AfieldIn this article
A polished demo can hide the work you will do after the call: remove unsuitable companies, verify a claim, correct a draft and find it again later. Put those tasks in the trial brief.
Give each tool the same assignment
Find five companies that sell B2B software to accounting teams.
Exclude bookkeeping firms, agencies and consumer budgeting apps.
For each company, retain its official domain, buyer audience and one
supporting product-page link. Mark uncertain classifications.
Choose one suitable company and draft a 90-word email offering a
comparison checklist for its buyers. Do not invent a contact or send it.
Save the list and the editable email so another person can reopen them.Replace the audience and offer with your actual business. Give both tools the same approved source material, time limit and output requirements. Record any extra explanation supplied during one trial.
Choose the acceptance checks before the demo
Start with three requirements that cannot be averaged away: the company must fit the requested business type, the email’s factual claims must be supported, and the accepted result must reopen. A quick draft with a false claim does not become usable because the interface earns a high design score.
| Trial step | Evidence to retain | Pass condition |
|---|---|---|
| Review five accounts | Official domains, product-page links and classification notes. | Every accepted account fits; uncertainty is visible instead of guessed. |
| Review one email | The exact draft and the sources for its company-specific claims. | The offer is clear and the claims survive source checking. |
| Make one correction | The instruction and revised version. | The requested change appears without losing unrelated accepted work. |
| Reopen after leaving | A link to the saved task and accepted output. | A colleague can retrieve the correct version without repeating generation. |
If two products have different scopes, keep the shared task small enough that both claim to support it. Do not score a contact database as failing article publication when publication was never part of the purchase. Record that missing capability as separate work your team will need to cover.
A filled score sheet
This is a fictional evaluation example, not a result from testing Afield or another vendor.
| Observed output | Acceptance check | Recorded decision |
|---|---|---|
| A bookkeeping firm appears in the five-company list. | Does it sell software, or provide the excluded service? | Reject the account; ask the tool to apply the existing exclusion. |
| The email says the selected company is entering a new market, but the linked page only lists available countries. | Is a dated expansion announcement present? | Remove the timing claim. Keep the supported product observation. |
| The revised email is useful, but reopening the task shows the old draft. | Can a colleague retrieve the accepted version? | Draft quality passes; recovery fails. Do not score the task as fully usable. |
Use the downloadable trial scorecard for the real observations. Record the correction itself and the minutes spent checking it, rather than assigning a vague “quality: 8/10.”
Test one correction without restarting the demo
Tell the operator: “This company provides bookkeeping; it does not sell software. Replace that record using the original brief, then keep the corrected list.” Ask the next person to open it without help. Watch whether the tool corrects the classification, loses the other records or repeats the original search.
For the email, change one real input—for example, the offer is now an implementation checklist rather than a comparison guide. Inspect the subject, body and closing question. All three should match the new offer.
Measure the operator’s work, including recovery
In this illustrative timing record, generating the first output takes four minutes. Reviewing five accounts takes eight, editing the email takes six, and finding the saved version takes another three. The usable result took 21 elapsed minutes, of which at least 17 involved checking, correction or recovery. Reporting “four minutes” would describe only the first output.
Trial timing record — illustrative
First output available: 4 minutes
Company review: 8 minutes
Email correction: 6 minutes
Reopen and verify: 3 minutes
Total elapsed: 21 minutes
Known checking/recovery: 17 minutesIn a real trial, record active work separately from unattended waiting; do not add overlapping intervals twice. Note which tasks needed specialist judgment. A tool may reduce drafting time while leaving a domain expert with the same amount of claim checking.
Then introduce an ordinary interruption: leave the task, reopen it and ask a teammate to find the accepted email. This exposes a cost that a continuous guided demo often avoids. If the operator has to repeat a search or copy content from a chat transcript into another tool, include that work in the buying decision.
Write the buying decision in one paragraph
We would use this tool for first-pass account research and draft preparation. The trial still required a person to check business type and remove an unsupported expansion claim. Before adopting it for the campaign, we need the accepted list and draft to reopen reliably. Sending and reply ownership remain in our current process.
That fictional decision names the work being purchased and the unresolved requirement. A larger annual commitment should wait until the people using the outputs can complete that requirement in an ordinary trial.