ChatGPT Business and Claude Team both state on their pricing pages that they don't train on your content by default.
A five-adjuster firm: the monthly bill and a break-even test
Say a five-adjuster firm opens 40 new instructions a month and writes about 60 reports. Two of the five write repair estimates; the other three review contractors' figures. The stack is a claims file system for everyone, Xactimate Pro with XactAI for the two estimators, and a business assistant seat for all five.
| Line | Working | Monthly cost |
|---|---|---|
| Claims file system (FileRidge list price) | 5 users at $40 | $200.00 |
| Xactimate Pro, two estimators | 2 x $2,690 a year, divided by 12 | $448.33 |
| XactAI add-on at its introductory price, two estimators | 2 x $348 a year, divided by 12 | $58.00 |
| Business assistant seats | 5 x $20 on annual billing | $100.00 |
| Total | $806.33 (about $9,676 a year) |
The AI share of that bill is the XactAI add-on plus the assistant seats, $158 a month. At an illustrative internal cost of $50 an hour for adjuster time, that is covered if the team saves about 3.2 hours a month, which across 60 reports is a little over three minutes a report. So the question isn't whether AI could save time in theory. It is whether you can measure three minutes a report, and the trial below does exactly that.
The other $648 pays for things AI doesn't replace: a file system and an audit trail, and estimates priced from a maintained database. Judge those by what they replace, usually spreadsheets, shared drives and manual invoicing, and by whether the diary and invoicing tests in the first table pass on your own files.
Demo questions that expose a weak AI claim
- Show one real instruction going in and one report draft coming out, with the source for each statement.
- Which model processes our files, where does it run, and is any of our data used to train it?
- Can AI be switched off for a file, a user or the whole firm?
- Is AI-written text marked in the audit trail, and can we see who approved it?
- What happens when the tool can't find a fact: does it leave a gap or fill one?
- Is each AI feature live today or planned, and is it priced as its own line?
- Can we export every file, with its documents and diary, if we leave?
Treat a vague answer to the third and fourth questions as a "no". A feature you can't switch off or audit is a risk to your signature, however good the demo looked.
A two-week trial on ten closed files
Take ten closed files of different types and run each through the tool as if it were live. Time every step, then compare the draft with the report you actually issued. Keep a tally of errors by type: wrong date, wrong name, wrong figure, invented statement, missing caveat. Do it in a sandbox or on a plan you have already cleared for claim data, never on a free personal account.
My suggested pass mark is cautious. If any draft contains an invented statement of fact, restrict the tool to summarising until the cause is understood. If dates and figures are wrong in more than one file in ten, keep AI out of intake. If the time saved is under the three-minutes-a-report line from the sum above, the add-on isn't earning its place yet, whatever its marketing says. If instead the tool saves visible time and the check list in this tutorial catches every error, roll it out one stage at a time and review the log monthly.
Questions adjusting firms ask before buying
Can AI write a loss adjuster's report?
It can draft sections from your notes and photos, such as the circumstances and inspection findings, but the adjuster must check every fact, add the opinion and sign. Tell the tool never to state cause, cover or figures it wasn't given, and to mark gaps for you. The signature should mean you verified the content.
Is it safe to paste claim documents into ChatGPT or Claude?
Not into a free personal account. Claim files hold personal and sometimes health data, and insurer instructions or panel terms may restrict third-party tools. Use a business plan that doesn't train on your content by default, or the AI built into your claims software, and read your instruction terms first.
Do I need Xactimate if I don't write estimates?
No. It is a property claims estimating tool, with the Pro subscription listed at $2,690 a year. If you appoint contractors or surveyors to cost repairs and only review their figures, a claim file system and an assistant cover most needs. Check that its pricing data covers the markets you work in.
What should a demo show me about the AI?
Ask to see one real instruction go in and one report draft come out, with the source for each statement. Ask which model processes the data, where it runs, whether it trains on your files, whether AI can be switched off per file, and whether AI-written text is marked in the audit trail.
Further reads
- Structural Engineer Inspection Report: Photos and Notes to a Draft With AI — Another signed report built from notes and photos
- Turning On-Site Voice Notes Into Job Sheets and Quotes With AI — Turn on-site voice notes into structured records
- Damp Survey Report: Write Findings and Quote Lines With AI — Findings written clearly from a survey
- Software for Insurance Agents and Brokers: Where AI Fits the Stack — How the same buying question looks on the broker side
- How Brokers Use AI to Compare Policy Wordings at Renewal — Compare policy wordings with AI without trusting it blindly
- AI Meeting Note-Takers Compared for Small Teams — Pick a note-taker for insured and contractor calls
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Verisk's Xactimate product page, Xactware's Xactimate and XactAI pages, VCA Software's home and pricing pages, the FileRidge site, and the pricing pages for ChatGPT Business, Claude, Microsoft 365 Copilot Business and CompanyCam, rechecked on 4 November 2026.