How AI Can Find Patterns in Lost Estimates

AI can help contractors find patterns in lost estimates by organizing recorded outcomes across services, salespeople, prices, timing and customer feedback. It cannot discover the true reason from silence. Reliable analysis begins with a consistent loss-reason process and distinguishes customer statements from staff assumptions.

Define when an estimate is lost

Use clear statuses: delivered, viewed, follow-up active, decision deferred, won, lost and expired. Do not mark every old proposal “price loss.” A customer may have paused, chosen another scope, become unreachable or never had authority.

Create a practical reason taxonomy

• Price or financing
• Timing or availability
• Scope or option mismatch
• Trust, proof or communication
• Competitor selected
• Customer postponed or canceled project
• Service or territory not a fit
• No decision or unknown

Allow one primary reason, supporting details and source. “Unknown” is better than invented certainty.

Capture evidence at the right moment

When a customer declines, ask a short, respectful question when appropriate. Let employees record the customer’s words and distinguish their interpretation. Use a close-the-loop message that makes “not now” easy.

Connect the full record

Analysis needs service, location, lead source, estimator, scope, options, price, margin target, delivery time, follow-up, customer response and final status. Remove duplicates and superseded versions.

Do not expose unnecessary personal data to the analysis tool.

Use AI to group themes

AI can cluster free-text reasons such as “start date too late,” “needed completion before move-in” and “competitor available next week” into a timing theme. Review samples to verify that the cluster preserves meaning.

Keep source notes accessible. A chart is a hypothesis generator, not proof.

Segment before drawing conclusions

Compare by service, project size, lead source, season, territory, estimator and customer type. A high loss rate for broad paid-search leads should not be attributed to proposal quality without controlling for fit.

Analyze price without discount panic

Compare estimated gross margin, scope and competitor information when actually known. “Too expensive” may mean outside budget, weak value explanation, different scope or a polite decline.

Test clearer options and scope before cutting price. Measure won-job margin and operational fit.

Analyze speed and follow-up

Look for data-quality patterns

High “unknown” rates, missing follow-up or inconsistent stage dates may reveal process weakness. Fix capture before buying more analytics. AI cannot reconstruct conversations that were never recorded.

Use call and email evidence carefully

Summaries can identify stated concerns, but verify the source. Do not infer an objection from sentiment or silence. Recording and privacy requirements must be reviewed.

Run controlled improvement experiments

Choose one high-confidence pattern. For example, if customers repeatedly ask how two options differ, revise the comparison for that service. Compare a representative period on questions, decision time, win rate and margin.

Do not change pricing, follow-up, proposal and lead targeting simultaneously; you will not know what helped.

Build a monthly lost-estimate review

1. Reconcile statuses and duplicates.
2. Review new loss reasons and unknowns.
3. Inspect source examples for leading themes.
4. Segment by service and lead quality.
5. Select one operational question.
6. Assign an experiment and review date.

Measure healthy improvement

• Known loss-reason rate
• Proposal turnaround
• Follow-up completion
• Win rate by comparable segment
• Gross margin on won work
• Change orders and cancellations
• Customer questions and decision time

Avoid common analytical errors

Survivorship bias: only studying jobs won. False attribution: assuming correlation caused the loss. Small samples: treating three projects as a market trend. Changing definitions: comparing periods with different status rules. Revenue-only thinking: pursuing high win rate at poor margin or bad operational fit.

Clean historical estimate data

Choose a defined period and reconcile customers, duplicate proposals, revisions, stage dates and final status. Link won estimates to completed jobs. Mark records with missing source or unclear outcome instead of forcing them into a clean category.

Standardize service and lead-source labels. “Google,” “Google Ads” and “web” should not be combined unless the tracking evidence supports it.

Create a cohort view

Group estimates by the month delivered and allow enough time for decisions. Comparing all open proposals from this week with completed decisions from six months ago creates a false decline. Long-cycle projects need a longer maturity window.

Show count, total proposed value, average scope, win rate, margin and decision time. Do not let one unusually large project dominate the interpretation.

Analyze option and scope patterns

Compare whether customers received one recommendation or several meaningful options. Review which options were selected and whether descriptions matched the operational differences. A low selection rate may reveal weak fit, confusing presentation or poor price positioning.

Check won projects for change orders and margin. An option that wins frequently but produces operational problems is not necessarily successful.

Interview frontline staff

Ask estimators and coordinators what the dashboard misses. They may identify financing delays, decision-maker absence, competitor scope differences or customers using proposals for insurance. Treat their observations as hypotheses and look for evidence.

Create a loss-review sample

Every month, select several losses from major segments, several unknowns and a few wins. Read the proposal, follow-up and source communication. This prevents summary categories from drifting away from reality.

Record whether the declared reason is direct customer evidence, employee interpretation or unknown.

Use privacy-aware analysis

Aggregate where individual identity is unnecessary. Limit access to proposal and communication data. Do not upload customer records to unapproved public tools. Review retention and model-training use.

Lost-estimate analysis checklist

• Stages and maturity windows are consistent.
• Revisions and duplicates are reconciled.
• Reasons retain evidence level.
• Unknown remains a valid category.
• Segments control for service and lead source.
• Price analysis includes scope and margin.
• AI themes are verified against samples.
• One change is tested at a time.
• Won-job quality is included.
• Customer information uses approved controls.

Calculate loss rates carefully

Use the number of mature decisions as the denominator, not every proposal sent. Report value-weighted and count-based results separately. A company can win many small jobs while losing most large projects, or the opposite.

Show uncertainty when the sample is small. Avoid ranking estimators from a handful of different opportunities.

Investigate “price” with scope normalization

Select proposals lost on price and compare service, quantities, materials, warranty, preparation and exclusions with any competitor information the customer voluntarily shared. Often the bids are not equivalent. Create a field for “scope comparison available” so analysis does not assume it.

When the company genuinely loses comparable work on price, review cost, production and positioning before offering automatic discounts.

Track follow-up quality

Count meaningful contacts, not only automated touches. A receipt confirmation, answered question and scheduled call have different value. Check whether the sequence stopped after a response and whether the salesperson addressed the stated concern.

Analyze the unknown category

Sample unknown losses and test a respectful close-the-loop message. Improve stage hygiene and employee notes. Do not eliminate unknown by requiring staff to guess. A lower unknown rate is useful only when evidence quality improves.

Create decision rules from evidence

A pattern should be specific enough to act on: “Replacement estimates above this scope take longer when the decision-maker misses the consultation,” not “customers need more nurturing.” Assign an owner, change, eligible segment, start date and success metric.

Protect against overfitting

After a change works in one period, validate it in another season or branch. Customer demand and competitor capacity change. Keep a control or baseline and stop changes that improve win rate while reducing margin or increasing cancellations.

Frequently asked questions

Can AI determine why a customer chose a competitor?

Only when reliable evidence exists. It can organize stated reasons, not read the customer’s mind.

Should “no response” count as price loss?

No. Use unknown or no decision unless the customer said price was the reason.

How much data is needed?

Enough comparable cases to see repeatable patterns. Start with process cleanup even when the sample is small.

Can lost-estimate analysis evaluate salespeople?

Use caution and control for lead type, service and opportunity. Review source evidence rather than one automated score.

What is the best first metric?

Known reason rate and time from qualified request to proposal often reveal actionable process issues.

How often should reviews occur?

Monthly for most teams, with faster review after a major process change or unusual loss spike.

Related Oivic guides

• Estimate Follow-Up Messages
• AI Sales Coaching
• Create Professional Proposals
• AI Estimating

Authoritative resources

• NIST AI Risk Management Framework
• FTC AI privacy guidance