How to choose a CRM: a practical evaluation process

Four numbers decide most of the shortlist, and they take ten minutes to establish.

1. How many people will actually log in and use it weekly? Not how many people are in the business — how many will use it.
2. How many contacts do you have today, and roughly how many will you have in two years?
3. How many open opportunities do you carry at once?
4. How many distinct sales processes do you run? Most small businesses have one, some have two, almost nobody genuinely has five.

These numbers rule products in and out faster than any feature list, because most CRM pricing and most CRM limits are built on exactly these dimensions.

Per-user-per-month pricing is easy to underestimate. Work out the annual figure with your own numbers:

users × monthly price × 12 = annual software cost

Then check four things that routinely change that number:

• Currency. If pricing is in US dollars, your cost moves with the exchange rate and the marketing-site number is never the invoice number.
• Tier jumps. Find the specific limit — users, contacts, pipelines — that would force you onto the next tier, and work out how far away you are from it.
• Add-ons. AI, extra storage, additional pipelines and integrations are sometimes separate line items. Ask explicitly what is not included in the headline price.
• Annual commitment. A discount that requires twelve months up front is a real cost if the product turns out not to fit.

A demo shows you a salesperson's best day in a perfectly-populated system. A trial with your own data shows you the truth. Insist on the second.

Day 1 — get your data in

Import your actual contact list, with all its inconsistency. If the import cannot handle your real file, that is a finding, not a setback. Note how long it took and whether you could see what would be created before it was created.

Day 1 — enter your live opportunities

Every deal you would be upset to lose this quarter. Give each a value, a stage and a next action with a date.

Days 2 to 10 — work from it

Do not configure anything. Log calls as you make them. Time yourself once: how long does it take to record a call and book the follow-up? If it is over thirty seconds, the CRM will not stay current in a busy week.

Day 10 — the handover test

Pick a customer a colleague has never dealt with. Ask them to open the record and tell you, in one minute, what is going on and what should happen next. If they can, the product works. If they have to ask you, the history is not surfaced well enough.

Day 12 — the phone test

Try to update a deal and log a call from your phone, standing up, on mobile data. This is where the majority of follow-ups genuinely get recorded, and a lot of CRMs are unusable in exactly this situation.

Ask these directly. The quality of the answer tells you as much as the answer itself.

• What does your product not do, that customers regularly ask for? A vendor who cannot name three things is not being straight with you.
• What can your AI do without a human approving it? Narrow answers indicate careful design; broad ones indicate risk you will carry.
• Where is our data stored, specifically, and who are your sub-processors? A good vendor has a published list. A vague answer is itself the answer.
• How do I get my data out, and in what format? Test this during the trial, not when you are leaving.
• Which security certifications do you actually hold? Watch for implied certification — "enterprise-grade", "bank-level" — which means none.
• What happens to my data at the end of a trial, or if I stop paying? Deleted, retained, or read-only?
• Is your price in my currency, and is it fixed for the term?

Most feature comparison is noise. These few genuinely differentiate products for a small business:

• Are companies a real record type, or a text field on a contact? This decides whether the relationship survives your contact changing jobs.
• Is the history one thread, or scattered across tabs by type? One thread is dramatically more useful in the sixty seconds before a call.
• Can you see, without running a report, which deals have no next action booked? This is the single highest-value signal in a CRM.
• Does closing a deal capture why you lost? Without it, win/loss reporting is a number with no lesson in it.
• Are permissions enforced by the server, or just hidden in the interface? Ask them to state which.
• Can you change stage names later without breaking your historical reporting?

At the end of two weeks, answer three questions honestly.

1. Did the specific failure you wrote down in step one get harder to make? Not impossible — harder.
2. Did logging feel cheap enough that you kept doing it on the busy days?
3. Could a colleague read a customer record and pick up the relationship?

Three yeses and the tool fits. Two and it is probably a habit problem rather than a product problem — try again with a smaller scope. One or none, and you have saved yourself a year of paying for something nobody opens.