Process Optimization

Business Process Optimization Services

Before you make anything faster, it is worth knowing how much of it should exist. Business process optimization is the work of finding where your operation loses time to steps nobody has examined in years — and the highest-yield finding is almost never "this could run quicker." It is "this could stop." Facet does that work with evidence rather than opinion, and we know what it turns up because we ran it on ourselves first: the act of specifying our own processes well enough to hand them to agents is what exposed which of them had no defensible purpose.

Specifying work for an agent is an audit you cannot fake

Facet's research, quality review, reporting and content delivery run through agent loops under human command. The first effect of that move was not speed. It was embarrassment.

You cannot hand a step to an agent without writing down what it is for, what triggers it, and what a correct output looks like. That is a low bar, and several of our own steps could not clear it. The categories will be familiar:

A recurring report with no nameable reader. Produced faithfully, on time, for years, by people who assumed somebody downstream needed it.
A review step that had never sent anything back. Not rarely. Never — which makes it a delay with a job title.
A required field that was required because it had always been required, and whose original purpose had left the company with the person who introduced it.
The same work happening twice in two places, under two names, because two people solved the same problem in the same month and neither knew.

None of those were dramatic discoveries. They were obvious ones, and that is the actual finding: they had survived years of intelligent people paying attention, because nothing in a normal working week ever asks why does this exist. Specifying the process for an agent asks it, unavoidably, on every step.

That is what we bring to your operation. Not a methodology we intend to try out on you — a pass we have already run against our own work, where being wrong cost us rather than a client.

Humans stay in command throughout. Agents produce the evidence; people make every call about what changes. The reason for that split is the next section, and it is the whole argument of this page.

Optimization has two moves. The industry sells one of them.

You can make a step faster, or you can remove it. That is the complete list.

Almost every optimization engagement on the market sells the first move, and not because anyone is being dishonest. Acceleration is simply the better product: it is safe, it is measurable, it produces a diagram, it can be delivered by people who do not work at your company, and nobody's job gets smaller. Subtraction is none of those things.

But automating a step you should have deleted is the most expensive outcome available to you — more expensive than leaving it alone. A slow, annoying, manual step is under permanent review by the people who have to do it; every one of them resents it, and resentment is a surprisingly reliable audit function. Automate it and the resentment disappears. The step becomes cheap, invisible, and permanent. It will still be there in ten years, generating a report with no readers, now with a monitoring line item and a maintenance budget of its own.

So the order of operations matters more than the tooling: subtract first, then automate what survives.

And subtraction is the one move an agent structurally cannot make. An agent can tell you a step adds four days of elapsed time and has not changed an outcome in two hundred instances. It cannot tell you the step exists because of a regulator, because one customer's contract requires it, or because of a bad afternoon in a year nobody wants to relitigate. More to the point: deleting a step makes somebody's job smaller, and that consequence lands on a person who has a name and a mortgage. Subtraction has consequences an agent does not carry — which is precisely why the evidence can be automated and the decision cannot.

The criterion that keeps this from becoming a cull: a step earns its place when removing it would cost more than keeping it. Some steps that look useless are load-bearing — they are insurance, or an audit trail, or the reason a regulator has never asked twice. Those survive the test easily and we will argue for keeping them. What changes is where the burden of proof sits. It belongs on the step, not on the person asking about it.

A related but separate question is where your human approval gates should sit inside a process you have already decided to keep, and what those gates cost you in elapsed time. That one is answered on our business process automation page. This page is about the prior question: whether the step should be there at all.

What the loop actually does

The evidence behind a subtraction decision is high-volume, mechanically checkable, and almost always uneventful — which is exactly why it has never been gathered by hand.

Time every step from timestamps, not from the interview. An interview returns the process as people understand it, which is reliably the tidier version. Agents read tickets, email headers, system logs and file histories, and return elapsed time per step from what actually happened — including the steps nobody mentioned because they do not think of them as steps.
Count how often each step changes the outcome. Over the last two hundred instances, how many times did this approval reject something? How many times did this review send something back? How many times did this field's value differ from the default? A person cannot count that. An agent counts all of it, and the counts are frequently the most uncomfortable page in the report.
Find the duplicates across teams. The same work done twice under two names in two departments is visible only to something that reads every process at once, which is not a thing anybody does.
Write down the process as it actually runs, against the process as documented, and mark every divergence. The gap between those two documents is usually the most useful artifact of the whole engagement.

What a person still decides

Everything above is evidence. None of it is a decision, and the engagement states that boundary in writing rather than letting it blur.

Whether a step that never changes the outcome is load-bearing anyway. Insurance, audit trail, a regulator's expectation, one client's contract. None of that is visible in the data — it lives in agreements and in people's memories.
Whose job gets smaller, and how that is handled. This is the real reason subtraction stalls in most companies, and it is a leadership act affecting named people. It must never be delivered as a technical finding.
Which of two duplicated processes survives. That is a decision between two teams, not between two workflows, and the better workflow does not always win it.
Whether the process should exist at all. The largest version of the question, and the one only an owner can authorize.
What the returned capacity is for. Removing work does not create value by itself. Pointing the freed hours at something is a separate decision, and it is yours.

Show me the math

These figures are illustrative. The value is in the method, which you should re-run on your own numbers.

(Illustrative) Take a weekly report. Five people contribute to it, forty minutes each; one person compiles it over two hours. That is roughly 5.3 hours a week — about 275 hours a year, or near $20,000 of loaded labor at $75 an hour, for one recurring artifact nobody has questioned since it was introduced.

Now ask the subtraction question, which is not how do we produce this faster. It is who opened it. Check the distribution list against the access logs. Suppose the answer is two people, and one of them only ever reads the summary line.

You now have three options, and the difference between them is the entire value of this service:

(Illustrative, second case) A three-gate approval chain running two hundred times a quarter. Read the history: gate one rejected fourteen, gate three rejected two, gate two rejected nothing at all. Gate two is not a control. It is a delay with a job title, and the evidence for that is countable rather than arguable — unless there is a reason outside the data, which is exactly the conversation the report is designed to start.

That is the method: count the instances, count the times the step changed the outcome, price the difference, and then have the conversation the numbers make unavoidable. It is what we do during the process audit, and it works on your data whether or not you ever hire us.

How could AI agents run your process audit? Book a 60-minute Process Audit — one real process, its history read before the call rather than reconstructed during it. You leave with elapsed time per step, a count of how often each step changed an outcome, a shortlist of candidates for deletion, and a written recommendation. If the read shows your process is tight and the steps are earning their place, we will tell you there is nothing here worth buying — that is a real outcome, it happens, and it costs us the engagement.

Where the removable work usually is

Business process optimization consulting at Facet covers four operational areas. The bullets are where we look first, because this is where undefended steps accumulate.

Sales and revenue operations

Approval chains on discounts and proposals that have never rejected one
Lead routing rules written for a territory model you no longer run
CRM fields required at stages where nobody has the information yet
Handoff meetings that exist because an integration does not

Marketing operations

Reporting produced on a cadence nobody chose, for a distribution list nobody curates
Review rounds that reliably change formatting and nothing else
Content workflows with two approval layers for work that publishes to twelve readers
Attribution reporting collected but never used in a decision

Internal operations

Onboarding and offboarding checklists carrying items for systems that were retired
Procurement steps scaled for a purchase size you rarely make
Document routing where the routing takes longer than the reading
Compliance checkpoints duplicating a control that already exists upstream

Technology and data operations

Alerts nobody actions, which is the same as alerts nobody receives
Manual reconciliation between two systems that should not both be authoritative
Incident escalation paths with a tier that has never been reached
Scheduled jobs still producing outputs for a dashboard that was decommissioned

How the engagement runs

1. Process audit

A structured read of one real process: where work enters, how it moves, where it queues, and — the part that is usually new — how often each step changed the result. The evidence pass is agent work; the session is a working conversation, not a questionnaire. You leave with a prioritized map ranked by what each step buys against what it costs.

2. The subtraction decisions

We bring the candidates and the evidence. You make the calls, with the people who carry the consequences in the room, because they will know the two reasons the data cannot see. This is the meeting the engagement exists for, and it is deliberately uncomfortable.

3. Redesign and implementation

What survives gets rebuilt properly, and only then does automation enter the conversation. Facet works across Microsoft 365, Google Workspace, AWS, Azure, Cloudflare and the major CRM and marketing platforms. No rip and replace — we optimize what you have and add automation where the leverage is real.

4. Measurement

Each retained process is instrumented: throughput, exception rate, time to completion. We review performance with you on a cadence and re-run the subtraction question against fresh evidence, because steps accumulate again. They always do, and not through carelessness: a step gets added every time somebody solves a real problem and leaves the fix in place. Nobody ever schedules the meeting to take one back out. That meeting, on a cadence, against evidence, is what you are buying — the subtraction is not a thing we do once and hand over.

What to expect

You will not get a deck and a handshake. You get evidence, a documented decision log naming what was removed and who agreed, and a measurement framework you keep.

You do not need to pause operations. Nothing in this work requires stopping. Deletion in particular is the only change on the list that can be tried and reversed inside a fortnight.

You do not need to replace your stack. New tooling enters only when the case is clear and you have agreed to it.

We will argue for keeping things. If a step is quietly protecting you, we say so — even where removing it would make our before-and-after look considerably better.

Why Facet

Founded 2014 — a boutique systems integrator, not a generalist agency
CIOReview Top 10 Digital Transformation Services Company (2021)
25 verified Clutch reviews
40 team members across 9 timezones and 6 continents
500+ specialist network — open-source web, CRM, data and IT talent on demand
Projects from $6,000 — structured to serve growth-stage companies, not only enterprise

We have run this pass across professional services, property management, recruiting, B2B wholesale and healthcare administration, which is where the pattern recognition comes from: the undefended step takes a different form in each industry and appears in all of them.

Where process optimization goes next

This is a ladder, not a bundle, and an operation with twenty percent of its steps removed and nothing automated is a complete, cheap and durable outcome.

Workflow Design & Implementation — for a surviving process worth running as a designed loop, including what happens when reality deviates from the drawing.
Business Process Automation — the build and run of the loops themselves, and where the human approval gates belong inside them.
Agentic Engineering — the build layer, when a process needs real software behind it rather than configuration.
AI Managed Services — the run layer, when you want a named Facet team accountable for keeping it working.
Agentic Operating System — the same discipline installed across every operating area rather than one process at a time.

FAQ

How much does business process optimization cost?

Engagements run from project-based work starting around $6,000 to ongoing operating-partner retainers for teams who want continuous review. The right structure depends on how many processes are in scope and how much ongoing measurement you want. The process audit clarifies scope before anything is committed.

How long does it take to see results?

Subtraction is the fastest result available in operations: a step removed this week returns its hours next week, with no build, no integration and no procurement. Redesign and automation of what survives typically starts producing measurable change four to six weeks after the audit, depending on integration complexity.

How is this different from an RPA or automation vendor?

An automation vendor starts from the question what can we automate. We start from what should exist, which is a question with a cheaper answer. RPA and agent loops both automate the steps you keep — and we build those too — but running that build before the subtraction pass is how businesses end up paying to maintain work they should have stopped doing.

Do you work with our existing tools, or will you require a platform switch?

We work within your existing stack by default, across Microsoft 365, Google Workspace, AWS, Azure, Cloudflare, the major CRM platforms and most marketing automation tools. New tooling enters only when the current stack genuinely cannot support something you have decided to keep.

What if we delete something we needed?

That is the right worry, and it is why deletion is tried rather than declared. A removed step is restored in an afternoon, and three weeks is usually enough to find out whether anyone depended on it. Compare that to the reversal cost of an automation build, and the risk profile is the opposite of what intuition suggests. The steps we will not experiment on are the ones tied to a regulator, a contract or an audit trail — and identifying those is what the human half of the engagement is for.

Who needs to be in the room?

Whoever can say "we are not doing that anymore" without checking with anyone. If that person is not in the session, the session produces a list instead of a decision.

Deletion is the only change that has no run cost

Every other improvement you can buy creates something. A loop has to be monitored. A tool has to be paid for, patched and explained to the next person who joins. An integration breaks when something upstream changes. A hire needs managing. All of them are worth it when the case holds — and every one of them shows up next quarter asking for a share of somebody's attention.

A deleted step creates nothing. It does not drift, it cannot break, it will not need a licence renewal, and it never appears on a roadmap again. It is the only line on any improvement list that is finished the day it is done.

It is also the only one you can authorize on a Tuesday. Nothing needs procuring, nothing needs integrating, and the capacity arrives the same week rather than the following quarter. The reason it does not happen more often is not cost, difficulty or technology — it is that exactly one person in a company has the standing to say we are not doing that anymore, and in a business your size that person is almost certainly you. The evidence to make that call confidently takes an afternoon to gather. What it is worth is the rest of the year.

How could AI agents run your process audit? Book a 60-minute Process Audit — one real process, its history read in advance, every step counted against what it actually changed, and a written recommendation including "leave this alone" where that is the honest answer.

Book a Process Audit or reach us directly: [email protected] — (323) 510-1984