How to tell if your recruitment CRM is still being built
Every recruitment CRM looks capable on a demo, because a demo shows the parts that were finished. Five questions tell you whether anyone is still working on it, and we answer all five about ourselves with numbers you can check.
Ask AI about this
Recruitment software does not fail loudly. It sits there doing what it did last year. The release notes get thinner and then quarterly. Support starts answering questions with that is how it works rather than we will fix it. One day you realise the thing you run your agency on has not meaningfully changed since you bought it, and by then you are three years of data deep and switching feels worse than staying.
That is the trap, and almost nobody walks into it carelessly. They walk into it because the software was genuinely good on the day they chose it. Nothing about a demo tells you whether the company behind it will still be building in year three, and year three is where you will spend most of your time with it.
What you actually want to know is a question about the last few weeks, not the last few years. Here are five that get you the answer, with our own from the week of 13 to 20 September 2026.
1. How often does the vendor actually ship?
Ask for the changelog and look at the dates. Not the roadmap. A roadmap is a list of things somebody intends to do, arranged by quarter, containing no dates that have already happened. A changelog is a list of things that were done. Every agency I know that ended up trapped got there by reading roadmaps, and the roadmap was sincere at the time.
A product with a real team behind it changes every week, in small pieces, in public. A product being run for cash changes twice a year and calls it a major release, because a major release is the only way to make a thin year look deliberate.
Ours. In that week we made 537 changes across 1,778 files, from seventeen people. Every one of them is written up before it lands, saying what it does and how it was checked, which is the only reason there is a number at all rather than an impression. That week produced 160 of those write-ups.
What a bad answer means for you. If a vendor cannot tell you what shipped last week, the gaps you can see in the product today are the gaps you will have in three years. Nothing about your contract changes that, and no amount of account management does either.
2. Can they change one thing everywhere at once?
You can ask this one without knowing anything technical, and it is the question that separates a real system from a pile of screens.
Think of the last small change you asked your vendor for, the kind that would save your team ten minutes a day, and remember how long it took. The reason it took a quarter is almost never the change itself. It is that the same thing is written by hand into two hundred separate screens, so a small change is two hundred small changes. You pay that tax on every request you will ever make, and it is why your sensible suggestions keep being described as bigger than they look.
Ours. We changed our brand colour twice in the same week, and it took one edit. One edit, because the colour lives in a single place that every screen reads from, and because 292 hand-written copies of it were tracked down and pointed at that source first. The build now refuses to run if anyone pastes a loose copy back in, which is the part that keeps it true next year when whoever set it up has moved on.
How to test it without taking anyone's word. Look at the product for screens that were obviously designed in different eras. Corners that are rounded on one screen and square on the next, two different shades of the brand colour, headings in three weights. That is not a cosmetic problem. It is the visible symptom of a company that can no longer change one thing everywhere, and everything you ask them for will be priced accordingly.
3. How fast do they get new AI into production?
Everyone in this market says they are an AI company. The test is not whether they have AI. It is how long it takes them to use something that did not exist last month, because that gap is what you will be living in for the next five years.
The gap also compounds, which is the part buyers underestimate. A vendor who takes eighteen months to adopt something is not eighteen months behind once. They are eighteen months behind on everything that arrives from now on, for as long as you are a customer, and the distance grows every year rather than closing.
Ours. TypeSafe released Jev on 18 September 2026. It is not a chatbot: it answers a typed question about a situation and tells you how sure it is, which is exactly what software needs when it has to choose between real options rather than write a paragraph. We had it running against live customer work the same day. By the end of that week it was sorting every reply to your outreach so the interested ones sit at the top, reading the tone of your calls and meetings, and filling seven fields on a job from your own lists rather than guessing at names. Hats off to the TypeSafe team.
One day was possible because the machinery it plugged into was built long before Jev existed. That is the part worth asking any vendor about, because speed on its own is recklessness. Nothing went live until a battery of 127 cases passed against the real model, including the cases that exist to prove the software refuses to guess.
How to test it. Name the newest thing in the market and ask when they will have it. You are listening for whether the answer is a date or a category.
4. Can they prove where the AI is, or only claim it?
Your candidates' details go through this software, and somewhere inside it a model is reading them. Ask the vendor to tell you every place that happens.
Most cannot, because nobody ever counted. A list somebody typed two years ago is a memory rather than a count, and it goes stale the week after it is written: every new feature adds to it and nobody updates the document. The revealing answer is not I will have to check, which is honest. It is a confident number that turns out to come from a slide.
Ours. It is counted by a program that reads every line of code we have written, here and in the sixty other systems we run, and it stops and complains if it finds a file that touches AI and does not match anything it recognises. The last count was 433 places across 221 files with nothing unexplained. The first time we ran it, it found more than forty places our hand-written list had quietly missed, which is not a flattering thing to publish and is the entire reason we stopped keeping one by hand.
Why this is about to become your problem rather than theirs. Where candidate data goes is already appearing in enterprise supplier questionnaires and PSL renewals. When a client asks you, the answer has to come from your CRM vendor. If they will not give it to you, you are the one explaining that to your client. A vendor who cannot say where their AI runs also cannot tell you what it does with a CV, and cannot tell you what quietly starts behaving differently the next time a model is updated underneath them.
5. Do they fix the small things nobody demos?
This is the question that decides whether your consultants like the software, and it never comes up in a sales conversation. Nobody demos a menu closing properly. You only meet it at four o'clock on a Thursday when it sits on top of the thing you were trying to read.
Ours, from the same week. The kebab menu on a contact now closes when you pick something, so it stops covering the window it just opened. Selecting every row in a list no longer loses your count when more rows load underneath. Deleting a job takes you to the next job instead of dumping you back at the grid. Adding a contact as a candidate carries their title, date of birth, skills and referrer across so you do not type them again. And a window opened by another window now has a way back to where you came from.
Eighty-seven of the 160 items that week were that kind. It is the least glamorous bar on the chart, it is not a flattering thing to publish, and it is the number I would look at first in anybody's changelog. A changelog that is all features is either a very new product or a partial view, because more than half the work of keeping a mature product good is things nobody asked for and nobody thanks you for.
Why it decides more than it looks. Small frictions are not annoyances, they are behaviour change. A consultant under pressure skips the step that is awkward. The call gets made and the note never gets written. The candidate gets moved in somebody's head and not in the pipeline. Six months later your database is a partial record of what your agency actually did, every report built on it is wrong in a direction nobody can measure, and an agency whose consultants keep private spreadsheets is not an agency with bad consultants.
What a no costs you
| The question | What a bad answer means for you |
|---|---|
| How often do you ship? | The gaps in the product today are the gaps in three years |
| Can you change one thing everywhere? | Everything you ask for is quoted in months, so you stop asking |
| How fast does new AI reach us? | You keep whatever the market had in the year you signed |
| Where is AI used in your product? | You cannot answer your own clients' questionnaires |
| Who fixes the small things? | Your consultants work around the software instead of using it |
Notice that none of those costs arrive as an incident. There is never a day when the decision to leave becomes obvious, which is precisely why agencies stay years longer than they should. The software still opens. Support still answers. Nothing is broken. It has simply stopped moving while your business did not.
Ask us the same five
Our changelog is public and dated, our engineering write-ups name the people who did the work, and every number in this piece came out of those same records rather than out of a marketing meeting. Ask any vendor you are talking to for the equivalent and notice how few of them have it.
And ask us. If our answers are not better than the alternatives on your shortlist, buy the alternative, because a wrong answer to any of these five is something you will live inside for the next five years and no discount compensates for it.
I have watched this industry ship software that stopped moving years ago and still charge per seat for it. That is why I say Recruitly is the best recruiting CRM in the world, and the week above is the only argument I want to make for it.
Gowri is the CEO of Recruitly. The figures in this piece cover 13 to 20 September 2026 and come from our own change records.
