AI for Employee Benefits Brokers: Renewal Decks and Benefit Guides

Give the AI the carrier's renewal PDF and ask for a table of every rate, plan feature and change, with the page number beside each figure. Check that table against the carrier's own totals, then build the client deck and the employee guide from the checked table rather than from the PDF.

That order is the whole method. Benefits renewals fail through small numerical slips: a rate for the wrong tier, a deductible read from the previous year, a footnote ignored. Letting AI summarise a 40-page PDF straight into slides hides those slips. Letting it extract to a table first, with page references, makes each one visible and checkable in minutes.

Follow me on Instagram@sagnikteaches

Why renewals are a good fit, and where AI stops

A renewal pack has a predictable shape: a carrier proposes new rates and sometimes changed plan terms, the broker compares them with the current plan and the market, and the client decides. Much of the effort is clerical: lifting numbers out of PDFs, laying them side by side, and rewriting jargon for a finance director or a team of employees. AI helps with that clerical layer.

Connect on LinkedInSagnik Bhattacharya

It does not decide which plan to recommend, negotiate with the carrier, or judge whether a rate increase is reasonable for this group's claims experience. Those stay with the broker. It also should not be trusted to recall a plan's terms from memory. Every figure in the output must trace to a page in a document you supplied.

Subscribe on YouTube@sagnikteaches

The related policy-wording comparison workflow covers the wording side of this; here the focus is the renewal pack itself.

Stage 1: Prepare the inputs

Collect what the AI will see, and strip what it should not.

• The renewal PDFs from each carrier: the current plan, the renewal offer and any alternative options. Use the original PDFs, not screenshots. If one is scanned, run it through OCR first and check the text for obvious mangling.
• The census summary as counts, not names: employees by coverage tier, age bands and any waiting-period categories. Remove names, birth dates and identifiers.
• The client brief: the contribution strategy, the budget ceiling, any priorities the client mentioned (a richer dental plan, a lower employee deductible).

Use a business-grade plan. On the vendors' pricing pages in November 2026, ChatGPT Business and Claude Team both start at $20 a seat a month billed annually ($25 monthly) with a two-seat minimum, and the Microsoft 365 Copilot Business add-on is $21 a user a month on an annual commitment ($18 in the first year if you buy before 31 December 2026). All three say business content is not used to train their models by default; verify that on the current page for the plan you pick, and check any client agreement that limits third-party processing.

Stage 2: Extract the comparison table

This is the stage that carries the risk, so spend your care here. Upload one carrier's renewal at a time and use a prompt like this:

You are assisting a benefits broker. Use ONLY the attached renewal document.

Extract a table with these columns:
Plan name | Coverage tier | Current rate | Renewal rate | Change (amount and %) | Deductible | Out-of-pocket maximum | Copay or coinsurance for primary visit | Network or other key change | Page number

Rules:
- Every figure must have a page number. If you cannot find a figure, write NOT FOUND, never estimate.
- Keep rates per month and state the billing basis (per employee, per member).
- After the table, list every change in plan design between current and renewal, quoting the clause.
- List footnotes, conditions or rate guarantees that limit a figure.

Run each carrier separately, then ask for a merged table in a second step, using only your verified extracts. Merging inside one giant prompt is where figures drift between carriers.

What a good extract looks like

An illustrative extract for a medical plan might read as follows. The numbers are invented for the example.

Gold PPO | Employee only | $612.00 | $689.40 | +$77.40 (+12.6%) | $1,000 | $4,500 | $25 copay | Pharmacy tier 3 copay raised | p.4
Gold PPO | Employee + family | $1,742.00 | $1,962.80 | +$220.80 (+12.7%) | $2,000 | $9,000 | $25 copay | No change | p.4

Notice that the percentage can be recomputed, which is the first check: if $689.40 minus $612.00 over $612.00 does not give about 12.6 per cent, a number is wrong.

Stage 3: Verify before anything becomes a slide

Treat the extract the way a reviewer treats a junior's spreadsheet. A short routine does the job.

1. Recompute the changes. Paste the table into Excel and recalculate each change and percentage yourself. If you work in Excel often, Claude for Excel can help build the formulas, but the arithmetic must still be yours to verify.
2. Tie out totals. Multiply each rate by the enrolled counts from the census summary and compare the monthly total with the carrier's stated premium on its summary page. A difference of more than a few dollars signals a wrong tier or a missed rate.
3. Spot-check five figures against the PDF by page number, picking at least one from each plan, one deductible and one footnote.
4. Read the NOT FOUND rows. They are the honest outputs. Fill them by hand or ask the carrier.

Log what you checked in one line on the file ("table checked against pp. 3-9, totals tied to carrier summary, 2 rows corrected"). That line is your defence if a client later queries a number.

An 85-employee renewal, timed and costed

Take an illustrative 85-employee client with a current medical plan and a renewal offer. The renewal PDF is 38 pages; two alternative options come from a second carrier at 22 pages each.

Working by hand, a benefits analyst might spend about 3 hours lifting rates and plan features into a spreadsheet and another hour checking, giving 4 hours for the comparison alone. With the extract-and-verify routine, a plausible split is 20 minutes preparing and uploading, 15 minutes of AI runs across three documents, 45 minutes of verification and correcting, so about 80 minutes. The saving of 2 hours 40 minutes per client is illustrative, and verification, not extraction, is the part that cannot shrink.

Suppose the checked table shows the current plan at $48,000 a month across the group and the renewal at $54,000, a 12.5 per cent rise. Alternative A from the second carrier comes in at $50,400 with a higher deductible. The table makes the decision visible: the cheaper option saves $3,600 a month, or $43,200 a year, but moves $500 of deductible onto each enrolled employee. That trade-off, not the raw rate, is what the deck needs to show.

Stage 4: Build the client deck from the checked table

Ask for one slide at a time, with the checked table as the only source. A sequence that works for a finance-minded client:

1. Where we are: current cost, headcount, average cost per employee.
2. What the carrier proposed: the renewal rate and percentage change.
3. What changed in the plan: design changes in plain words, ranked by impact on employees.
4. The options: three columns for renewal, alternative A, alternative B, with annual cost and the employee-side differences.
5. What each option means for employees: the deductible and copay differences, shown for a typical employee and a high-use family.
6. Our recommendation and next steps: yours, written by you.

Using only the verified comparison table below, write the text for slide 4 ("The options").
Three columns: Renewal, Alternative A, Alternative B.
For each: monthly group cost, annual group cost, change vs current, deductible, out-of-pocket maximum, one sentence on the main employee-side trade-off.
No recommendations. No figures that are not in the table.

[paste verified table]

Paste the generated text into your own template. Keep visuals simple: one table, one bar chart of annual cost by option. Every number on a slide should match the table to the cent, so do a final "slide against table" read before the deck leaves.

Stage 5: Plain-English benefit guides for employees

The employee guide is where AI earns most of its keep, because translating plan jargon into readable text is slow for humans and cheap for a model. The risk is shifting from numbers to advice. Constrain it.

Write a one-page guide for employees explaining the plan below in plain English at a reading age of about 12.
Use only the plan summary supplied. Explain: what each term means (deductible, copay, out-of-pocket maximum), what changed this year, and three examples of what an employee would pay.
Do not recommend a plan. Do not predict claims. End with: "For your own situation, check the plan documents or contact [team name]."

[paste plan summary]

Before and after: one sentence rewritten

Before, from a plan summary: "Covered services are subject to the annual deductible, thereafter coinsurance applies at 20 per cent up to the member's out-of-pocket maximum." After, from the AI, lightly edited: "You pay the first $1,000 of your care in a year. After that you pay 20 per cent of each bill, until your total spending in the year reaches $4,500. Beyond that the plan pays." The rewrite is only safe because the figures were in the source; check each one.

Worked cost examples for employees

Ask for three scenario boxes built from the plan figures: a routine year (two primary visits and a prescription), a moderate year (a specialist and a scan) and a high-cost year (a hospital stay). Use illustrative amounts you supply, and have the AI show the arithmetic of what the employee pays under the plan. Check each sum yourself. Employees remember "in a moderate year you would pay about $1,450" far better than a list of definitions, but a wrong sum in a guide is a bigger problem than a wrong sum in an internal table.

More ways to use the same table

Spot the renewal changes buried in footnotes

Run the renewal PDF a second time with a narrow prompt: "list every sentence that limits, excludes or conditions a benefit, with page number". Footnotes about rate guarantee periods, waiting-period changes or new prior-authorisation rules often matter more to the client than a headline percentage. This second pass takes about five minutes and can surface design changes that the first, table-focused run skipped.

Draft the cover email to the client

From the verified table and your recommendation, ask for a 150-word cover email stating the headline figure, the decision needed and the date it is needed by. Write the recommendation yourself and paste it into the prompt; the AI only tidies the language.

Prepare answers for the client's likely questions

Ask for ten questions a finance director would ask about this renewal, then answer them in your own notes before the meeting. You will not use all of them, but two or three often reveal a number you have not prepared, such as the cost of keeping the current plan for another quarter.

Build a year-on-year tracker

Once the verified table exists, store it in a simple sheet with one row per plan per year. In 12 months, asking the AI to compare the two years' tables is quick and reliable because both are already structured. This is the beginning of the dataset you need for automating renewal follow-ups.

Three client types, three different decks

The same checked table supports different conversations. Adjust the emphasis, not the numbers.

A 25-person professional firm

The owner usually decides alone and cares about cost per head and whether the team will complain. Keep the deck to five slides, lead with the annual cost difference, and show one scenario box for an employee with a family. The plain-English guide matters more here than the deck, because the owner will hand it straight to staff.

A 150-person manufacturer with shift workers

The decision involves a finance lead and an HR manager, and employees are less likely to read long documents. Ask the AI for a one-page guide with a short section per plan and a table of what an employee pays in a routine, moderate and high-cost year. Shift workers often compare out-of-pocket maximums more than premiums, so put that line near the top.

A 400-person group with a contribution strategy debate

The renewal is a board-level conversation about how much of the increase the employer absorbs. Use the table to model three contribution splits (employer pays 100, 90 or 80 per cent of employee-only cover) and show the annual cost to the company and the monthly payroll deduction per employee. The AI can draft the narrative, but you should build and verify the model in a spreadsheet, because a wrong split here is a very expensive mistake.

A filled-in verification log

A verification log should be short enough that people actually fill it in. This illustrative entry shows the level of detail that is enough.

Client: [client name] | Renewal date: [date] | Prepared by: [your name]
Source documents: Carrier 1 renewal (38 pp), Carrier 2 options A and B (22 pp each)
AI extract run: 3 documents, 3 runs, table merged by hand
Checks: recomputed all % changes (2 rounding differences fixed); tied monthly total to carrier summary p.3 (matched within $1.20); spot-checked 7 figures against PDF (all matched); 3 NOT FOUND rows filled from carrier email
Corrections made: Alternative B family-tier rate was read from the wrong column (p.9), corrected to $1,688.00
Reviewer: [second broker's initials]

The corrections line is the useful one. Over a quarter it tells you which kinds of pages cause errors, which is how you improve the prompt.

What to do when the carrier's PDF is a mess

Some renewal documents are scans of printed tables, or exports where columns collapse into a single line of text. In those cases extraction quality drops and the verification burden rises. Three practical responses help. First, ask the carrier or your account manager for the rate sheet as a spreadsheet; many will send one. Second, if only a few pages are poor, re-type those figures by hand and let the AI handle the rest. Third, if the whole document is poor, use the AI only for the plan-design narrative and build the rate table manually. The method is flexible; the rule that no unverified figure reaches a client is not.

It also helps to keep a record of which carriers produce clean documents. After a few cycles you will know that one carrier's tables extract cleanly and another's always need a manual pass, and you can plan the time accordingly.

Handling mid-year questions and employee queries

The renewal pack does not end the work. In the weeks after open enrolment, the same verified table becomes the reference for questions from the client and from employees. Keep it in one place and use it as the only source when drafting replies. An illustrative query: an employee asks whether the new pharmacy copay applies to a long-term prescription. The reply should quote the plan clause and its page, state what changed against the previous year, and avoid any statement about what the employee should do. If the answer is not in the documents, say so and route the question to the carrier.

A short bank of approved answers to the ten most common questions saves repeated work. Draft them with the AI from the plan summary, check each against the source, and store them with the date and the plan year, so that next year's team does not reuse an out-of-date answer. Label every answer with the plan year in its title; stale benefit answers are a quiet source of complaints, and a dated label makes them easy to retire.

A quick sum to explain the premium change to a client

Clients often ask why the headline percentage differs from what their employees will feel. Take the illustrative 12.5 per cent rise on a $48,000 monthly premium: that is $6,000 a month or $72,000 a year. Now split one tier. On the Gold PPO employee-only rate from the extract, $612.00 rising to $689.40, an employer paying 80 per cent sees its share go from $489.60 to $551.52, an extra $61.92 a month per employee, while each employee's payroll deduction goes from $122.40 to $137.88, an extra $15.48. If 50 of the 85 employees sit on that tier, the employer absorbs $3,096 a month more on that tier alone, while each of those employees notices about $15 a month. That is why the client feels the rise far more than the staff do. Ask the AI to show each step of the sum in words, then verify it in a spreadsheet. A step-by-step explanation lets a finance director follow the logic without taking your word for it.

Five renewal-table errors and where they surface

• A rate from the wrong tier. The tool lifts the two-person rate into the family column. It shows up when the totals do not tie to the carrier's summary. Fix by checking tier labels and re-running that page.
• Last year's figure. Renewal PDFs sometimes include a comparison with the expiring plan; the tool picks the wrong column. Fix by naming the columns in the prompt and the page in the output.
• A confident "NOT FOUND" replaced by a guess. Models occasionally fill a gap anyway. If a figure has no page number, treat it as invented.
• Advice sneaking into the guide. "This plan is best for families" appears. Delete, and keep the prohibition in the prompt.
• Census data left in. A name or birth date slips into an upload. Strip before the run, not after.

One page that keeps every renewal built the same way

Write the extraction prompt, the verification checklist and the slide sequence into a single page and keep it in the place your team already works. Time each stage on the first three renewals and compare against how long the work took before; the table of those times is what tells you whether to extend the approach to open enrolment communications. If you are choosing tools and costs, the year-one cost of AI for a small brokerage is a useful companion, and if the question turns to the rest of your workload, where AI saves time in a brokerage is the next read.

For help mapping which of these steps is worth automating in your own process, my AI implementation consultation covers that, and you can book a call through the contact page.

Checking it worked at the next renewal

After the first full cycle, ask four questions. Did any figure in a client-facing deck need correcting after it left? How many minutes did the comparison take against the old way? Did employees ask fewer basic questions after the plain-English guide? Did the client decide faster? If a figure was corrected after sending, tighten the verification routine before expanding. If employees still ask what a deductible is, the guide's examples were too abstract, and one worked cost scenario usually fixes it.

Questions benefits teams ask about AI renewals

Can AI read a carrier renewal PDF reliably?

It reads text-based PDFs well and scanned ones less well. Tables that span pages, footnotes and tiered rates are where errors creep in. Always have it output to a table with the page number for each figure, then check the totals against the carrier's own summary page.

Is it safe to upload client census data?

Employee-level census data is personal information. Remove names and identifiers, work with ages, tiers and counts, and use a business-grade plan with no training on your content. If your agreement with the client restricts third-party processing, check it before uploading anything.

Will the deck look generic?

It will if you ask for the whole deck at once. Ask for one slide's text at a time, tied to the client's priorities, and build the visuals in your own template. The AI supplies the narrative and the numbers table; your branding and judgement carry the rest.

How do I stop AI giving employees coverage advice?

Tell it to explain what a term means and what the plan document says, never to recommend a plan or predict claims. Add a line at the end of each guide pointing to the plan documents and your team for personal questions.

Further reads

• AI Submission Intake: Stop Re-Keying Data Into Insurer Portals — Stop re-keying census and submission data.
• How to Answer Staff HR Questions With an AI Helpdesk — Turn the benefit guide into an employee question helpdesk.
• How Insurance Brokers Use AI to Handle Claims Enquiries — Handle the claims questions that follow open enrolment.
• How to Track Contract Renewals and Deadlines With AI — Track renewal dates and notice periods.
• AI for HR in Small Businesses: 12 Tasks You Can Hand Over Safely — See what HR teams hand over to AI.
• How Small Law Firms Use AI to Draft Letters and Routine Documents — Borrow the checking habits professional firms use.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business pricing page (OpenAI), Claude Team pricing page (Anthropic), Microsoft 365 Copilot Business pricing page and Microsoft Learn's Copilot privacy page, checked November 2026. Plan figures in the examples are illustrative.