A sales AI business case should answer a straightforward question: what would have to improve for this purchase to be worth it?

Start there, not with a vendor's biggest percentage.

A rep finding an answer faster can be useful. That does not automatically mean lower payroll, more pipeline, or a higher win rate. A credible proposal separates what you can observe, what you can reasonably model, and what you still need to prove.

Here is a practical way to build that proposal without turning assumptions into results.

Define the work you are buying back

Choose one workflow before opening a spreadsheet.

For live sales AI, that might be finding approved product answers during technical discovery. It should not be “make every seller more productive.” The narrower version gives you something you can observe and a fair way to compare alternatives.

Write down the current process. Who searches? Where do they look? When does the question become a follow-up? Who checks the answer before it reaches the buyer?

Your baseline should capture the complete effort, including verification and correction. If the assistant makes the first response faster but creates another round of checking, measure both. A confidently wrong answer is not time saved.

Use comparable questions and calls. Note differences in rep experience, question difficulty, source quality, and call type. A practice session with an expert beside the rep is not the same as routine use.

For quality and workflow acceptance criteria, use the 20-question scorecard for a live sales AI pilot. The business case starts after you know what you are measuring; it does not replace that evaluation.

Build the cost side before the benefit side

Use a current quote for the package you would actually buy. Record its date, seat commitment, billing term, usage limits, and anything charged separately. A public starting price is not necessarily your cost.

Then add the work your own team will do.

Initial work: source cleanup, permission review, setup, rep training, internal rehearsal, procurement, and any required security review.

Recurring work: keeping sources current, reviewing answer quality, handling exceptions, supporting reps, administering access, and maintaining the workflow.

Other incremental costs: required equipment, integration work, extra usage, or overlapping subscriptions during a transition, where applicable.

Some of those costs are cash payments. Some are existing employees' time. Show them separately, even when you combine them for an economic comparison. Internal time is not free, but valuing it does not mean payroll will increase by that amount.

Include an alternative. Could a better product FAQ, a narrower training exercise, or a tool you already pay for solve the same problem? Compare the incremental cost and benefit against that alternative, not against an artificially expensive version of doing nothing.

The U.S. Government Accountability Office's cost-estimating guide is a useful methodological reference for documenting assumptions, estimating life-cycle costs, testing sensitivity, and updating estimates with actual costs. It is not evidence of sales AI performance.

Keep three benefit categories separate

Recovered capacity means people have time available for other work. Estimate it from observed net time changes and the number of relevant occasions. Explain what the team plans to do with the recovered time.

Cash savings require an actual spending change. Examples could include retiring a redundant subscription or reducing paid overtime, if those changes are feasible and approved. Recovered employee minutes alone do not establish cash savings.

Revenue impact needs a separate evidence standard. A win after a helpful answer does not show that the assistant caused the win. Buyer fit, pricing, competition, deal stage, and rep experience still matter.

Do not add the same benefit twice. If recovered rep time is valued as capacity and then assumed to create additional deals, you cannot simply add both values without resolving the overlap. Likewise, an avoided follow-up should not be counted once as rep time and again inside a broader time-savings total.

For an initial proposal, it is often more useful to leave revenue upside unpriced than to invent a conversion lift. The UK Government Efficiency Framework provides a useful distinction between cash-releasing savings and monetizable non-cash benefits. The distinction is useful here; its public-sector setting is not evidence of a sales team's returns.

A hypothetical model you can inspect

The following figures are entirely illustrative. They are not Tenali pricing, customer results, benchmarks, or a forecast.

Assume a team buys 20 seats for a fictional tool at $75 per seat per month. Twelve reps use the target workflow on 20 working days each month. For this simplified example, those inputs stay constant for 12 months.

The team uses a hypothetical loaded labor rate of $60 per hour to value internal time. Finance should supply an appropriate rate and explain what it includes in a real model.

First-year economic cost:

• Subscription: 20 seats × $75 × 12 months = $18,000.
• Initial preparation and training: 40 hours × $60 = $2,400.
• Ongoing administration and quality review: 10 hours per month × $60 × 12 = $7,200.
• Total: $27,600, comprising $18,000 of subscription cash cost and $9,600 of valued internal effort.

This illustration includes no other incremental expenses. That is a simplifying assumption, not a recommendation to omit them.

Now test three assumptions about net time recovered per active rep per working day: five, ten, or fifteen minutes. “Net” means after the rep's additional interaction, checking, and correction time. The separate administration allowance above covers different work; do not charge the same review time twice.

• Five-minute case: 12 reps × 20 days × 5 minutes ÷ 60 = 20 hours per month. At $60 per hour, annual modeled capacity value is $14,400.
• Ten-minute case: the same calculation gives 40 hours per month and $28,800 of annual modeled capacity value.
• Fifteen-minute case: 60 hours per month and $43,200 of annual modeled capacity value.

Against the $27,600 first-year economic cost, the modeled capacity-value differences are negative $13,200, positive $1,200, and positive $15,600 respectively.

Those are not cash returns. They show what the chosen assumptions imply when recovered capacity is valued at the stated rate. If nobody can explain how that capacity becomes useful work, the spreadsheet has not established a business benefit.

These cases vary one input to make the arithmetic visible. They are not probabilities, confidence intervals, or a complete risk analysis. A real downside case should also test slower adoption, higher maintenance effort, and a delayed start.

Calculate the threshold, not just the upside

Under those same assumptions, the first-year economic cost averages $2,300 per month. That includes the initial $2,400 spread over 12 months for comparison; it is not the vendor's monthly invoice or the actual timing of cash payments.

At $60 per hour, covering $2,300 in modeled capacity value requires about 38.3 recovered hours per month.

Across 12 active reps and 20 working days, that is about 9.6 net minutes per active rep per day.

Now you have a question worth testing: does this workflow consistently recover roughly that much useful time, at acceptable quality, after the extra work it creates?

Notice how narrow the ten-minute case is. Its modeled annual surplus is only $1,200. A small change in adoption or upkeep could erase it. That is a reason to investigate the inputs, not round the result into a confident ROI claim.

Also test affordability separately. Can the team fund the actual payments when due? An attractive capacity model does not pay an annual invoice.

Replace assumptions as the pilot produces evidence

Keep a short record for every important input: definition, source, owner, observation period, and confidence level.

For time recovery, retain the before-and-after observations and the checks used to make them comparable. For adoption, count use of the target workflow, not just activated accounts. If your average already includes inactive days, do not apply an adoption discount again.

For costs, replace estimated setup hours with actual effort and update recurring maintenance. Include the ramp period rather than assuming full benefits from day one.

For quality, record unsupported answers, missed conditions, unnecessary interruptions, and cases that required escalation. An improvement in speed cannot cancel a serious factual problem. Decide the acceptable boundaries before using the financial model to justify expansion.

If revenue remains part of the hypothesis, keep it in a separate section. Define the outcome, comparable cohort, measurement window, and attribution limits. Any financial treatment should use a finance-approved incremental contribution measure, rather than treating gross bookings as profit.

Take a decision memo to the budget owner

A useful recommendation fits on a page:

• The workflow and team in scope.
• The baseline and the alternative you compared.
• First-year and ongoing costs, with cash and internal effort separated.
• Observed benefits versus assumptions still being tested.
• Low, base, and high cases, with the inputs that change the decision.
• The quality conditions that must hold.
• A named review owner and the evidence required to expand, revise, or stop.

“Continue a limited evaluation because the cost is known but the benefit is still uncertain” can be a sound recommendation. So can deciding that a better-maintained FAQ is enough.

If you are evaluating Tenali, start with the company-knowledge questions your reps actually face during calls. Tenali's live answers use prepared sources and eligible FAQ answers; do not build the business case around an assumed live CRM lookup or a revenue result you have not measured.

Book a Tenali demo with a few recurring buyer questions and the approved material that answers them. Use the demonstration to define a testable workflow, then build the business case from your team's costs and observations.