[ 01 ] Free reference card PDF · A4 + Letter

Twelve questions
before you pay for AI in any CRM.

A vendor-neutral printable for the 1 to 50 person team evaluating an AI tier this quarter. The questions a buyer should bring into the demo, not the ones the deck answers on its own.

Get the printable checklist

4 pages · printable Vendor-neutral No phone, no upsell

syncek.com / resources / ai-crm-checklist.pdf PDF

Reference · v1 · 2026

12 questions. One good-answer pattern. One red flag. Per row.

[ 01 ]Measurable outcome

[ 02 ]Off-switch

[ 03 ]Your data shape

[ 04 ]Native vs. bolt-on

[ 05 ]Data residency

[ 06 ]Prompt or rule

[ 07 ]Price at GA

[ 08 ]Output ownership

[ 09 ]When AI is wrong

[ 10 ]Human approval

[ 11 ]Accuracy evidence

[ 12 ]Exit cost

Question · Good answer · Red flag Sheet · 1 of 4

The recurring quote

“The AI demo only worked because the vendor seeded the data.”

Every operator evaluating a CRM in 2026 has sat through three demos with AI in the pitch. Each one ran on clean, complete, synthetic data the vendor controlled. Then the trial started, the real CSV got imported, and the AI surface that summarized deals beautifully in the demo went quiet, or worse, started hallucinating fields.

The twelve questions below close the gap between demo and Tuesday morning. Take them into every vendor call this quarter.

[ 02 ] The twelve at a glance

Four risks, twelve questions, four pages.

The questions group around four risks an SMB operator carries home from a pitch: feature risk (will the AI do what was demoed), data risk (will it work on your data, where does your data go), price risk (what happens at GA), exit risk (what comes with you if you cancel). The full reference card pairs each question with what a good answer sounds like and what a pitch sounds like.

[ 01 ] Feature risk

What measurable outcome does it move?

The vendor should name a metric and a direction. Hours saved per week, deals closed, time to first reply. Not a feeling.

[ 02 ] Feature risk

Can you turn it off without breaking the CRM?

If disabling the AI breaks the table, the pipeline, or the import, the AI was the product. You are buying a chat surface, not a CRM.

[ 03 ] Data risk

Will it run on your actual CSV?

Demo data is not your data. Ask the vendor to run the feature on a CSV you export from your current tool, before you pay.

[ 04 ] Feature risk

Native to the table, or bolted-on chat?

An AI feature inside the table you already use is a utility. A separate AI chat tab next to the CRM is two products bundled to defend a price.

[ 05 ] Data risk

Where does your data go when the AI runs?

Vendor servers, third-party LLM provider, EU region only. The answer should be specific, not “industry-standard security.”

[ 06 ] Feature risk

Deterministic rule, or fresh prompt every time?

A rule returns the same answer on the same input. A prompt does not. Both can be useful. The vendor should be honest about which one you are buying.

[ 07 ] Price risk

What happens to the price at GA?

Beta-tier AI almost always moves up at general availability. Ask for the ceiling in writing, and ask what happens to your renewal if it crosses it.

[ 08 ] Exit risk

Does the AI's output export with your data?

Summaries, scores, suggested fields. Your data, generated by the AI, on your records. If the export drops them, you are renting the output, not owning it.

[ 09 ] Feature risk

What is the recovery path when AI is wrong?

Edit in place, undo on the record, audit log, version history. There should be a clear way back. “The model rarely makes mistakes” is not an answer.

[ 10 ] Feature risk

Does a human approve every write?

Autonomous AI writes to your records are silent failures waiting to happen. A human-in-the-loop is not a workflow tax. It is the audit trail.

[ 11 ] Feature risk

How does the vendor measure accuracy?

A named test set, an accepted ground truth, a percentage. If the only evidence is “customers love it,” the vendor has not evaluated.

[ 12 ] Exit risk

What is the exit cost after six months?

CSV export, JSON export, AI outputs included. If any of those is hostile, the AI tier was the lock-in, not the feature.

[ 03 ] Question by question

What a good answer sounds like. What a pitch sounds like.

Each question gets a one-line good-answer pattern (what a real product sounds like in a vendor call) and a one-line red-flag pattern (what a pitch sounds like, even when the words are confident). The printable card carries the same three columns.

Q	Ask the vendor	Good answer	Red flag
[ 01 ]	Measurable outcome What metric does this AI feature move, and in which direction?	Good answer“Reply-time on new inbound, from 18 hours median to 4. Measured on N=240 accounts. Here is the dataset.”	Red flag“It makes your team more productive across the entire customer lifecycle.”
[ 02 ]	Off-switch Can I disable the AI feature without breaking the table, pipeline, or import?	Good answer“Yes. Workspace setting. The table, pipeline, and import behave identically with it off. Existing AI outputs stay on the records, you can delete or keep them.”	Red flag“Nobody turns it off, it is the main reason teams choose us.”
[ 03 ]	Your data shape Can you run the feature on a CSV I export from my current tool, live in this call?	Good answer“Yes. Drop the file in the trial workspace, give it three minutes, share your screen.”	Red flag“Our demo data shows it better. We can schedule a custom POC with our solutions team.”
[ 04 ]	Native vs. bolt-on Does the AI live inside the table and record pages I already use, or in a separate chat surface?	Good answer“Inside the record. Suggested-field chip in the sidebar, click to accept or dismiss, written to the field on accept.”	Red flag“Open the AI panel and ask anything about your data.”
[ 05 ]	Data residency Where is my data processed when the AI runs? Which providers, which regions?	Good answer“Anthropic via AWS Frankfurt, EU region only. Logged in our audit trail. No training on your data, contractual.”	Red flag“We use industry-standard encryption and best-in-class security.”
[ 06 ]	Prompt or rule Is this a deterministic rule that returns the same answer on the same input, or a fresh prompt every time?	Good answer“Prompt. Outputs vary run to run. We log every call so you can audit drift, and we surface the diff.”	Red flag“Our AI is consistent because of our proprietary model architecture.”
[ 07 ]	Price at GA What is the price ceiling when the AI moves out of beta? What does that do to my renewal?	Good answer“Ceiling of +€12 per user per month at GA, written into the order form. Twelve-month grandfather on current pricing for beta accounts.”	Red flag“We are still finalizing GA pricing. It will be competitive.”
[ 08 ]	Output ownership When the AI writes a summary or score into my records, does that exit with my CSV export?	Good answer“Yes. AI-generated values live on the field, exported in the same row as the rest. CSV and JSON both.”	Red flag“AI insights are a premium feature, surfaced through our UI.”
[ 09 ]	When AI is wrong When the AI writes a wrong value, what is the recovery path for the operator who finds it?	Good answer“Edit in place, the field reverts. Audit log keeps both versions and the timestamp. One-click revert on the record.”	Red flag“Our model accuracy is very high, false positives are rare.”
[ 10 ]	Human approval Does the AI write to my records autonomously, or does a human approve each write?	Good answer“Suggested by default, written on accept. Auto-write is opt-in per field, off by default.”	Red flag“The AI is autonomous, that is the value.”
[ 11 ]	Accuracy evidence How do you measure accuracy? Can I see the test set and the percentage?	Good answer“1,200-row held-out test set, ground-truth labeled by three reviewers, 87% agreement with reviewer consensus. Updated quarterly.”	Red flag“Our customers tell us it is highly accurate.”
[ 12 ]	Exit cost If I cancel after six months of using the AI tier, what comes with me, and what stays behind?	Good answer“Everything. CSV and JSON export include all AI-generated fields. Thirty-day grace window after cancellation to re-export.”	Red flag“Your records export as expected, AI insights are part of the premium tier.”

[ 04 ] Download · free PDF

Get the printable checklist.

Four pages, A4 and Letter. Cover, risk map, the twelve questions across two facing pages, and a marking sheet you keep for the buying record. Print it, take it into the next vendor call, hand a copy to whoever sits in the demo with you.

Download the PDF

The checklist is on its way. Check your inbox in a minute. If it does not arrive, look in spam or write to hello@syncek.com.

Reference card · v1 · 2026 1 / 4

AI in a CRM, a buyer's checklist

Twelve questions. One good-answer pattern. One red flag. Per row.

Ask

Good answer / red flag

[ 01 ]

Measurable outcome

A named metric and a number.

[ 02 ]

Off-switch

Disable, table still works.

[ 03 ]

Your data shape

Live run on your CSV.

[ 04 ]

Native vs. bolt-on

Inside the record, not a chat tab.

[ 05 ]

Data residency

Provider and region, named.

[ 06 ]

Prompt or rule

Honest about which one.

Vendor-neutral · printable syncek.com/resources/ai-crm-checklist

A4 + Letter

[ 05 ] How to use the card

A buyer's checklist, not a vendor scorecard.

The card belongs to the buyer, not the procurement team. Print it once, run it across every vendor call this quarter, mark the answers as they come.

[ 01 ]

Print one copy per vendor you are seriously evaluating.

Two to four vendors is the typical SMB shortlist. The card runs four pages, light enough to staple and take into the demo. Write the vendor name on the cover, the date in the corner, keep the marking sheet on top.
[ 02 ]

Ask the twelve in order. Mark the answer next to each question.

Use three marks. Green for an answer that looks like the good-answer pattern. Amber for a hedge. Red for the red-flag pattern. Twelve marks per vendor, twenty minutes per demo.
[ 03 ]

Tally the marks by risk family.

Three reds in feature risk and one in exit risk is a different problem than four reds spread across data and price. The card groups questions by risk family so the pattern shows up at a glance.
[ 04 ]

Ask for the live CSV run before you sign.

Question 03 is the single highest-signal question on the card. If the vendor refuses to run the AI on your actual data live in a call, the AI is not ready for your data. Mark, move on.
[ 05 ]

Keep the marked cards in the record.

Six months from now, when the renewal arrives or you are talking to a successor vendor, the marked cards tell you why you said yes. They are the audit trail for the buying decision.

[ 06 ] The companion cards

Fields, stages, hygiene. Then the AI tier.

Three earlier cards cover the CRM you already run. Fourteen fields for who the client is. Five stages for where the deal is. A monthly calendar for keeping the records honest. This card covers the next decision: which CRM (and which AI tier inside it) you should be paying for at all.

See all the cards

[ 07 ] FAQ

The questions we get most.

Short answers. The long-form article covers the rest.

What should I ask a CRM vendor about AI?

Twelve questions cover the gap between a demo and Tuesday morning: what outcome the AI measurably moves, whether you can turn it off, whether it works on your data (not the vendor's seeded sample), whether it lives inside the table and pipeline you already use, where your data is processed, whether the AI is deterministic or a prompt every time, whether the price changes at GA, who owns the AI's output, what happens when the AI is wrong, whether a human approves writes, how the vendor proves accuracy, and what comes with you if you cancel.

The printable card pairs each question with what a good answer sounds like and what a pitch sounds like.

Is this checklist tied to Syncek?

No. The twelve questions are vendor-neutral by design. The same checklist holds whether you are evaluating HubSpot, Pipedrive, Attio, Notion with an AI add-on, Salesforce Einstein, or any other CRM with an AI tier.

Syncek itself does not ship AI features today. The checklist could not be a self-promotion if it tried. It is written from the buyer's seat, because that is the seat we sat in when we evaluated CRMs in 2024 and could not find a list like this.

Who is the checklist for?

Operators at 1 to 50 person teams who own the client book and are actively evaluating a CRM with AI in the pitch. Founders, agency owners, ops leads, freelancers.

If you have a fifty-person sales floor and a dedicated CRM admin, this checklist is below your floor. It is for the operator who already wears five hats and needs a reference, not a procurement project.

Why twelve questions, not ten or twenty?

Twelve is the band where every question earns its place and no important one is missing. Below ten, you skip pricing-at-GA or fallback-when-wrong and pay for it later. Above twenty, the buyer never gets through the list and the checklist becomes paper.

Twelve fits on four printable pages (cover, risk map, the questions, marking sheet), takes about twenty minutes in a vendor call, and covers the four real risks: feature risk, data risk, price risk, exit risk.

Will I be added to a sales sequence?

No. Email is used to send the PDF and the occasional Syncek update. No phone, no company size, no drip nurture, no just-checking-in follow-ups. You land on the closed-beta nurture list, segment ai-crm-checklist-EN. Unsubscribe is one click.

Honest pricing. No per-seat surprises. Same posture on the email list.

What format is the download?

A single printable PDF, A4 and US Letter sized so the same file works on either side of the Atlantic. Three columns. Question. Good-answer pattern. Red-flag pattern. Numbers in the brackets.

The English and Spanish versions are separate files. Tick whichever boxes you want on the form.

Does Syncek have AI features?

Not today. AI features are on the roadmap and will live as utilities inside the existing table and pipeline UX, never as a copilot layer that replaces the product.

The checklist is written from the buyer's seat. When AI utilities ship in Syncek, they will be evaluated against this same card before they go live.

How does this connect to the other cards?

The 14-fields card tells you who the client is. The 5-stages card tells you where the deal is. The data-hygiene calendar tells you how to keep the records honest. This checklist tells you which CRM (and which AI tier inside it) you should be paying for at all.

Same buyer, same posture, four different design problems. See all the cards.

Where can I read the long-form article?

The full article, with one question per section, the rationale, the good-answer pattern, the red-flag pattern, and an example transcript of how the question plays out in a vendor call, is at syncek.com/blog/ai-in-a-crm-buyers-checklist-for-smbs. The card is the printable. The article is the reference.

[ 08 ] One last thing

The AI demo is not the product. Your CSV is.

Twelve questions, four printable pages, twenty minutes per vendor call. The buyer walks in with the questions; the product answers them; the contract reflects the answers. That is the entire posture.