Markides Advisory
Use CaseSurrogate AI Lead Sprint
CustomerMid-Market Commercial
Insurance Brokerage (Anonymized)
Selected examples of how we partner across growth, operations, and AI.
About these examples. These examples are illustrative. Some are drawn from prior work, some combine more than one engagement, and some were created to demonstrate how the work runs. Identifying details have been removed or changed, and some of this work predates Markides Advisory. What is real is the structure, the reasoning, and the format of the deliverables. Any figures shown are directional rather than audited.
02Current situation

A well-governed brokerage with no AI pilot history at all.

Growth comes organically and through selective, competitive producer hires. Producer hiring is slow; cash position is healthy.
The business
A commercial insurance brokerage serving mid-market business clients, roughly 38–45 full-time staff. Service lines: commercial property and casualty brokerage, risk management advisory, and claims support.
Systems and data
Policy and client data sit in an agency management system, treated internally as a strong, trusted single source of truth. Carrier portals support day-to-day servicing. Board-level KPI reporting is finished by hand in Excel from a manual quarterly export.
Governance already in place
An operations committee chaired by the COO, established roughly two years earlier partly to own decisions like this one. The COO and reporting lead propose, the committee reviews, the founder signs off on larger initiatives, anchored by a monthly committee meeting and a quarterly board-style KPI review.
The constraint that shapes everything
Tool tolerance is moderate, moderated specifically by regulatory sensitivity rather than cost or change fatigue. Any new tool that touches client policy data requires compliance review before it can go live.
Markides Advisory - Use CaseCurrent situation
03The trigger

A reissued certificate, and a board deck that took days to rebuild.

The incident
A certificate of insurance had to be corrected and resent to a client's lender
A coverage detail was entered incorrectly during a routine, high-volume issuance week. It was caught and fixed quickly, but it was enough for the reporting lead to flag, at the next committee meeting, how often small errors like it were quietly being absorbed rather than tracked.
The slower strain
The quarterly board deck took several days to assemble, every cycle
Built by hand by the reporting lead from an agency-management-system export, despite the underlying data being good. As the book of business grew, that manual rebuild began competing directly with the reporting lead's other responsibilities.
The committee's stated concern
Open to AI and automation in principle, with a specific, well-founded caution. They had watched peer brokerages roll out AI tools at the producer level without coordination, creating exactly the kind of scattered tool sprawl they wanted to avoid.
Why an outside-led sprint
The COO had already begun researching AI-assisted reporting tools, but had piloted nothing, this would be the brokerage's first formal AI-related pilot of any kind. Rather than choose a tool and bring it for after-the-fact approval, the COO proposed establishing with evidence where to start, and how to sequence a first pilot inside the existing compliance process.
Markides Advisory - Use CaseCurrent situation
04Mandate and guardrails

The boundaries were set before the evidence was read.

Surrogate AI Lead Sprint, run as a founder-plus-operator intake: a founder interview plus input from the operations committee.
Objective
The fastest, cheapest, directionally reliable view of what to do first in AI, automation, and analytics, specifically identifying a defensible, compliance-compatible first pilot rather than surveying every possible use case.
Sponsor and stakeholders: the COO sponsored on behalf of the operations committee, with the reporting lead as day-to-day contact. The founder participated as CEO and retained sign-off authority for any initiative of meaningful scale, consistent with existing governance.
Explicit boundaries
Boundary 01
No AI tool may touch client policy data without compliance review before rollout.
Boundary 02
No producer-level tool rollout without ownership and guardrails first being communicated down from the operations committee.
Boundary 03
No more than one to two structured, compliance-reviewed tests running at once, given governance maturity but limited pilot experience.
Markides Advisory - Use CaseEvaluation
05Evidence base

What was reviewed, and what was deliberately not.

Collected
  • Founder interview, growth intent and sign-off expectations for larger initiatives.
  • COO interview, governance, compliance requirements, and prior research into AI-assisted reporting tools.
  • Reporting-lead interview, board-reporting mechanics, the agency management system, and day-to-day operational pain.
  • Agency-management-system export, active policies, renewal timing, producer book-of-business detail.
  • Existing quarterly KPI review deck, current board-reporting mechanics observed firsthand.
Intentionally not collected
No individual account-executive or producer-level interviews were conducted.
That gap was logged explicitly rather than assumed away, and confidence in producer-level sales-behaviour findings was lowered accordingly. The sprint treats the declared input set as the evidence base for analysis rather than assuming visibility into functions that were not actually reviewed.
Evidence thickness
Rich
Strong, consistent cross-functional visibility across leadership and operations.
Overall confidence band
Medium-high
A rich evidence band can support medium-high to high confidence; here it supported medium-high.
Where confidence is lower
Producer-level sales behaviour
Thinner and partly inferred rather than directly observed.
Markides Advisory - Use CaseEvaluation
06Evaluation

Seven move families, scored the same way every time.

Each family was evaluated independently against severity, leverage, effort, information readiness, and confidence, then compared across families for overlaps, magnifiers, conflicts, and sequencing dependencies.
Move familyWhat the evidence showedRead
Leadership and ownershipThe operations committee explicitly owns AI, analytics, and automation decisions, with a clear proposal–review–signoff pattern. Ownership has not yet been communicated down to producer teams, a governance boundary, not a governance failure.Low severity at leadership level; medium once producer activity is considered
Workflow redesignSeveral hundred active policies, renewals concentrated around anniversary dates. Certificate-of-insurance issuance is the sharpest pain: several staff-hours per week firm-wide, with occasional detail errors requiring reissuance. Renewal prep and claims tracking likely stay hybrid.Severity high · leverage high · effort medium
Reporting and KPIsMeaningful KPIs already tracked, retention rate, new business growth, book by producer, loss ratio by segment. The friction is mechanical: days to assemble the quarterly deck despite good data, and no real-time retention-risk visibility between cycles.Severity medium-high · leverage high · effort medium
AI-assisted sales and marketingProducer-driven relationship selling with limited central marketing support; a few producers already use AI informally for drafting. Producer time is the main new-business constraint. Evidence thinner here, no producer interviews.Severity medium · confidence low-medium
Data plumbingThe agency management system is a strong, trusted single source of truth. The break is narrow and well scoped: a manual quarterly export into the board deck, with no live integration today.Severity medium-high · leverage high · effort medium
Tools and experimentsNo formal AI or automation pilot of any kind. Tool overlap minimal and deliberately conservative. Appetite for structured testing is strong, but stated absorption capacity is one to two compliance-reviewed tests at a time.Capacity-constrained
Risk and guardrailsTwo guardrails governed everything: no AI tool touches client policy data without compliance review, and the committee wants to avoid producer-level tool sprawl. Spend tolerance comfortable; compliance-risk tolerance low.Binding on every move
Markides Advisory - Use CaseEvaluation
07The choice

What was deferred, and the three reasons why.

What the synthesis ruled in
Workflow redesign and data plumbing pointed at the same pattern: concrete, quantifiable operational friction sitting on top of an already-trusted data source, addressable without first fixing governance, which was already sound at the leadership level. Because of that, the highest-confidence leverage was in two operational fixes that could clear the mandatory compliance review cleanly.
Selection rule applied
The operating system defaults to two primary moves in the large majority of cases. That default applied cleanly here: two primary moves, with a third candidate documented as a next-best move rather than elevated.
Deferred, producer-level AI-assisted renewal-letter drafting
Not elevated to a primary move, despite real, already-existing informal use among a few producers.
  • Thinner evidence, no producer-level interviews were conducted, so findings here are partly inferred.
  • Governance had not reached that far, ownership of AI decisions has explicitly not yet been communicated down to producers.
  • Stated committee wariness of the producer-level tool sprawl seen at peer firms.
Sequencing this behind the two higher-confidence operational moves avoided compounding a governance gap with a new, ungoverned pilot.
Markides Advisory - Use CaseChoice
08Primary move 01

Automate certificate of insurance issuance.

The move
Build a semi-automated certificate issuance workflow to replace the current manual, admin-heavy process.
Why now
The strongest automation candidate the sprint identified. Consumes several staff-hours weekly firm-wide and generates occasional costly reissuance errors, the same pattern behind the incident that triggered the engagement.
Current state
A dedicated admin function manually issues certificates, with occasional detail errors requiring reissuance.
Proposed state
Semi-automated issuance pulling directly from agency-management-system policy data, with admin staff reviewing exceptions.
Suggested test
Pilot on one line of business's certificates for four to six weeks, tracking error and reissuance rate against the current baseline, with operations-committee sign-off, the brokerage's first formal AI-related pilot.
Leverage
High
Effort
Medium
Directional value
Medium
Time to scope
Three to four weeks, including compliance review
Time to traction
Two to three months
Confidence
High
Ownership
Reporting lead day to day; COO as executive sponsor; committee and compliance sign-off before rollout
Startability
Mixed internal and external support, given the compliance review requirement
Tool families
Workflow and orchestration; integration, sync, and data plumbing
Evidence for
Explicitly named as a strong automation candidate; the agency management system is a trusted single source of truth for the underlying policy data.
Caution
This would be the brokerage's first formal AI or automation pilot, execution discipline (defined owner, goals, stop-go rules) matters more than usual, and compliance review is mandatory before rollout.
Key missing data
No detailed breakdown yet of which specific certificate errors are most common, which would sharpen where automation adds most value.
Markides Advisory - Use CaseAction sequence
09Primary move 02

Automate the reporting link, and add a retention-risk flag.

The move
Replace the manual quarterly export from the agency management system into the board deck with a more automated pull, and add a simple retention-risk flag ahead of renewal dates.
Why now
The quarterly board deck takes the reporting lead several days to assemble despite good underlying data, and leadership explicitly lacks real-time retention-risk visibility between reporting cycles.
Current state
The reporting lead manually rebuilds the board deck each quarter from an agency-management-system export.
Proposed state
A more automated data pull plus a simple renewal-date-based retention-risk flag, reviewed monthly by the operations committee.
Suggested test
Run the automated pull and retention flag alongside the manual deck for one quarter to validate accuracy before replacing the manual process.
Leverage
High
Effort
Medium
Directional value
Medium-high
Time to scope
Three to five weeks
Time to traction
One to two quarterly cycles
Confidence
Medium-high
Ownership
Reporting lead owns; COO sponsors and ensures compliance sign-off
Startability
Mixed internal and external support; possible light specialist help for the automated pull
Tool families
Integration, sync, and data plumbing; analytics, dashboards, and reporting
Evidence for
The agency management system is explicitly described as a trusted source of truth; the reporting gap is a well-defined, addressable manual break rather than a data-quality problem.
Caution
Any change touching policy data flows must stay within the compliance review process the brokerage has already committed to for AI-related pilots.
Key missing data
No detail yet on the agency management system's export or API capabilities, which will shape how automated this can become.
Markides Advisory - Use CaseAction sequence
10Action sequence

Prove the process on trusted data first then extend governance.

Two moves start now; one governance step unlocks the third. Deliverables at close: a slide deck for the committee and founder, a one-to-two-page written summary built to move through compliance review, and a roadmap.
Now, concurrent
Certificate issuance automation
Scope in 3–4 weeks including compliance review. Pilot one line of business for 4–6 weeks against the current error and reissuance baseline.
→
Now, concurrent
Reporting link and retention flag
Scope in 3–5 weeks. Run alongside the manual deck for one quarter to validate accuracy before replacing it.
→
Precondition, then next
Extend AI governance to producer teams
Communicate ownership and guardrails down from the committee, the explicit precondition for the deferred move.
→
Deferred
Producer renewal-letter drafting
A compliance-safe template and review step, formalizing informal use, once governance reaches producers and a first pilot is complete.
Why this order
Both first moves lean on agency-management-system data the brokerage already trusts, and both are compatible with the mandatory compliance review process. That is what makes them defensible as a first formal pilot in a regulated environment.
What is deliberately not claimed
The directional impact expected is a reduction in certificate errors and reporting assembly time, plus earlier retention-risk visibility. No precise dollar or retention-rate impact is claimed, the evidence does not support that level of precision.
Markides Advisory - Use CaseAction sequence
11Outcomes

Two compliance-cleared pilots, and a governance gap that closed itself.

Move 01, certificate issuance
Piloted on one line of business for six weeks
Cleared compliance review before launch and ran with no reissuance-triggering errors across the pilot window, enough of a signal for the committee to approve extending it to the brokerage's other lines of business.
Move 02, reporting and retention
Ran alongside the manual deck for one full quarter
Reconciled cleanly against it, at which point the reporting lead retired the manual rebuild. The retention flag surfaced one at-risk renewal early enough for a producer to intervene before the policy anniversary, cited directly by the committee when reviewing results.
What happened next
Governance extended to producer teams
With its first two formal pilots complete, the committee began communicating AI ownership and guardrails down to producers, the explicit precondition the sprint had identified. The brokerage's first formal AI use policy was drafted shortly after.
Why this work is credibility-relevant
The engagement required operating credibly inside a regulatory constraint that shaped every recommendation, rather than treating compliance as a footnote added after the fact, a materially different kind of judgement than an unregulated commercial sprint requires.
What it demonstrates
Sequencing a first AI pilot responsibly in a business with zero prior automation experience and real regulatory exposure: choosing moves that could clear compliance review cleanly, and explicitly declining a plausible producer-facing move until the governance precondition for it existed.
Markides Advisory - Use CaseOutcomes
Markides Advisory
Book a discovery call to learn more.
markidesadvisory.com•michael@markidesadvisory.com
Book a Call