Analytics and Reporting Skills for AI Agents

Purpose

Find which discovery questions reps ask but never get a real answer to, so the call script can be trimmed.

Inputs

• Call transcripts or export from Gong or similar
• The current list of standard discovery questions
• Time window and rep or team filter

Steps

1. Pull transcripts for the chosen window and segment by rep.
2. For each transcript, detect when a question from the standard list was asked, allowing for reworded phrasing.
3. Classify the prospect response as answered, deflected, or ignored based on the following turns.
4. Tally per question the ask count, answer count, and answer rate.
5. Flag questions with high ask volume but answer rate under 40 percent as low value.
6. Note questions that reliably produce long answers as high signal keepers.
7. Compare rates across reps to separate script problems from delivery problems.

Output

A table ranking each discovery question by answer rate with keep, reword, or drop recommendations and rep level notes.

Purpose

Show how long each segment takes to move from first outreach to a booked meeting so you can spot slow cohorts.

Inputs

• CRM or sequencer export with contact ID, first touch date, meeting booked date
• Segment fields such as industry, company size, persona, and channel
• Reporting window, for example the last 90 days

Steps

1. Load the export and keep only contacts that booked a meeting inside the window.
2. Compute lag in days as meeting booked date minus first touch date for each contact.
3. Drop rows with negative or missing dates and log them as data quality flags.
4. Group by each segment field and calculate median, 25th, and 75th percentile lag.
5. Rank segments from slowest to fastest median lag and flag any above the overall median.
6. Note segments where sample size is under 10 as low confidence.

Output

A table of segments with median and percentile lag, a slow-cohort flag list, and a short note on data quality gaps.

Purpose

Track the gap between booked and held meetings so you can find where prospects ghost and fix it.

Inputs

• CRM or calendar export with meeting booked date, scheduled date, held flag, source, and rep
• Date range to analyze
• Minimum sample size per segment (default 10)

Steps

1. Load the meeting records and confirm each row has a booked timestamp, scheduled slot, and a held or no-show status.
2. Compute overall no-show rate as no-shows divided by total scheduled meetings.
3. Break the rate down by lead source, rep, day of week, and days between booking and scheduled slot.
4. Flag any segment above the overall rate by 10 points or more, but only if it clears the minimum sample size.
5. Correlate no-show rate with booking lead time to see if far-out bookings ghost more.
6. Suggest two concrete fixes per flagged segment, such as tighter reminder cadence or shorter booking windows.

Output

A ranked table of no-show rates by segment with flagged leak points and specific reminder or scheduling fixes.

Purpose

Tie booked meetings back to the campaign, channel, and sequence that actually created them.

Inputs

• A meetings export with contact, booked date, and owner
• Campaign and sequence membership per contact
• The channels in play, for example email, LinkedIn, cold call

Steps

1. Match each meeting to the contact record and pull the campaigns and sequences that contact was in.
2. When a contact touched more than one channel, apply a clear rule such as last touch before the booking.
3. Roll meetings up by channel, by campaign, and by sequence.
4. Pair each meeting count with the volume that produced it to get a meetings per 100 contacts rate.
5. Flag meetings with no traceable source so the tracking gap is visible, not hidden.
6. Rank campaigns by meetings booked and by efficiency, since the highest volume is rarely the most efficient.

Output

An attribution table of meetings by channel and campaign, with a per 100 contacts efficiency rate and a count of untracked meetings.

Purpose

Show which buyer personas convert best so reps target the right titles and messaging.

Inputs

• CRM export of closed deals with primary contact title, stage, amount, and close date
• Persona mapping rules that group raw titles into standard personas
• Date range for the analysis window

Steps

1. Load the closed deal export and normalize each primary contact title against the persona mapping rules.
2. Bucket every deal into a persona and flag any title that fails to match a rule for manual review.
3. For each persona, count total deals, won deals, and lost deals, then calculate win rate and average deal size.
4. Compare persona win rates against the overall blended average to highlight over and under performers.
5. Segment win rate by deal size tier to check if a persona only wins small deals.
6. Rank personas by a combined score of win rate and average amount.

Output

A ranked table of personas with win rate, deal count, average size, and a short note on which titles to prioritize or deprioritize.

Purpose

Show where each rep's open pipeline falls short of the coverage needed to hit quota this quarter.

Inputs

• CRM export of open opportunities with owner, stage, amount, and close date
• Quota target per rep for the current quarter
• Standard win rate by stage

Steps

1. Load the opportunity export and filter to deals with a close date inside the current quarter.
2. Group open amount by rep and by stage, then apply the stage win rate to get a weighted value.
3. Divide each rep's total weighted pipeline by their remaining quota to compute a coverage ratio.
4. Flag any rep below a 3x raw coverage ratio or below 1x weighted coverage.
5. For flagged reps, list the missing dollar amount and the count of new deals needed at average deal size.
6. Rank stages that hold the most stalled value to point at where to add or advance deals.

Output

A per rep table with coverage ratios, gap dollars, needed deal count, and a short prioritized note on where to focus.

Purpose

Explain why a campaign reply rate is what it is by splitting it into cohorts instead of one flat number.

Inputs

• A send level export with contact, industry, title, sequence, send date, and reply status
• The reply categories you track, for example positive, neutral, negative

Steps

1. Compute the overall reply rate and positive reply rate as a baseline.
2. Break reply rate down by industry, by seniority, and by sequence, and rank each cut best to worst.
3. Flag any cohort with a sample too small to trust, for example under 50 contacts.
4. Compare positive reply rate, not just total replies, so angry replies do not read as wins.
5. Call out the two cohorts pulling the average up and the two dragging it down.
6. Suggest one concrete action per weak cohort, such as a new angle or a tighter filter.

Output

A cohort table with reply and positive reply rates, sample sizes flagged, and a short read on where to double down or cut.

Purpose

Detect segments where deals are closing slower than a prior period so you can act before pipeline stalls compound.

Inputs

• CRM export of closed won deals with created date, close date, segment, deal size, and owner
• Two time windows to compare, for example last quarter versus the quarter before
• Optional stage timestamp fields if available

Steps

1. Compute median days from created to close for each deal in both windows.
2. Group by segment, deal size band, and owner, then calculate median cycle length per group per window.
3. Flag any group where the current median grew more than 15 percent versus the prior window.
4. If stage timestamps exist, isolate which stage absorbed the added days.
5. Rank flagged groups by combined impact of drift size and deal count.
6. Write a short note per flagged group naming the likely bottleneck stage or owner.

Output

A ranked table of segments with growing cycle length, the days added, the stage where slowdown occurs, and a plain summary of the top three drifts to investigate.

Purpose

Pinpoint the exact step in a multi-touch sequence where reply and open momentum collapses so you can rewrite or retime it.

Inputs

• CSV export of sequence activity with columns for step number, sent, opened, replied, bounced
• Sequence name and total enrolled contacts
• Copy text for each step, if available

Steps

1. Load the CSV and compute per step open rate, reply rate, and bounce rate against contacts still active at that step.
2. Calculate step to step survival, meaning how many contacts advance versus fall out at each touch.
3. Flag the single step with the largest drop in reply rate relative to the prior step.
4. Cross reference that step against its copy to note length, ask type, and send day.
5. List two likely causes for the drop, such as weak call to action or send timing clustering.
6. Recommend one concrete change per flagged step and predict the recovery direction.

Output

A ranked table of steps by drop-off severity plus a short fix list for the worst two steps.

Purpose

Turn raw sending data into a one page weekly scorecard leadership can read in a minute.

Inputs

• This week's send, reply, and meeting numbers per campaign
• Last week's numbers for comparison
• Your targets for reply rate and meetings

Steps

1. Pull the core metrics per campaign: sends, bounce rate, reply rate, positive replies, and meetings booked.
2. Show each metric next to last week and mark the change up or down.
3. Compare every metric to its target and flag red, yellow, or green.
4. Write a two line summary of what moved and why, in plain language.
5. List the top three actions for next week based on the reds.
6. Keep the whole thing to one page or one Slack message, numbers first.

Output

A one page scorecard with per campaign metrics, week over week deltas, target status, and three next actions.