An Unsold Estimate Is an Abandoned Cart

01What the abandoned cart comparison gives us

Baymard Institute calculates an average documented ecommerce cart-abandonment rate of 70.22% across 50 studies. That describes online shopping carts across the studies it compiled. It gives us no estimate-abandonment rate for roofing, tree care, electrical work, or any other trade. [1]

The useful connection is the stage of the buying process. An inquiry expresses interest. A delivered proposal puts a defined offer in front of someone. Approval records a commitment. Completion and payment establish later outcomes.

For analysis, I would treat each transition as a separate question. If proposals are prepared but never delivered, investigate the handoff. If homeowners receive them and ask unanswered questions, examine communication. If approvals arrive but jobs never reach completion, follow the record into scheduling and delivery.

The analogy also has limits. Consider a replacement roof: the contractor must define the assembly, inspect conditions, explain exclusions, and establish how additional damage will be handled. A homeowner comparing that proposal with another is evaluating work whose scope may differ materially.

That comparison deserves attention before we prescribe a reminder sequence. “You left something in your cart” tells us very little about whether the homeowner understands the recommendation or has received a comparable alternative.

There is also a limit to the money supposedly waiting to be recovered. Imagine three contractors each submitting a $12,000 proposal for the same roof. The homeowner hires one. The other two record $24,000 of unsold estimates between them, while the household purchases one roof. Adding those losses together would count competing offers as though they represented separate projects the household intended to buy.

At the individual company, winning that project could still have been valuable. Establishing the opportunity requires asking what the company could have changed and whether winning would have justified the cost. The unapproved dollar amount cannot answer that question. A sales target built from every rejected offer would ask the business to win decisions that were mutually exclusive from the beginning.

The first task is to locate the unfinished decision precisely enough to investigate it.

02Start by questioning the close rate

Jobber’s 2026 Home Service Trends Report surveyed 1,050 U.S. home-service business owners in December 2025. It reports that 69% said they won more than half their quoted jobs. The report combines survey findings with separate platform benchmarks. These are useful directional observations, with different evidence behind each. [2]

They do not establish that a particular contractor should close a particular percentage. The owner responses cover different trades, company sizes, and business circumstances. A reported win rate also leaves questions about how each respondent counted proposals and decisions.

I would settle those definitions before comparing performance.

Take an illustrative cohort of 100 purchasing opportunities. Forty are approved, twenty explicitly declined, fifteen deferred by the customer, and twenty-five have no reliable outcome recorded.

Approval across the full cohort is 40%. Approval among the sixty opportunities with an explicit approval or decline is 66.7%. Both calculations are correct. They describe different populations.

Neither calculation tells us why the twenty-five unknown opportunities remain unknown.

Count the underlying purchasing decision consistently. Three mutually exclusive roof options belong to one opportunity. A revised painting proposal should remain connected to its earlier version. Otherwise, a customer choosing one option can leave several apparent losses in the reporting.

Then give comparable opportunities equal time. Choose the observation window for the work being studied and report approvals within it. Comparing yesterday’s landscape installation estimates with last quarter’s mature cohort would mix decision time with sales performance.

These definitions determine what the number means. A benchmark becomes useful only after that meaning is stable.

03Keep the customer decision separate from the record

I would record the commercial outcome separately from the explanation and the evidence supporting it.

“Declined” records a decision. “Chose another contractor” describes an outcome the customer reported. “Price” is a reason that needs context. “No response” describes the company’s contact history.

Putting all four into one status field makes the analysis harder to interpret.

Suppose a homeowner tells a fencing contractor, “We are waiting until after the driveway is replaced.” Record the deferral and the stated dependency. Agree on a suitable next contact. There is no reason to turn that statement into a price objection.

Suppose another homeowner never replies. Preserve the unknown outcome. Three unanswered calls establish three attempts to make contact. They establish no customer motive.

My minimum record would connect the opportunity to the proposal version, delivery date, quoted scope and amount, customer statements, contact attempts, and eventual job. Keep the source and date beside any recorded reason.

An expired proposal also needs its own meaning. Expiration describes the validity of an offer. It does not explain whether the customer postponed the project, selected another company, or stopped considering it.

This discipline matters when deciding what to fix. If “price” includes every unanswered estimate, a pricing problem has been built into the reporting before anyone investigates it. If the customer’s actual words remain available, the team can examine the interpretation.

04What the homeowner still needs to decide

The Federal Trade Commission advises homeowners to obtain multiple estimates. Its guidance says written estimates should describe the work, materials, completion date, and price, and recommends asking about substantial differences between bids. Comparison is part of the consumer guidance contractors are selling into. [3]

For the contractor, that raises a specific question: can a homeowner explain the difference between this recommendation and another offer?

Consider an illustrative exterior painting project. One proposal includes washing, scraping, spot priming, two finish coats, and defined treatment of damaged surfaces. Another says “paint exterior” beside a lower price. The homeowner needs enough information to compare the promised work.

I would make scope review part of investigating unapproved estimates. Check what is included, what remains uncertain, what completion means, and how changes will be discussed. Where duration affects access or disruption, explain the expected schedule. A fixed price can accompany a detailed explanation of the work.

Affordability deserves a separate investigation. Angi’s 2025 State of Home Spending Pulse survey covered 1,000 U.S. homeowners who had hired a professional during the previous year. It reported that 71% had postponed a planned project that year and 58% said high material or labor costs caused delays. The survey ran April 15–16, 2025. [4]

Those responses concern project postponement. They cannot tell us which of your delivered estimates stalled because of cost. They do establish a reason to take household budget constraints seriously when customers describe them.

A homeowner postponing window replacement has a different situation from one selecting a competitor for tree removal. A customer asking how an HVAC recommendation was reached needs an explanation of the diagnosis. Repeating the original offer addresses neither question by itself.

I would ask what remained unresolved: the recommendation, the scope, the budget, the timing, or the choice of provider. Let the answer determine the next conversation.

05Follow the estimate through the economics

Here is a simplified illustration, entirely invented. A residential service company issues 100 estimates averaging $6,000. Each represents one purchasing opportunity with one selected scope. All completed jobs in the model have the same price, and no partial work earns revenue.

The model defines direct delivery costs as the labor and related employment costs assigned to completed work, materials, subcontractors, and other job-specific delivery expenses. The separate $32,000 includes acquisition and estimating costs for the entire cohort. Nothing is counted in both groups.

Five approved jobs remain unfinished. Their $30,000 contract value stays outside completed revenue in this illustration. The $12,000 receivable belongs to completed work and remains separate from cash received.

The company has $360,000 of unapproved quoted value. Calling that lost revenue assumes every remaining opportunity could have become completed work at the quoted price. The model supplies no basis for that assumption.

It also supplies no basis for applying the completed jobs’ 40% gross margin to every unsold project. The remaining scopes could require different labor, access, equipment, or subcontracting.

The acquisition and estimating costs already incurred still matter. They belong in evaluating the full process and deciding which opportunities deserve future investment. When deciding whether to contact an existing prospect again, compare the additional cost and likely additional contribution from that next action.

That distinction protects the decision from a sunk-cost argument. Spending money to prepare an estimate creates a reason to learn from the outcome. Further spending needs its own justification.

06What would follow-up have to earn

I would assess a follow-up program through a defined comparison: how much additional contribution does it generate relative to the process the company would otherwise use?

Here is another illustration. A business has 100 eligible open opportunities. A proposed follow-up process costs $1,200 in additional staff time and tools across those opportunities. Assume each additional completed job contributes $1,800 after its additional delivery costs. Assume available capacity and no displaced work.

With no additional jobs, the program reduces contribution by $1,200. One additional completed job produces $600 after the program cost. Four produce $6,000. The result depends on jobs that would otherwise have remained uncompleted with this company.

That last condition is essential. Four customers purchasing after receiving messages does not establish four additional purchases. Some could have approved the work under the existing process.

Discounting changes the calculation again. If all ten eventual buyers receive a $300 discount, the reduction is $3,000, including discounts given to customers who would have purchased anyway. Count that full reduction when evaluating the intervention.

Capacity belongs in the same model. If a tree-care crew has an unfilled day, additional work can use that opening. If the recovered project displaces another profitable job, subtract the contribution given up. Include extra travel, equipment rental, or overtime when the added work requires them.

For constrained operations, I would also compare contribution per scarce crew hour. Specify whether that hour includes travel and setup. A job’s contribution divided by an inconsistent time measure gives an unreliable comparison.

This is where sales analysis becomes an operating decision. The work worth recovering depends on the margin, resources, and commitments of the business expected to deliver it.

07Test the change against the actual problem

Start with one segment and one intervention. For example, investigate residential replacement estimates that remain unresolved after a defined period. Review the actual records and speak with the team before deciding what customers need.

If the recurring question concerns scope, test a clearer proposal and an opportunity to discuss it. If customers ask about scheduling, test a process that provides an accurate scheduling answer. Define the change tightly enough that the eventual result has a useful interpretation.

Where volume permits, I recommend randomly assigning eligible households to the existing process or the revised process. Keep multiple estimates for the same household together. Preserve essential service communication in both groups and record what each actually receives.

Measure approval within a predetermined window, then follow the jobs through completion, delivery cost, and payment. Use the same observation period for both groups. If jobs remain unfinished when the test ends, report that limit explicitly.

Evaluate contribution per eligible opportunity, including the cost of applying the process to people who never buy. Also record opt-outs, complaints, and staff workload. Those observations help determine whether an economically promising process is suitable to continue.

Random assignment gives the comparison a stronger basis than contacting only the prospects staff already believe are most likely to purchase. With small samples, report the uncertainty around the difference. A handful of extra approvals warrants continued investigation before a precise annual return is claimed.

If a concurrent comparison is impractical, use a limited pilot and name the other changes during it. Weather, pricing, staffing, and demand should remain visible in the interpretation. A before-and-after improvement alone cannot isolate the contribution of follow-up.

08Use AI to organize evidence you can inspect

There is a useful AI application here: propose structured labels from existing customer messages and notes, then let a person inspect the supporting text.

NIST’s Generative AI Profile identifies confidently presented false information as a risk and recommends verifying sources and citations in model outputs. For estimate analysis, that supports keeping every proposed classification connected to its source and checking performance before relying on the results. [5]

I would begin with a manually reviewed sample. Define separate labels for stated outcome, stated reason, and missing evidence. Ask the system to extract the relevant passage, identify its source, and return “unknown” when the record supports no conclusion.

Then compare its classifications with human review. Examine the errors by category. A system that repeatedly converts “left voicemail” into “customer declined” would manufacture sales losses from contact attempts.

Keep customer statements separate from employee interpretations. “Customer said another bid was lower” supports a reported price comparison. “Probably shopping around” records an employee’s hypothesis. The analysis should preserve that difference.

Limit the customer information supplied to what the task requires. Retain corrections so the team can trace how a summary changed. The proposed benefit is less manual sorting with evidence still available for inspection. Any claim that it saves time or improves decisions needs measurement in the actual workflow.

09Decide what to change from what you learn

I would begin with a manageable cohort from one service category. Reconcile the proposals with customer communications, completed jobs, and payment records. Count explicit decisions separately from unknown outcomes, and identify which explanations come directly from customers.

Then choose a change that addresses a documented problem. That could mean answering scope questions earlier, improving delivery of proposals, following up at an agreed time, or recording decisions more consistently. It could also mean spending less effort pursuing work the business cannot deliver profitably.

Judge that change by the additional work completed and the contribution retained, alongside its cost and effect on customers and staff. Keep unresolved outcomes visible until the evidence changes.

The $9,000 estimate at the beginning gives us an opportunity to ask better questions. What did the homeowner decide? What could the company have done differently? Would that change have produced work worth performing?

Those answers give an owner something useful to act on. The dollar amount sitting in the pipeline cannot supply them.