Where the previous article explained why attribution under-reports AI influence, this one is the practical answer: a framework you can implement in a fortnight. It assumes you accept that upstream attribution is partial and want to measure what is genuinely measurable, properly.
Three layers of measurement
Separate metrics by how directly you can observe them. Mixing layers in one report is how visibility problems get blamed on sales performance and vice versa.
- Layer 1 — Visibility and influence. Partially observable, trend-only, upstream of your systems.
- Layer 2 — Workflow efficiency and quality. Fully observable inside your own tools.
- Layer 3 — Commercial outcomes. Fully observable, but slow and affected by many factors.
Layer 1: visibility and influence indicators
| Metric | Definition | Source | Cadence |
|---|---|---|---|
| Assistant mention rate | Share of a fixed set of category questions where the brand is named | Manual monthly audit log | Monthly |
| Description accuracy | Share of those mentions describing the business correctly | Same audit, scored against approved positioning | Monthly |
| Branded impressions | Search impressions for brand-name queries | Search Console | Monthly |
| Deep direct sessions | Direct sessions landing on non-homepage URLs | Analytics | Monthly |
| Unlinked brand mentions | Public references to the brand without a link | Monitoring tool or manual search | Quarterly |
| Self-reported AI source | Enquiries citing an AI assistant in the free-text source field | Form data | Monthly |
Layer 2: workflow efficiency and quality
| Metric | Definition | Source | Cadence |
|---|---|---|---|
| Median time to first response | Enquiry received to first outbound contact | CRM timestamps | Weekly |
| Preparation time per meeting | Minutes spent preparing, sampled not universal | Weekly sample of five meetings | Weekly |
| Selling time share | Customer-facing hours divided by total working hours | Calendar and activity data | Monthly |
| CRM completeness | Open opportunities with a next step and a close date set | CRM report | Weekly |
| 24-hour follow-up rate | Meetings followed by a logged follow-up within a day | CRM activity | Weekly |
| Weekly active use | People using the AI-assisted step at least once that week | Tool telemetry | Weekly |
| Output quality score | Sampled outputs scored for accuracy, completeness, tone | Manual review of ten samples | Fortnightly |
| Rework signals | Corrections, clarification emails, disputed scope | Manual log | Monthly |
Layer 3: commercial outcomes
| Metric | Definition | Source | Cadence |
|---|---|---|---|
| Qualified enquiry volume | Enquiries meeting the written qualification criteria | CRM | Monthly |
| Enquiry-to-meeting rate | Share of qualified enquiries reaching a first meeting | CRM, by cohort | Monthly |
| Stage conversion rates | Progression between each defined pipeline stage | CRM, by cohort | Quarterly |
| Win rate | Closed won as a share of closed decisions | CRM, by cohort | Quarterly |
| Average deal value | Mean value of closed won opportunities | CRM | Quarterly |
| Sales cycle length | Median days from qualified enquiry to close | CRM, by cohort | Quarterly |
| Pipeline coverage | Open pipeline value against target for the period | CRM | Monthly |
Establishing the baseline
- Write the definition for each metric you will use. One sentence, agreed in writing, before collection starts.
- Name the source system and the person responsible for extracting it.
- Collect four weeks of data with no changes to the workflow.
- Record the mix as well as the numbers: lead sources, sectors, deal sizes. A mix change will otherwise be read as a performance change.
- For Layer 3, use the last two complete cohorts rather than the last two months.
Review cadence and reporting
| Frequency | Audience | Content | Decision |
|---|---|---|---|
| Weekly, 15 minutes | Sales manager and operations | Layer 2 operational metrics | Fix friction, adjust prompts |
| Monthly, 45 minutes | Sales and marketing leadership | Layers 1 and 2, plus enquiry volume | Continue, adjust or extend |
| Quarterly, 90 minutes | Leadership team | All three layers by cohort | Scale, redirect or stop |
Report each metric as current value, baseline value and direction. Three columns. Dashboards with thirty metrics and no baselines get skimmed; a short table with a baseline column gets discussed.
Interpretation rules
- Do not attribute Layer 3 movement to AI alone. Pricing, market conditions and personnel changes all move win rates.
- Require sustained change. Three consecutive weeks or two consecutive cohorts before treating a shift as real.
- Watch quality alongside efficiency. A speed gain with rising rework is a net loss.
- Investigate improvements you cannot explain as carefully as deteriorations; they are often definition or data problems.
- Never report adoption as licences issued. Weekly active use, or nothing.
The measurement discipline here is the same one that governs our 90-day implementation approach, and it is what makes the day-90 scale-or-stop decision a matter of evidence rather than preference.
Key takeaways
- Separate visibility, workflow and commercial metrics; do not mix them in one report.
- Every metric needs a written definition, named source, collection method and cadence.
- Baseline for four weeks with no workflow changes, and record the mix as well as the numbers.
- Selling time share is the metric that proves efficiency gains became real value.
- Report current value, baseline and direction, and require sustained change before acting.