Operations Automation · Insurance Agencies · Wholesale Distribution

The work your team
does by hand
should be software.

Most of it already could be. We find it by timing it — not estimating it — then build it, and wire the decisions that need judgement to a model score instead of a rule somebody guessed once.

6 Workflows timed, not estimated
1 wk Kickoff to teardown in hand
1:1 No juniors, no handoffs

Is this
right for you?

We work in two industries, because the second engagement in an industry costs a third of the first — and that discount belongs to you, not us. You have a system of record that holds your data, and the actual work happens somewhere else: in an export, a spreadsheet, and a person who checks.

  • 10–100 staff, commercial lines, running Applied Epic, AMS360, EZLynx or HawkSoft — plus a dozen carrier portals, email, and a shared drive
  • Certificates and endorsements are still keyed by hand, and a CSR spends a quarter of the week on them
  • Renewal outreach runs on a calendar rule — 90 days before expiry, the same touch for every account regardless of who is actually at risk
  • $5–75M revenue on NetSuite, Epicor, Acumatica or SAP Business One — with orders arriving as email bodies, PDFs, spreadsheets and texts
  • Inside sales re-keys those orders against an item master that only one person fully understands
  • Reorder points were set years ago on an assumption, and nobody has revisited them since

You are probably not a fit if one vendor already owns the whole workflow end to end. If your platform ships what you need as a feature, buy the feature — we will tell you so on the call.

Three rungs. Nobody buys the third one first.

Fixed scope and a fixed price at every rung. We don't sell hours, and we don't scope on a call — the first rung exists to make the second one accurate.

01 One-Time

Ops Teardown

One week. We sit with your team and time six recurring workflows with a stopwatch — frequency times observed duration times your loaded hourly cost. Not a survey, not an estimate, not a maturity score. You get the six ranked by annual recoverable spend against build difficulty, and a build plan for the top three. The fee comes off the build if you book one within 30 days, and the document is yours either way.

  • Six workflows timed and costed in labour dollars
  • Ranked by annual recoverable spend vs. build difficulty
  • Build plan for the top three, with what we'd refuse and why
  • Baselines captured before any work starts
03 Retainer

Run

Optional, and we would rather you decided at day 60 than sign for it now. Monitoring, breakage repair, quarterly model retraining, and one new or extended workflow every month. The thing that actually earns it is the monthly one-pager: what ran, what broke, what it saved, what's next — because an automation that has worked silently for four months is one nobody remembers paying for.

  • Row-count and freshness checks, so failures are loud
  • Quarterly model retrain and threshold review
  • One new or extended workflow per month
  • Monthly one-page report. Three-month minimum

A score decides the action.
Not a rule someone guessed.

Automation is mostly conditional plumbing — if this, send that. It works fine for mechanical work and badly for anything that depends on which record it is. Here is what the alternative looks like, running.

Model-Scored Routing — the Score Selects the Action

● Live

A working scoring engine that does not stop at a probability. Every record is ranked, binned into a named action tier — Priority, High Reach, Suppress — and the tier is the field the downstream workflow actually reads. The same pattern that decides which account a service team calls first, or which order gets held for review.

  • The top-ranked 10% of the file converts at 2.95× the population average — so the tier, not a recency filter, decides who gets worked and who gets skipped
  • Every score is attributed to three underlying signals, so the person acting on it can see why a record ranked where it did. A score nobody trusts is a score nobody uses
  • Paired suppression logic handles the other half of the decision — who should never be actioned at all, which is the part rule-based automation almost always gets wrong
Python Scikit-learn Pandas SQL Streamlit

Enterprise rigor.
Boutique attention.

The measurement discipline of a $B+ operation, delivered with the direct access and senior focus you'll never get from a high-volume agency.

3+ Years as dedicated analytics consultant inside a multi-billion-dollar operation
$B+ In revenue informed by models and measurement built as the sole technical point of contact
1:1 Direct partnership — no shared queues, no junior analysts, no hand-offs

Most people who build automations cannot build the model. Most people who build models will not do the plumbing. The work that matters here needs both — and it is the reason this page has real numbers on it instead of ranges.

Data.
Decisions.
Dollars.

Data
We time the work with a stopwatch. We don't estimate it, and we don't ask your team to remember how long something takes.
Decisions
Where the work needs judgement, a model score decides which path fires — not a threshold somebody picked once and never revisited.
Dollars
Every workflow carries hours per month and a dollar figure attached — so you know what a build is worth before anyone builds it.
See a score select the action →

Timed, Not Estimated

We sit with the people doing the work and time six recurring workflows — frequency, duration, loaded cost. Nine competitors sell a diagnostic; every one of them delivers a workflow inventory built from estimates. So what: You get the hours it actually takes, not what somebody remembers, which is the only way the ranking underneath it means anything.

Scored, Not Ruled

The decisions that depend on which record it is get a real probability, fitted to your own history and written back into the system your team already uses. No new tool, no migration, nothing to retrain anyone on. So what: The automation keeps working when the exception rate moves, because it was never pinned to a number somebody guessed in 2019.

Measured Against a Baseline

Baselines are captured during the teardown, before a line of anything is built, and the model-scored build ships with a held-out control group. So what: The result can come back negative — and if it does, you'll hear it from us, which is the only reason to believe the ones that come back positive.

About

A focused solo practice — not an agency juggling 30 accounts. Every engagement gets direct senior attention. No junior analysts. No hand-offs. The same person who scopes the work does the work.

Brendan Hoffman

Brendan Hoffman

Founder & Lead Analyst

Three years as an analytics consultant at a boutique practice, as the sole technical point of contact for a multi-billion-dollar e-commerce operation — building the models, the pipelines and the measurement behind decisions at that scale. I run automation pipelines in production that score records against a versioned rubric and route on the result, unattended, on a schedule. That combination is the whole practice: most people who build automations cannot build the model, and most people who build models will not do the plumbing. University of Maryland, Business Analytics & Information Systems.

Process Mapping Workflow Automation Model Scoring Python SQL API Integration Data Pipelines Executive Communication

Tell us what your
team does by hand

One sentence is enough to start — the thing someone on your team exports, re-keys, or checks every week. We'll tell you whether it's worth a teardown, and we'll tell you if it isn't.

For implementation partners: if you consult on an agency management system or an ERP, we run the Ops Teardown under your brand — you own the client and you bill them. Say so in the message and we'll send the partner terms.