An AI BDR pilot should tell you whether a defined sales workflow is ready for broader use. The scorecard needs to capture what happened to each account, how much work people still did, and whether sales accepted the result. Email volume alone cannot answer those questions.
Use the worked example below to prepare a pilot specification, test its exceptions, and make an expand, adjust, or stop decision. The company, case records, and outcomes are hypothetical teaching examples, not customer results or tests of a named product.
Fill in the campaign specification
Consider a B2B software team evaluating outbound to operations leaders. Its reps already work from a CRM. The pilot covers one offer and a fixed list of eligible accounts, with messages reviewed before sending.
Its completed specification reads:
- Trigger: an account on the approved pilot list becomes eligible for outreach.
- Eligible work: research the account, prepare a message using approved facts, follow the approved sending process, and route replies.
- Completed outcome: the reply and agreed next step are recorded, with the correct rep assigned. Meeting attendance and qualification are tracked separately.
- Exclusions: customers, active opportunities, suppressed contacts, and accounts with unresolved ownership.
- Process owner: the sales manager, who defines qualification and accepts the result.
- Integration owner: the RevOps lead, who checks permissions, records, and failed updates.
- Exception owner: the account's rep, with the sales manager as backup.
- Limits: the approved account list and a daily sending cap agreed before launch. No expansion of the audience during the test.
Add actual names and a review date in your own version. Choose a run window that allows replies and meetings to mature. If the review arrives before that happens, record pending outcomes rather than counting them as failures or successes.
Establish the human baseline
Review comparable recent campaigns before changing the process. For each eligible account, record time spent researching, drafting, checking, correcting data, and handling replies. Keep time waiting for an approval separate from active work.
Retain difficult cases: missing contact information, conflicting ownership, and replies that need escalation. Removing them would make the pilot look easier than the process it is meant to improve.
Use the same definitions in both periods. Record the offer, audience, campaign dates, and follow-up window. If these change, show the difference beside the results. A smaller or more receptive audience is not evidence that automation caused an improvement.
Turn permissions into testable rules
List what the AI may read and change. For this example, the pilot can use approved account data and product facts, prepare outreach, and update the permitted activity records. It cannot change commercial terms or invent a customer result. A pricing exception must reach the sales manager before anyone makes a commitment.
Artisan's Ava is one implementation option: it supports lead research and enrichment, personalized outreach, reply handling, and meeting booking. Artisan’s campaign documentation describes approval before sending, autonomous replies, escalation instructions, sender routing, and exclusions. Some settings may require a higher plan. Map those controls to the pilot specification; do not treat the existence of a setting as proof that your configuration passes.
This example does not assume an integration between Ava and WIQ. The scorecard evaluates a sales workflow and can be used with whichever system the team has authorized.
Run normal and exception cases
Prepare controlled cases before sending to live prospects. Use test contacts and authorized test accounts for actions that could send a message or write a record. A simulated pass still needs an integration check before live use.
The minimum test set for this example is:
- Eligible account: prepare a factual message, hold it for approval, and preserve the correct owner.
- Duplicate: recognize an account or contact already in the workflow without creating another outreach action.
- Missing information: hold the case when eligibility cannot be established; do not fill the gap with a guess.
- Opt-out: record the request and prevent further outreach under the team's suppression rules.
- Commercial exception: preserve the prospect's question and assign it to the sales manager without promising terms.
Also test an unavailable receiving system and a repeated delivery of the same event. Inspect what was saved and assigned, including any duplicate action. The wording of an escalation is only part of the result.
Record a failed handoff and its retest
Here is a filled scorecard entry for the hypothetical commercial-exception case:
- Case: a prospect asks for a contractual commitment outside the approved offer.
- Expected: no commitment is sent. The conversation, account owner, and question reach the sales manager's review queue.
- Illustrative observation: the outgoing reply is held, but the review item has no assigned owner.
- Result: fail. The message control worked; the handoff did not. The prospect's request could be left unanswered.
- Correction owner: RevOps repairs the routing rule. The sales manager confirms the receiving queue and backup owner.
- Retest: repeat the exception and inspect the saved assignment and notification. Leave the result pending until that evidence exists. Do not turn a planned retest into a pass.
Reuse these fields for every case: input, expected action, observed action, evidence location, result, correction owner, and retest result. Keep failed and abandoned cases in the log. A final average must not conceal an unauthorized send or lost account ownership.
Review quality, effort, and sales outcomes
Review operational results alongside commercial outcomes. Track correct workflow completions out of attempted eligible cases, unauthorized actions, duplicate touches, and unresolved handoffs. Record review and correction time as well as the work the AI performed.
Separately report replies, meetings booked, qualified meetings attended, and opportunities accepted by sales. Define each stage before the pilot. Keep pending meetings visible so the review does not confuse a booking with a completed sales conversation.
For economics, include software, data, setup, review, exception handling, and maintenance within an agreed cost scope. Divide that total by qualified attended meetings only when there are qualifying outcomes. Report spend and zero outcomes otherwise. Do not label this early measure CAC: customer acquisition cost requires acquired customers and the corresponding acquisition costs.
Make a bounded rollout decision
Set thresholds with the process and budget owners before the run. Stop or pause affected actions for unauthorized sends, suppression failures, or lost ownership. Do not let a high overall score cancel those failures.
Expand only the case types that meet the agreed requirements. Adjust a narrow failure when someone owns a testable correction. Stop when the workflow cannot meet the required controls or economics. In the hypothetical failed-handoff case, fix and retest routing before increasing volume.
If evaluating Artisan's AI BDR, bring the completed specification and exception cases to the walkthrough. Ask to inspect the actual message state, receiving queue, and CRM record, then use that evidence in the rollout decision.
