A CFO's first thirty days
Did the money make it to the bank?
A new CFO inherited 47 stores, three payment rails, and an accounting team answering that question by hand every single morning. This is what happened in their first thirty days, and what it would take to do the same for your team.
The short version
A new CFO at a 47 store Southern California automotive service group brought in Michael Bowers of Heed AI Solutions within their first ten days on the job. Kickoff with the accounting team was July 1, 2026. Working, secured software was live on July 2, one day later. More than 40 numbered releases shipped over the following two weeks, built session by session with the two accounting veterans who knew the rules. On July 15 the team worked the tool all day and unreconciled deposits fell from more than $1,000,000 to $435,000. The engine now matches cash, card settlements, and consumer financing deposits to the bank every morning, explains every exception in plain English, and escalates anything unresolved on a schedule leadership set.
Chapter one
Why can nobody answer this question fast?
Every multi location business lives with one deceptively simple question: did every store's money actually make it to the bank? For this company, the answer took a team of people most of every day, every day.
Cash from dozens of stores deposited through ATMs and bank branches. Card batches that bundle Friday and Saturday into Monday, split at a nightly cutoff, and sometimes fund a day late. A consumer financing program that pays the whole company in one lump sum, net of fees that vary by promotion. Three payment rails, three different clocks, none of them lining up with a calendar day.
Underneath it, a point of sale from another era feeding an accounting server that predates the cloud. A separate system for wholesale. Monthly exports to yet another system for reporting. None of it connected.
And one detail that says everything about manual work in 2026: the bank's downloadable file drops the one identifier that ties a cash deposit to a store. The identifier is visible on the bank's website. So every day, someone copied and pasted screens from the bank website into spreadsheets, store by store.
It's just copy and paste. Yeah.
A member of the accounting team, describing the daily processChapter two
The real cost was never the hours
Before any tool existed, the manual process was already collecting its real price, and it was not measured in hours. It was measured in one person.
The team member who carried the daily cash reconciliation produced those numbers by hand every single morning, no matter what. Every vacation needed a coverage plan. Every sick day created a backlog. Every month end stacked a thirty tab workbook on top of the daily grind.
Workload like that has fallout: fatigue that compounds, morale that erodes, and a single point of failure quietly holding knowledge no one else has. This company had already watched twelve years of institutional knowledge walk out the door when its longtime controller departed.
That is the quiet math of manual accounting work. It does not just consume hours. It wears down the people who hold everything together, and sooner or later it puts them at risk of walking too.
There's some obvious frustrations there with the team, just looking for some kind of relief.
The incoming controller, on why Heed was brought inWhat happened in thirty days
No requirements document. No steering committee. The CFO picked the first target deliberately: daily deposit reconciliation, because it touched money, people, and every location, every single day.
The new CFO's first day.
Within their first ten days, approval to bring in Heed. First target chosen.
Kickoff working session with the accounting team. The rules of the business live in a few veterans' heads and a thirty tab workbook.
The reconciliation portal is live and secured. The team logs in. Not a mockup, not a proof of concept deck. One day after kickoff, and a day ahead of the promised date.
A morning walkthrough produces five requests. All five are live by that evening. That same night an upload mystery is root caused remotely: the bank file had arrived without its header row. Instead of asking the team to change, the system learns to read it anyway.
A 10:30 call about upload errors becomes funding day rules, next day netting, write off workflows, and an unpair button. Live that afternoon.
The team works the tool all day. Unreconciled deposits fall from more than $1,000,000 to $435,000. Twelve releases ship that same day.
A Monday upload surge exposes a bottleneck. Fixed by that afternoon: uploads confirm in about two seconds. The CFO walks into their Monday leadership meeting with a live, current dashboard.
Chapter three
The back and forth was the method
This was not a build handed over a wall. It was a conversation with software in the middle.
Two veterans of the accounting team, one with more than a decade at the company, were instrumental from the first session. The rules of the business lived in their heads, and session by session, screenshot by screenshot, those rules became the engine.
Cash deposits get a grace amount and a realistic window of banking days, because stores are human. Friday and Saturday card sales bundle into Monday. A batch that misses the nightly cutoff lands a day later. One location runs on its own bank account with its own rules. The financing program funds the whole company in one deposit, net of fees, so the engine reverse engineers every store's share and the fee math down to the penny.
When the CFO reviewed manual matching, their instinct became a design rule: if a link is off by even a few dollars, the system should say so and ask before proceeding. More than 40 numbered releases shipped in two weeks. The feedback loop was hours, not sprints.
It has to be accurate, never a guess.
The senior accountant, whose rule became a design law of the productWho you actually work with
Not an account manager. Not a delivery team you never meet. You work directly with Michael Bowers, and that is the part that makes the rest of it possible.
The CFO in this story did not find Heed through a search. They have brought Michael into almost every company they have joined over the last ten years, across a bicycle manufacturer, an avocado grower and packer, and an industrial laundry equipment manufacturer. Different industries, different systems, the same reason every time: the thing that is quietly costing the most is almost never the thing anyone has written a requirements document about.
That is the part most software engagements get backwards. A vendor arrives with a product and asks where it fits. Michael arrives and asks what your morning actually looks like, then listens for the answer underneath the answer.
You can hear it in the recordings from this engagement. On the kickoff call, a member of the accounting team described a process the whole company had accepted as normal in five words: it's just copy and paste. Nobody there had ever framed it as a problem worth solving, because it was simply how the work got done. Naming it out loud was the first thing that changed.
A few calls later, Michael said the goal was to give the team some relief internally, and the controller finished the sentence: right, so we can just breathe. That is the moment a project turns. The client stops being a buyer evaluating a tool and becomes a co author describing what they actually need. Everything after that ran faster, because they were building it too.
Getting to that moment is the skill. It takes showing people what is possible before they can picture it, so that the conversation moves from what we have always done to what we could have by Friday. Then it takes handing the design back to them, because the person who has closed the month forty times knows things no discovery document will ever capture.
Years of sitting with CFOs, controllers, and owners is what makes that fast instead of slow. Knowing that a deposit window needs a grace amount because stores are run by people. Knowing that a senior accountant saying it has to be accurate, never a guess is not a preference to note down, it is the constraint the entire system has to be built around. Knowing that when a bank file shows up without its header row, the right move is to teach the software to read it, not to hand the team one more rule to remember on a Monday.
The nuances are the job. A generic product cannot hold them, and a team that has never carried a close cannot hear them. That is the whole difference between software that gets installed and software a team actually opens on a Monday morning.
What a CFO actually gets
Two things you cannot buy off a shelf, because they only exist once the rules of your business are written down: control over the process, and insight into what it is telling you.
Control
Nothing moves without a name on it
- Never a guess. The matching is exact math against written rules, not a model's opinion. When a shortage cannot be explained with certainty, the tool says exactly that and hands it to a human.
- A full audit trail. Link, offset, write off, unpair, correct a source number. Every action records who did it and when, and every action can be undone.
- An escalation ladder you set. Items that sit unresolved move up automatically: to the team at day three, to leadership at day five.
- One list for the top. The company president sees exactly one thing: the short list of stores where money is genuinely missing.
- Uploads that police themselves. The system tracks what should arrive each day, flags gaps by store and day, warns when data goes stale, and refuses to let a stale re upload silently overwrite corrected numbers.
- Month end becomes a report. Write offs print as a clean, reviewable document instead of archaeology through a thirty tab workbook.
Insight
Answers, not more rows to read
- Exceptions that explain themselves. Every row says in plain English what was matched, what was ignored, and why. No silent failures, ever.
- Shortages pinned to a cause. A shortage gets tied automatically to the single invoice and customer that created it, so the store makes one phone call instead of opening an investigation.
- Mis keys caught on sight. A sale rung on the wrong payment type gets flagged with the matching offset already suggested.
- Money recovered from your processors. Two real funding shortfalls, $607.58 and $862.19, were caught to the penny and documented to take back to the processor.
- A look back you have never had. The first historical run flagged more than $100,000 across two months of recorded deposits for investigation, each one traceable to a cause: timing, a mis key, a broken deposit card, or a genuine follow up.
- Ask it in plain language. Which locations are late to deposit. What is still open from last Tuesday. The assistant answers from your data, and the judgment stays with you.
The numbers, dated and real
Everything below happened between July 1 and July 20, 2026. Nothing here is a projection.
Chapter four
So we can just breathe
On an early call, Michael described the goal as giving the team some relief internally. The controller answered before the sentence was finished.
That is the result that matters. The backlog is measured and shrinking. The morning question has an answer by the second cup of coffee. The veterans who used to spend their days copying and pasting now spend them resolving the short list of real exceptions, with the tool doing the hunting.
New stores can be added without adding headcount to count the money. The knowledge that lived in two people's heads now lives in an engine the whole company shares, with the humans still firmly in charge of every decision.
Right. So we can just breathe.
The controller, finishing the sentenceQuestions every CFO asks first
These come up on every call, so here are the answers in advance.
Does my financial data train an AI model?
No. Your data lives in a private, login gated environment and connects to the model by secured API only. Anthropic does not train on this data by default, and we lock it down further from there.
What data does the engine actually touch?
Transaction amounts and dates. Never cardholder data, never customer PII. That constraint comes from a compliance background, and it is designed in from the first day rather than bolted on.
Are the numbers AI guesses?
Never. The matching is deterministic: exact math against rules your team wrote. AI is why this got built in days instead of quarters, and it powers the assistant that answers questions in plain language. The arithmetic is not up for interpretation. When a shortage cannot be explained with certainty, the tool says so and hands it to a human.
Do we have to replace our point of sale or accounting system?
No. The engine sits on top of the systems you already own, including a point of sale from another era and an accounting server that predates the cloud. Nothing gets ripped out and nothing gets installed on your machines. It runs behind zero trust access, which suits a company with no in house IT.
How long does this take to build?
In the engagement above, working secured software was live one day after kickoff, and more than 40 numbered releases shipped over the following two weeks. For a new client, plan on two to four weeks after the proof of concept, depending on how many source systems are involved.
Who actually does the work?
Michael Bowers works directly with your team. The CFO in this case study has brought Michael into four companies over ten years. The build happens alongside the people who carry the work rather than being handed off to a delivery team you never meet.
What does it cost?
A proof of concept on your real data starts at $1,250 and credits toward the build. A managed build starts at $5,000 plus $450 per month, and outright ownership of the code starts at $10,000. Where you land inside that depends on scope, which is broken down below.
How this starts
The same way it started for the CFO above. One narrow workflow, proven on your real data, before anyone signs up for a build.
Start here
Proof of Concept
Starting at $1,250
One workflow, your real data
Built on your own stores, your own deposits, and your own gaps. Not a demo dataset.
If it does not deliver, you get it back. If it does, it credits toward the build.
Recommended
As a Service
Starting at $5,000 to build,
then $450 per month
We host it, your team uses it
Everyone on your accounting team can use it. Hosting, support, updates, and reconciliation help included.
You own your data and your IP. This is the option that keeps your upfront cost down.
Alternative
Own it outright
Starting at $10,000, one time
The code is yours to keep
You own all the code and can hand it to anyone, at any time.
Support still included.
Why every number says starting at
Because no two of these builds are the same, and quoting a flat price before seeing your operation would either overcharge you or set a number neither of us could keep. Five things move the scope:
- Vendors and source systems
- One bank and one processor is a very different build from four processors, two banks, a consumer financing program, and a separate wholesale system.
- Technology and access
- A modern API is straightforward. A server from another decade with no export scheduler, or a point of sale that only writes files to a share, takes longer to reach safely.
- Data condition and cleanup
- If the history is clean, we move fast. If it has to be reconstructed and reconciled before anyone can trust a number, that is real work and it comes first.
- Documentation
- Some companies have their rules written down. Most keep them in two people's heads, and getting them out, in writing, is part of the engagement and part of what you keep.
- Ongoing data management
- How much arrives automatically each morning versus how much someone still has to export and hand over, and what it takes to close that gap.
You get a fixed number before any build starts. The proof of concept is what makes that number honest, because by then we have both looked at your actual data.
If someone on your team answers this by hand, that is not a cost of doing business.
It is a solvable problem, whether you run two locations or two hundred. The conversation is free, and you will know inside thirty minutes whether there is something here worth building.
Heed AI Solutions is the AI practice of Heed Business Solutions: business, technology, and people, connected.