AI sales assistant ROI: measure the path from conversation to revenue

Start with a result your sales team accepts

An AI conversation is not automatically a qualified lead. A visitor may be exploring, looking for support or asking a question that has no buying intent. Before launching a sales assistant, agree what makes an enquiry worth a salesperson’s time: a relevant need, a plausible fit and a requested next step.

Use that definition consistently when reviewing both AI-assisted and other enquiries. Record why a lead was accepted or rejected. A short, accurate conversation that leads to a useful human discussion may be more valuable than a long transcript or a completed form.

Keep the funnel stages separate

Measure eligible visitors, conversations started, useful answers, consented handoffs, sales-accepted leads and won customers as distinct stages. Connect records only through an appropriate identifier and the data your team is authorised to use.

A website can measure a signup click without knowing whether registration was completed in a separate application. Label that metric as a click until a verified completion event is available. Similarly, do not count a generated handoff summary as a CRM record unless the receiving system confirms it.

• Connection rate: connected sessions divided by conversation starts.
• Accepted handoff rate: sales-accepted enquiries divided by eligible conversations.
• Win rate: won customers divided by the relevant accepted-lead cohort, after enough time has passed.
• Quality: unsupported claims, missing context and reviewer corrections.

Calculate value without double counting

Estimate incremental contribution from additional won business, then subtract the full cost of the programme. Use the same time period and include subscription or usage fees, implementation, integrations, maintenance and human review. If the pilot has not produced closed deals, report leading indicators rather than declaring revenue ROI.

For illustration only: suppose a comparable pilot produces two additional customers, each contributing $600 after variable delivery costs. The incremental contribution is $1,200. If the pilot costs $800 in total, the net contribution is $400 and the illustrative ROI is 50%: ($1,200 minus $800) divided by $800. These are invented figures, not Aivah customer results.

Do not add the full sales value and its contribution margin as two separate benefits. Treat time saved separately unless it demonstrably reduces spending or creates additional capacity with a measurable value.

Make the comparison fair

Compare similar pages, audiences and acquisition sources over a suitable period. A campaign launch, seasonal demand or a change in your sales process can affect the result independently of the assistant. Where practical, use an experiment that assigns comparable visitors to different experiences.

Exclude staff tests and obvious automated activity from your decision dataset. Keep small samples visible: a few conversations are useful for finding defects, but they do not establish a reliable conversion uplift. Review failures before increasing distribution.

Run a focused Aivah pilot

Choose one sales question your team repeatedly answers. Configure the role’s knowledge and instructions, define what it must not promise, and test a handoff using your intended destination. Start with a small, reviewed workflow and expand only when the evidence supports it.