Business workflow automation: how Fornida cut month-end reconciliation from 5 days to a few hours

At month end, Fornida's accounting team was taking company credit-card statements, matching transactions against purchase orders and sales orders, then manually pushing each line into the right GL bucket. It was repetitive, slow, and easy to get wrong. Across five credit cards, the process could eat twenty to twenty-five hours a month. Today that same workflow runs through an internal automation. Export the transactions, drop the file into a folder, let the system classify the bulk of the entries, then have a human review the anomalies. The job that used to take four or five days now takes a few hours.

This is what business workflow automation is supposed to look like in a real SMB. Not a demo. Not a chatbot. A finance workflow that got materially faster without losing human judgment.

The 60-second version

Client: Fornida, internally. The accounting team was reconciling company credit-card activity by hand at month end.
The old workflow: Five different credit cards, often hundreds of transactions, each line manually classified into the correct GL account through spreadsheets and system entry.
The problem: Too much repetitive data entry, too much review time, and too many small manual errors compounding over the month.
What got built: A custom reconciliation workflow that ingests exported transaction files, maps transactions to the right accounting buckets, and flags the uncertain entries for human review.
The outcome: What used to take four or five days now takes a few hours, with the reviewer focused on anomalies instead of typing every line item.

The old process

Carlos described the original work plainly. A monthly credit-card statement was not one or two pages of simple expenses. It was a business statement, often with a hundred or more transactions, spread across categories that all needed to land in the right place.

This is a business account where you can have, we usually have from a hundred transactions that happen in one month. Sometimes it will have more and the statement will look like with 19 or 20 pages sometimes.

— Carlos, Accounting, Fornida

The hard part was not getting the statement. It was what came next.

Every transaction needs to be separated and put into each GL account.

— Carlos

Meals and entertainment went one place. Travel reimbursement went somewhere else. Product purchases landed in another bucket. Every line required a judgment, then a manual entry.

And because this was happening across five company credit cards, the time stacked up fast.

We have five different credit cards. So imagine doing that five times a month adding three to four or five hours per credit card statement.

— Carlos

That is how a normal-looking finance task quietly turns into twenty to twenty-five hours a month of repetitive work.

Why this was the right workflow to automate

Steven's explanation is the cleanest one. Accounting workflows like this are exactly what AI is good at when the business gives it the right structure.

Accounting functions that require a lot of data entry, a lot of that stuff can be automated, because that's what AI is actually really good at. Taking data and analyzing it for one, and then taking data and transforming it.

— Steven, Senior Engineer, Fornida

This workflow had the right shape. High volume. Repetition. Recognizable patterns. And real business value if the time got cut down.

It also had a practical constraint that made the problem more interesting. There was no direct integration between Fornida's accounting system and the credit-card source. So the workflow could not rely on an out-of-the-box connector. It had to work with exports.

That meant the automation needed to fit the real environment, not the ideal one.

What got built

The actual operator experience is simple.

We export the transactions directly from the credit cards. He drops it into a little folder, and AI will pick it up and read it and transform it and spits it out.

— Steven

Underneath that simple handoff is the useful part. The workflow maps the transaction patterns to the accounting buckets they usually belong to, then narrows the human review burden down to the entries the system has not confidently identified yet.

So instead of him having to go manually review all, let's say, 800 transactions in these credit card statements, the AI flags the ones that it has not identified yet. So it could be as little as 30 transactions versus the 800 transactions that he's going through.

— Steven

That is the right human-in-the-loop design for finance work.

The system does not pretend to be perfect on day one. It does the bulk classification. The human checks the outliers. The workflow gets better from the corrections.

What changed for the accounting team

Carlos's expectation for the workflow was never "I disappear." It was "I stop doing the dumb part manually."

It's not gonna do the whole job for me, but it's gonna get me to a point where it's complete for us just to review it and make sure everything's actually falling in the right bucket.

— Carlos

That distinction matters because it is how a lot of useful SMB automation works. The repetitive entry goes away. The review step stays. The operator becomes the verifier, not the typist.

Before the workflow, mistakes were normal because the process depended on manual entry. A wrong number in the spreadsheet could throw off the totals and force another pass through the account.

There are errors to be expected. If I enter a three instead of a five or something like that and it will mess up the total, so I will have to go review that account and make sure everything still line up.

— Carlos

After the workflow, the review effort shifts away from checking whether every single line was typed correctly and toward checking whether the remaining anomalies make sense.

The outcome

Farzad gave the clearest before-and-after across the full reconciliation process.

Before, they would get the credit card statement, all our sales orders, all our purchase orders, and they would be reconciling at the end of the month. It would take them about a week. If not, four to five days.

— Farzad Vahid, Founder & CEO, Fornida

And then the new state.

Now that job is done in a few hours. You put it into the workflow that we have, it spits out an answer for you and you're looking for anomalies, things that aren't right, that are in the wrong GL.

— Farzad

That is the win in one sentence. The workflow did not eliminate accounting review. It compressed the time spent getting to the review-worthy part.

Steven's expectation, framed from the operator side, was similar.

Realistically, it's gonna cut down his total time probably from a couple days to maybe an hour a day, per month.

— Steven

Even taking the more conservative version of the story, the result is the same. A workflow that used to consume days now consumes hours.

Why the automation got better over time

This was not a one-shot build. It was an iterative one.

It's not like it's done overnight. It's something that we've put the actual software in place and it's taken time to tweak it to make sure everything gets into the right GL.

— Farzad

That matters because it is where a lot of SMB owners get the wrong expectation. They assume automation either works immediately or it failed. The more accurate pattern is:

Ship the first workable version.
Watch where it misses.
Correct the mapping and logic.
Reduce the exception count over time.

Farzad said it directly in the April session.

When you first automate things, it's not gonna be perfect. If you do something in accounting, it might not get the right GL at first, but as you get further along, it'll make less and less mistakes to where it gets to a point where there's no mistakes made.

— Farzad

That is the real business case for ongoing workflow automation. The value is not just the first deployment. It is the ability to keep tuning until the system is better than the manual baseline.

What this means for other SMBs

Three things this case proves about business workflow automation for small and midsize companies.

The best first workflows are boring. Credit-card reconciliation is not flashy, which is exactly why it is a good first target. The time waste is measurable and the adoption hurdle is low because everybody already feels the pain.
Human-in-the-loop is usually the right design. Finance teams do not need an AI that replaces judgment. They need one that removes the repetitive sorting so humans can spend their time on exceptions.
The real ROI is time plus error reduction. A workflow like this does not just save hours. It reduces the quiet manual mistakes that otherwise get found later, when they are more expensive to unwind.

That is also why this case works as the opening proof point for a broader automation program. Once one team sees that repetitive work can drop from days to hours, they start noticing the next workflow that should be rebuilt.

If you want the broader cluster around this case, the companion reads are:

About Fornida

Fornida is a Plano-based managed IT, cybersecurity, and AI-services company. This case study is intentionally internal. The point is not that Fornida sells an abstract AI service. The point is that Fornida used workflow automation on its own accounting team first, learned the iterative-debug pattern in-house, and now uses that same pattern to think through automation work for clients.

What this kind of workflow usually needs underneath it

This accounting automation did not appear in a vacuum. It depends on the same foundation Fornida pushes in every AI Advantage conversation:

Clean source data
Approved AI usage
Guardrails around who can access what
A workflow owner who knows the business nuance

Without that, the model can classify transactions, but it cannot know which classifications actually reflect how the business wants to run.

That is why Steven's implementation note matters.

You talk to your team. What are some tedious functions that they are doing that takes a lot of time that doesn't seem to make a whole lot of sense nowadays?

— Steven

The operator still teaches the system what good looks like.

That foundation has two parts beyond the workflow itself:

clean source-of-truth data before the model touches anything
approved tools and access controls before employees start improvising with AI

Those two supporting pieces live here:

If your accounting team is still doing this by hand

If somebody on your team is still spending days every month sorting transactions into buckets, reviewing spreadsheets line by line, and catching the same kinds of manual errors over and over, that is exactly the kind of workflow worth looking at first.

The point is not to chase "AI" as a category. The point is to identify repetitive work with a visible cost and rebuild it into something faster and easier to review.

Talk to Fornida if you want to look at one workflow that is already hurting and figure out whether it is ready for automation. Accounting is usually a good place to start because the pain is obvious, the before-and-after is easy to measure, and the human review step is already built into how finance teams think.