Janitorial Intelligence: What We Are Building at taskforse, and Why

Janitorial intelligence is what we are building at taskforse: software that keeps the record of a commercial cleaning company and acts on it. It knows your clients, jobs, crews and rates, and it uses them to do work: cover a call-off, send a quote, chase an unpaid invoice. You approve and supervise instead of carrying every step.

This article is the plain version of the plan. What we are making, why we think the software most cleaning companies already own has stopped paying off, what order we are building in, and what we will not promise.

Key takeaways

• Nucleus Research found the return on record-keeping sales software fell from $4.90 to $3.10 per dollar spent, a 37% decline over ten years.
• In a Harvard Business Review study, firms that answered a lead within an hour were nearly seven times as likely to reach a decision maker as those that waited one hour longer. 23% never answered at all.
• Gartner predicts at least 15% of day-to-day work decisions will be made autonomously by agentic AI by 2028, and also that over 40% of agentic AI projects will be canceled by the end of 2027.
• A cleaning company that depends on its owner sells for less: Breakwater's 2026 guide puts small operators at 1.5x to 2.5x earnings and mid-size commercial ones at 3x to 4.5x.
• The work itself stays human. BLS counts 2.4 million janitors and building cleaners in the US, at a median of $17.71 an hour.

What are we building?

Two systems that work as one.

The first is a system of record. One place that holds every client and every job: the scope, the agreement, the crew assigned, the rate and the history. Around it sit the things a cleaning company does every day: recurring work scheduled once, crews clocking in from the job, inspections with before and after photos that become the report the client receives, and quotes priced from production rates and loaded labor.

The second is a system of action: an agent that reads that record around the clock and does something about it. These are the jobs we are building it to take:

• Cover a call-off. A cleaner says she cannot make tonight. The agent works the backup list on its own, in each person's language, and tells you when the shift is covered.
• Flag what needs you. It tells you what changed, what is at risk and what needs a decision, before you go looking.
• Find work and bid on it. It looks for jobs that fit your routes, then drafts the bid and sends it for your approval.
• Chase unpaid invoices. It spots what has not been paid and follows up with the client until the money comes in.
• Answer the phone. A receptionist in English and French picks up when the office cannot, and logs the details.
• And more.

Put together, the record and the agent become a system of intelligence: it understands the operation well enough to run parts of it, without the owner being the bottleneck. We call it janitorial intelligence because it is built for one trade. A general tool does not know what a day porter is, why a medical office is priced differently from a warehouse, or that a bid is production rates and loaded labor rather than a guess.

Why is keeping good records not enough?

Because the record was never the problem.

The clearest evidence comes from the category that invented record-keeping software. Nucleus Research, which has tracked the return on CRM for more than a decade, re-examined 63 case studies in 2023 and found the return had dropped from $4.90 to $3.10 for every dollar spent, a 37% decline. Their explanation was not that the software got worse. It got more complicated, and a tidier database stopped producing new gains.

Cleaning company owners describe the same thing in plainer words. More than 1,000 companies have joined the taskforse waitlist, and when we ask what hurts most, about two thirds name quoting that takes too long and about two thirds name call-offs and coverage. Neither is a filing problem. Nobody needs a neater record of the shift that went uncovered.

A record tells you what happened. The owner is still the one at 6pm, on the phone, finding somebody to cover a job.

What does that change on a real night?

Here are four jobs every cleaning company does by hand, and what we are building it to do instead.

The job	With records only	With janitorial intelligence
A cleaner calls off at 5pm	You log it, then start phoning	The backup list is worked in order, and you are told when someone confirms
A walkthrough is done	The quote waits until you have an evening free	The quote is drafted from the scope and your rates, ready for you to approve
An invoice is 15 days late	It shows as overdue, if you look	The client gets a reminder, then another, until it is paid or flagged to you
A client goes quiet before renewal	You find out when the notice arrives	The account is flagged while there is still time to fix it

In every row the information is the same. What differs is who moves it forward.

We wrote up the manual version of the first row in building a callout chain that covers a night shift, and it is worth having on paper before any software runs it.

What order are we building it in?

The honest order, because the last step depends on the first.

Now through December: the foundation. The beta began the week of September 28, 2026 with our first design partners, real companies on real nights. These weeks go into making the record solid underneath them. It is unglamorous work, and everything after it depends on it. An agent acting on a wrong record is worse than no agent.

Next: it helps, and you decide. You tell it something out loud and it does it. It flags the account that is drifting and the night that is short. It prepares the quote and you press send.

Then: it runs the jobs you have learned to trust. For the work where it has earned that, it acts and tells you afterward. You supervise the exceptions instead of working every one of them, and you choose which jobs graduate.

None of this is finished, and we would rather say so. taskforse is pre-launch. What is described here is what we are building, in the order we are building it.

Does being faster actually win work?

Yes, and the size of the effect surprises people.

Researchers writing in Harvard Business Review audited 2,241 US companies to see how fast each one answered a new sales inquiry. Only 37% responded within an hour. 24% took more than a day, and 23% never responded at all. Looking at 1.25 million leads, the same study found that firms that tried to make contact within an hour were nearly seven times as likely to have a real conversation with a decision maker as those that waited even one hour longer.

That study is from 2011 and covers many industries, not cleaning specifically. But the mechanism is the one you already know. A property manager who asks three companies for a price is not waiting for the best quote. They are waiting for the first credible one.

The stakes are easy to work out. One account at $3,000 a month is $36,000 a year, and $108,000 over a three-year contract. If the quote goes out four days late because the owner was covering shifts all week, that is the number at risk. The pricing method stays the same, and it is laid out in how to bid a commercial cleaning contract. What changes is that the quote leaves the same day.

Is this just AI hype?

Some of what is sold under that name is, and the people who study it say so.

Gartner predicts that at least 15% of day-to-day work decisions will be made autonomously by agentic AI by 2028, up from essentially none in 2024. In the same release it predicts that over 40% of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear value or weak risk controls. It also names the problem directly: "agent washing", where vendors relabel an existing chatbot or automation as an agent. By Gartner's estimate, only about 130 of the thousands of vendors selling agentic AI are real.

Both predictions can be true, and together they explain how we chose what to build. The technology works for narrow, well-defined jobs with clear rules, and it fails when it is pointed at something vague. Covering a call-off is narrow and well defined. So is sending a reminder on day 15. That is why the list above is short and specific.

It also will not clean a floor. The Bureau of Labor Statistics counts about 2.4 million janitors and building cleaners in the US, earning a median of $17.71 an hour as of May 2025. That work is not going anywhere. What software can take is the office work around it.

Three questions sort the real from the relabeled. Ask them of us too:

1. What does it do without me? If the answer is "it suggests", it is a record with a chat window.
2. What does it do when it is unsure? A real system stops and asks. A bad one guesses.
3. How do I turn one job off? You should be able to hand over call-offs and keep quotes, or the reverse.

What does it do to what the company is worth?

This is the part owners tend to think about last, and buyers think about first.

Breakwater M&A's 2026 valuation guide for cleaning companies puts small residential operators at 1.5x to 2.5x seller's earnings, mid-size commercial companies at 3x to 4.5x, and platform-ready businesses at 5x to 7x EBITDA. On $200,000 of annual earnings, 2x is $400,000 and 4x is $800,000.

The guide is direct about what moves a company between those ranges. Recurring contract revenue is "the most important factor", with buyers looking for retention above 85% a year. Owner dependence works the other way: if you are personally managing crews, handling complaints, doing estimates and running payroll, that "compresses your multiple". And it lists scheduling and dispatch software, quality inspection systems and time tracking among the systems buyers look for.

A company that runs through its owner's phone is hard to sell. A company where the night gets covered, the quote goes out and the client gets a report whether or not the owner is in the country is a different asset. Proof matters here too, which is the subject of how to keep a cleaning contract at renewal.

What can you do before any of this exists?

Three things, this month, with no software at all. They are also what makes software like this work when it arrives.

Get the record in one place. Every client, every job, the scope, who is assigned and at what rate. If that lives in your head and three group chats, nothing can act on it, including a new supervisor.

Measure two numbers. How long from walkthrough to quote sent, in days. How long from call-off to shift covered, in minutes. Most owners have never timed either, and both are the numbers that software which acts is supposed to move.

Pick the first job you would hand over. For most companies it is call-offs, because the rules are clear and the cost of being slow is immediate. For others it is invoice reminders, because it is the job everyone avoids. Start with one.

taskforse is building janitorial intelligence for commercial cleaning companies: one system that holds the record of your clients, jobs and crews, and acts on it. If you are still the one covering the shift and sending the quote at night, join the waitlist.