Signs Your Business Needs Custom Software (and Signs It Does Not)

The signs your business needs custom software are measurable, and most of them take an afternoon to check. One workflow that earns money carries more than about 450 hours a year of hand-maintained workarounds (our rule of thumb, roughly a quarter of one person's working year). A documented product limit is less than a year away and no higher plan removes it. Or you pay full seats so people can use one screen. Check the cheap exits first. At QuickBooks Online list prices in September 2026, running out of classes and locations on Plus costs $2,400 a year to fix by moving to Advanced. That is a pricing decision, and many "we need custom software" moments turn out to be one.

Below are seven signs, each with a test you can run this week, then four complaints that are not signs and a four-week test to run before you commit money. If several signs land on one workflow, that workflow is the candidate, not your whole software estate.

1. The workaround has its own instructions

Somewhere there is a document, a pinned message or a long-serving employee explaining what to do because the system cannot do the obvious thing. Save the job status twice so the dispatch email fires. Copy the job number into the notes field because the report cannot see the real one. A workaround with instructions is an unfunded piece of software that your business depends on and nobody maintains.

The test: build a workaround inventory. Sit with the person who does the work for one full cycle of it: a day for daily work, a week for weekly work. Add one row for every step that exists because of the tool rather than the business. Then ask them to log the minutes for two weeks instead of estimating, because habitual steps are the ones people forget to mention.

Workaround inventory (the four rows are an illustrative example)
Step	Why it exists	Who does it	Minutes per week	What breaks if they are off sick	Who covers
Re-key approved quotes from the CRM into the job system	G	Sales coordinator	150	Jobs start without materials ordered	Nobody trained
Save each job status twice so the dispatch email sends	L	Dispatcher	20	Technicians are not notified	Operations manager
Merge two timesheet exports and fix the job codes	G	Payroll administrator	120	Payroll runs late	External bookkeeper
Check a "do not ship" list kept in a spreadsheet	R	Credit controller	30	Goods ship to customers on credit hold	Nobody

Code the "why" column with one letter: L for a product limit, G for a gap between two systems, R for a business rule the product cannot hold, and H for habit with no current reason. Then convert the minutes into hours per year, using 48 working weeks. With minutes in column D, this formula does it:

=SUM(D2:D40)*48/60

Run it in the inventory spreadsheet. The four example rows total 320 minutes a week, which is 256 hours a year.

What the result means: H rows cost nothing to stop, so stop them. L rows belong to sign 3 and G rows to sign 2. R rows are where custom software earns its keep, if anywhere. Any row with "Nobody" in the last column and a revenue, customer or compliance consequence is key-person risk, and it counts however few minutes it takes.

2. A spreadsheet sits between two systems that should exchange information

Someone exports from one system, adjusts the data and imports or re-keys it into another. The spreadsheet is doing translation, and it usually holds rules that exist nowhere else, such as which product codes map to which delivery category.

The test (ten minutes, done by whoever owns the file): find the rules hiding in it.

In Excel, go to Home → Find & Select → Go To Special and choose Formulas to highlight every calculated cell. Run it again with Constants. A typed number sitting inside a column of formulas is a manual override somebody made once and nobody tracks.
Next to any calculated column, add this check. It shows FALSE wherever someone typed a value over a formula:
```
=ISFORMULA(C2)
```
Count the lookup tabs (one code mapped to another) and the colors that carry meaning, such as "orange means waiting for a PO".

What the result means: if the rules are code-to-code mappings and copied fields, you need an integration, not custom software. Our guide to connecting business systems without replacing everything covers how to scope one. If the spreadsheet holds decisions, such as who gets which price, what gets held back or what counts as complete, write each rule down as one sentence. A rule nobody can write as a sentence cannot be automated by anyone, with custom software or without it.

3. You are splitting data to stay under a documented product limit

Every product publishes ceilings in its plan documentation, and they are what outgrowing a tool looks like in practice. Some that this audience meets often, as of September 2026:

Documented product limits, as of September 2026
Product	Limit	What happens at the limit
QuickBooks Online	Billable users: Simple Start 1, Essentials 3, Plus 5, Advanced 25. Time-tracking-only users do not count, and neither do reports-only users (not available on Essentials).	Intuit's two options are to upgrade or to reduce usage. Advanced is the top QuickBooks Online plan.
QuickBooks Online	Chart of accounts: 250 on Simple Start, Essentials and Plus. Unlimited on Advanced.	Only active accounts count, so making accounts inactive buys room.
QuickBooks Online	Tracked classes and locations: 40 combined on Plus. Unlimited on Advanced. Not available on Simple Start or Essentials.	Upgrade, or deactivate classes and locations.
Airtable	Records per base: 1,000 on Free, 50,000 on Team, 125,000 on Business, counted across all tables in the base.	The base stays readable but takes no new records until you upgrade or delete. For the Free plan, Airtable spells out that this includes records from forms and the API.
HubSpot	Custom objects, for things like sites, equipment or contracts, need an Enterprise subscription.	On Starter or Professional, that data has to live in standard objects or properties.

Sources: Intuit's QuickBooks Online usage limits, Airtable's plans overview and HubSpot's custom objects documentation. Vendors change these, so check the pages before you rely on them.

The sign is not hitting the limit. It is the workaround people invent to stay under it, because each one quietly breaks something else:

Out of classes and locations on QuickBooks Plus, people put the location into account names ("Sales - Denver"). That spends the 250-account allowance, and every report by location now needs someone to regroup accounts by hand.
Out of records in an Airtable base, people start a second base for "last year". Every report that spans both halves now needs a person to combine them.
Without custom objects in HubSpot, a site or a piece of equipment gets stored as a text property on a company record. A property is a field, not a record, so the site cannot have its own owner, pipeline or activity history.

The test: for each tool, write down the documented limit, your current usage (active items only, where that is how the vendor counts) and how much usage grows each month. Then:

Months to the ceiling = (limit − current usage) ÷ monthly growth

An illustrative example: 31 active classes and locations on QuickBooks Plus, adding one a month for new sites and job types, gives (40 − 31) ÷ 1 = 9 months.

What the result means: if a higher plan removes the ceiling, price that plan and compare it with nothing else yet. QuickBooks Online Plus was $140 a month and Advanced $340 a month at list on Intuit's pricing page in September 2026, a difference of $2,400 a year. A ceiling becomes a sign only when there is no higher plan (25 users is the top of QuickBooks Online), when the next plan's ceiling is also in sight, or when the workaround has already started breaking reports.

4. You pay full seats for people who use one screen

Per-user pricing assumes every user needs the whole product. A warehouse team recording three fields costs the same as a power user, and the cost grows quietly with headcount.

First, check for a lighter seat. Many vendors sell one, and some are free. HubSpot's View-Only seats are free and unlimited on Starter, Professional and Enterprise, but they cannot edit records. QuickBooks Online time-tracking-only users do not count toward the user limit. Microsoft sells a Dynamics 365 Team Members license for light, designated scenarios. If a lighter seat covers what the group does, change the licenses and stop here.

Then do the arithmetic. An illustrative example, with 35 warehouse and field users who each use one screen:

Annual seat cost       = users × price per user per month × 12
                       = 35 × $110 × 12 = $46,200

Three years, rising 7% at each renewal
                       = 46,200 + 49,434 + 52,894 = $148,528

Narrow internal tool   = build + 3 × annual running cost
                       = $55,000 + 3 × $12,000 = $91,000

Three-year difference  = $148,528 − $91,000 = $57,528

Break-even group size  = $91,000 ÷ ($110 × 12 × 3.2149) ≈ 22 users
  (3.2149 = 1 + 1.07 + 1.07², the three years of price increases)

Put in your own price, group size and renewal increase. Below the break-even group size, keep the seats.

Then read the licensing terms, because they can erase the saving. If the narrow tool reads and writes the platform's data through one integration account, the vendor may still require a license for every person behind it. Microsoft's multiplexing guidance states that multiplexing "does not reduce the number of licenses needed". Its own example is colleagues editing Dynamics 365 data through Power Platform, who then need Dynamics 365 licenses as well as Power Platform ones. In the example above, if each of the 35 users still needs a $25 light license, the three-year saving falls from $57,528 to $23,772. If they need full seats, the tool saves nothing. Get the vendor's answer in writing before anyone quotes a build.

5. A recurring report takes someone most of a day

Someone assembles the same report every week or month from two or three systems and reconciles the differences. The person is usually senior, the figures are usually questioned, and decisions wait for it.

Cost it first. Annual cost = hours per cycle × cycles per year × loaded hourly rate. For the loaded rate, the US Bureau of Labor Statistics reports that in June 2026, employers paid full-time private-industry workers $54.00 per hour worked in total compensation against $36.97 in wages and salaries (Employer Costs for Employee Compensation). That is a ratio of 1.46, so:

Loaded hourly rate ≈ annual salary ÷ 2,080 × 1.46

What recurring reporting costs (an illustrative example)
Line	Inputs	Annual cost
Monthly management pack	14 hours × 12 months × $74 (finance manager on $105,000)	$12,432
Weekly operations figures	5 hours × 48 weeks × $44 (coordinator on $62,000)	$10,560
Arguing about the numbers	5 people × 45 minutes × 12 meetings × $74	$3,330
Total		$26,322

Treat the total as capacity rather than cash. It becomes a saving only if something changes in staffing, overtime or outside spending.

Then run the definitions test, before you buy anything. Ask three people who use the report, separately and in writing: "How many active customers did we have last month?" and "What was our on-time delivery rate last month?" Ask each of them to write down the rule they used as well as the number.

Three numbers, three rules: a definitions problem. "Active" might mean ordered in the last 12 months, has an open contract, or was invoiced this month. Software cannot fix that. A one-page definition can.
Three numbers, one rule: a data problem. The systems disagree, which points to an integration or a reconciliation, not a new application.
One number: the report is sound and only the assembly is manual. That is the case where automation pays.

On-time delivery hides a trap worth checking on its own. The same shipments give different rates measured against the customer's requested date, the first promised date or the current promised date. Measuring against the current promised date hides every delay that was re-promised before it happened, so the rate looks best exactly when customers are waiting.

Settle each disputed figure on a card like this before anyone automates it:

Metric:     On-time delivery rate
Owner:      Operations manager
Counts:     Order lines shipped complete on or before the FIRST promised date
Excludes:   Delays the customer requested; cancelled lines
Source:     Shipment date and original promise date, from the ERP
Period:     Calendar month, by ship date
Refreshed:  Daily, 6:00 a.m.

This matters even more before putting an AI assistant over the reporting data, because an assistant that turns questions into queries picks a definition silently each time. Our account of making AI answers about numbers checkable shows what catching that takes.

6. New staff learn the workaround before they learn the job

When onboarding covers tool quirks before the work itself, the complexity has moved from the software into people. It appears in no budget line, but it slows new staff down and concentrates risk in whoever knows the most tricks.

The test: a 20-minute conversation with your newest hire, someone who joined in the last three months, without their manager in the room. Ask:

What did someone have to show you that was not written down anywhere?
Which step do you still check with a colleague before doing?
What did you get wrong in your first two weeks, and who noticed?

Mark each answer T if it is about the tool (a field filled in a particular way, a screen to avoid, a button pressed twice) or B if it is about the business (a customer's terms, a pricing rule, a safety step). Then ask their manager how many weeks passed before the new hire could run the workflow alone, and which step came last.

What the result means: more T answers than B answers means the tool is being taught before the job. If the answer to the third question is "a customer noticed", the workaround has no check built into it, and that row goes to the top of your inventory.

7. You are waiting on someone else's roadmap for something central

You raised a feature request that matters to how you operate. The answer was "under consideration", "planned" or "available on a higher tier" that does not otherwise suit you.

The test: assume the answer is a definitive no, and cost it.

Ask the vendor three questions in writing. Is this on a committed roadmap with a target release? Which plan will it be on? Can we do it through your API today, and within what rate limits?
Treat anything short of a named release on your current plan as a no. A roadmap is not part of your contract, so if it slips you have no remedy unless the order form names the feature and a date.
Cost the permanent workaround: annual hours from your inventory × loaded hourly rate × 3 years.
Decide. If the three-year cost is small, accept the workaround and stop asking. If it is large and the API can do the job, a small extension on top of the product is the narrowest build available. If it is large and the API cannot, this is the strongest sign on the list, because the vendor has told you the ceiling will not move.

A rule we use: a request that has sat at "under consideration" for a year is a no.

Four things that are not signs

Complaints that do not justify custom software by themselves
Complaint	Why it is not a sign	What to check instead
Staff dislike the interface	Preference is not obstruction. A custom replacement usually becomes a differently disliked tool.	Ask them to log the disliked task for two weeks in the inventory. If it does not show up as minutes, it is preference.
The software does more than we need	Unused features cost nothing operationally. This is a purchasing question.	Whether a lower tier drops only features you do not use. Moving from QuickBooks Plus to Essentials, for example, loses class and location tracking entirely.
We have too many tools	Consolidation rarely pays for itself, and fewer systems can mean worse fit in each one.	List the handoffs between tools and count the fields re-keyed at each. The handoffs cost time. The number of tools does not.
A competitor built their own	You cannot see what it cost them or whether it worked, and their constraints are not yours.	Run the seven tests above on your own workflow.

Before you commit: a four-week test

Collect four weeks of evidence before a vendor or developer is in the room, on the single workflow where the most signs landed.

Keep the workaround inventory running for four weeks, with minutes logged rather than estimated. Multiply the four-week total by 12 for an annual figure.
Write down every product limit the workflow touches, with its months-to-ceiling figure.
Run the definitions test on every figure the workflow reports.
Get the vendor's written answers on lighter seats, higher plans, licensing for indirect access and the roadmap.

Then read the results against this table. It is our rule, not an industry standard.

Reading the four-week results
Result	What it usually means	Next step
Under 450 workaround hours a year, and no single-person step with a revenue, customer or compliance consequence	The tools are adequate	Fix configuration, training or licensing. No project.
Most of the hours sit in one handoff between two systems (G rows)	An integration problem	Scope one integration
A ceiling within 12 months that a higher plan removes	A pricing problem	Price the plan
The definitions test returned different rules	A definitions problem	Write the metric cards, then re-run the test in a month
Three or more signs on a workflow that earns money; over 450 hours a year or a single-person step that stops revenue; and vendor answers that rule out a lighter seat, a higher plan or a roadmap fix	A real case for building something narrow	Compare the options properly with our integrate, replace or build guide

Why 450 hours: a custom tool has running costs every year, for hosting, support and changes when the systems around it change. Below roughly a quarter of one person's working year, those costs tend to eat the benefit.

Keep the inventory after you decide. Re-run it three months later, whatever you chose. The rows that disappeared are your measured result, and the rows that remain are the next list of work.

If several of these signs land on one workflow, bring the workaround inventory, the vendor's written answers and one recent example of the work going wrong to a conversation. We will look first at where the hours sit, and whether a lighter license, a higher plan or one integration removes them, before discussing custom business software.