Should you talk to an AI advisor before buying an AI tool?

By KYN AI Advisory Team — AI implementation specialists, Singapore

Yes, if you cannot yet state the problem in one sentence or your systems do not share data. A tool bought before that point tends to add a login without removing any work. A short advisory conversation settles three things first: the problem, the data it depends on, and whether you need a new tool at all. If you can name the problem and one tool clearly fits, buy the tool.

Disclosure: KYN builds custom systems, so we have an interest in this question. The checks below work with any adviser, including ones who sell nothing.

When is buying a tool without advice fine?

Buy directly when all four are true:

The problem is one task, such as transcribing meetings or drafting first copy.
The tool works on its own, without reading your other systems.
The data is not sensitive, or the vendor's terms cover it.
You can cancel within a month with nothing lost.

An AI writing aid for one person fits. An agent that reads your bank feed and CRM does not.

When does advice earn its cost?

Two situations account for most wasted spend.

Misaligned tools. The real issue is that systems are not connected, data is not structured for reporting, or the workflow is unclear. A new platform leaves those untouched and adds one more place to look.
No problem statement. "We should automate more" cannot be scoped, priced or tested. Teams then buy for whichever department shouted loudest.

Both come from starting at the tool. Start at the workflow, the data and the decision it feeds.

What does an advisory conversation cover?

A first conversation follows four parts.

Context and goals. Your model, your products, and why you are looking now.
Systems and data. Accounting platform, ERP, CRM, e-commerce, bank feeds and any portals. How data moves between them, and where someone retypes it.
Pain points. Manual reconciliation, slow reports, scattered customer data, inconsistent follow-up.
Outcomes. What would count as success, stated as a number a manager can check.

You leave with a short list of defined problems and a view of which existing systems can be reused. Problems read like "daily report from five systems" or "capture and follow up leads from email and WhatsApp", not "adopt AI".

How do you turn "we need AI" into a problem statement?

Fill in this sentence: "[Role] spends [hours per week] on [task] across [systems], and the result is [error or delay]."

Then test it:

Can you measure it today, even roughly?
Does it repeat weekly or more often?
Is there a right answer a person can check?
What happens if the automation is wrong, and who catches it?

A problem that fails the third test is a poor first project. A problem that passes all four is a candidate.

What questions should you ask any adviser?

| Question | A good answer | A warning sign | |---|---|---| | What would you build first, and why? | A narrow, checkable task tied to a number | A platform-wide roll-out | | Which of my existing systems stay? | Most of them | Replace everything | | Where does the system run, and who owns it? | In an account in your name | The vendor's account | | What happens if I stop paying? | Everything keeps running | Access ends | | What will a person approve? | Outbound messages, money, filings, deletions | Fully autonomous | | How is each action recorded? | Logged with the actor | Not logged |

Also ask to see something live from a client in your industry, described by industry if the client is confidential.

Which problems tend to come first?

Finance. Reconciliation against bank feeds, bill capture from email, receivables chasing at set intervals.
Operations. A morning report that joins production, stock and cash from systems already in use.
Sales. One inbox for email and WhatsApp, with replies sorted and logged.
Marketing. Content drafted from checked facts, with a person approving publication.

Each sits on data you already hold. That is why they come first.

Do government and legacy systems change the answer?

Yes. Where you file through a government portal or run an older ERP, the hard part is data correctness and integration, not the AI. Reference data has to match the external system's format before any transaction is accepted, and a vendor sandbox will not show you the live rejections. Ask any adviser how they test: against real historical records, in a separate copy, before anything writes back.

What does a bad purchase cost?

The figures below are illustrative. Replace them with your own.

A 30-person company buys an AI sales platform at an assumed S$150 per user per month for 10 users. It signs a 12-month contract.

| Cost item | Amount | |---|---| | Licences, 10 users x S$150 x 12 months | S$18,000 | | Setup and data import, internal time at 40 hours x S$50 | S$2,000 | | Training, 10 users x 3 hours x S$50 | S$1,500 | | Parallel work while staff keep their old spreadsheet | S$3,000 | | Total in year one | S$24,500 |

If the tool saves nothing because enquiries still arrive on WhatsApp and the tool reads only email, the company has paid S$24,500 for a login. The cost is not the licence alone. It is the licence plus the hours spent finding out. A two-hour advisory call before signing costs a small fraction of that and would have raised the WhatsApp question on day one.

The same logic holds for custom builds. A build started before the problem is stated costs more, because the scope shifts as the problem becomes clear.

A worked first conversation

This example is illustrative. The company and the answers are invented.

A distributor says: "We want AI for customer service." The adviser asks four questions and gets these answers.

| Question | Answer | |---|---| | What happens today when a customer asks about an order? | Staff check three places: the ERP, a shared spreadsheet and the courier site | | How many such questions a week? | About 120, mostly "where is my order" | | What does a reply take? | Around 4 minutes, because of the three lookups | | What goes wrong? | Different staff give different answers; some replies take a day |

The adviser restates the problem: "Two staff spend about 8 hours a week answering order-status questions across three systems, and answers are inconsistent."

That sentence changes the shape of the project. The need is not a chatbot. It is a single order-status view that joins three sources, plus drafted replies that a person approves. The adviser also notes what to avoid: do not replace the ERP, and do not let the system reply to customers without review at first.

The conversation produced a problem, a data map and a boundary in under an hour. A tool vendor, asked "do you do customer service AI?", would have said yes and quoted a platform.

What should you expect to leave with?

Ask for a written output, not a promise to follow up. A useful first review produces five things.

A problem statement for the top candidate, in one sentence with numbers.
A systems map showing where the data lives and where people retype it.
A ranked list of the options, with the scoring visible.
A recommendation that says build, buy or wait, with the reason.
A scope outline for the first project, covering what a person approves, what the system may do alone and how success is measured.

If you receive a slide deck of general benefits and no scope, the advice has not been given.

How do you tell whether the advice is independent?

Every adviser has an interest. The test is whether the interest changes the advice.

Ask what they would recommend if you did not hire them. A fair answer names a package or a do-nothing option when it fits.
Check whether the recommendation matches the problem. A recommendation that mentions a particular product before the problem is clear is a sales pitch.
Look for the "no" answers. An adviser who has never told a client to wait or buy off the shelf is selling.
Ask who pays them. A fee from the vendor you are about to buy is a conflict, and you should be told about it.
Get the recommendation in writing, so you can show it to a second adviser.

A second opinion costs little. If two advisers agree on the problem and differ on the solution, you have learned where the real decision lies.

How should you compare vendors after the advice?

Use a scorecard with the same rows for each vendor. The weights below are illustrative; set your own.

| Criterion | Weight (of 20) | Vendor A | Vendor B | Build option | |---|---|---|---|---| | Covers the top problem without workarounds | 6 | | | | | Reads and writes the systems you already use | 4 | | | | | Data and code stay in an account in your name | 3 | | | | | Total cost over three years | 3 | | | | | Approval steps for outbound, money, filings, deletions | 2 | | | | | Exit terms and data export | 2 | | | |

Fill it in with evidence: a live demonstration on your data, a written term, a quote. Leave a cell blank where you have none. A blank cell is a question for the vendor, not a score of zero or five.

Which contract terms matter most?

Read these before you sign anything, whether it is a subscription or a build agreement.

Term and renewal. Does the contract renew automatically, and with what notice?
Price changes. Can the price rise mid-term, and by how much?
Data ownership. Who owns the data the system produces, and can you export all of it?
Code ownership. For a build, when do rights pass to you?
Access to the cloud account. Is the account in your name?
Subprocessors. Which companies handle your data, and where?
Service levels. What response time applies to a failure, and what is the remedy?
Termination. What happens to the system and the data if you stop paying?

Ask a lawyer to read anything with a term longer than twelve months or a clause on data transfer. For personal data, check the rules in each country where your customers sit.

What security questions should you ask?

You do not need to be technical to ask these. A vendor who cannot answer them in plain words has told you something.

Does each person have a named account, or do people share a login?
Can one user see another customer's data? How is that prevented?
What can the AI parts of the system do alone, and what waits for a person?
Is every action recorded with who or what did it?
Where is the data stored, and in whose account?
How do you revoke access when someone leaves?
What happens if a model provider has an outage?

A system that sends messages, moves money or deletes data without a person's approval carries a different risk from one that drafts for review. Be clear which one you are buying.

What are the warning signs in a sales conversation?

The vendor names a product before asking about your problem.
The demo uses invented data, not a sample of yours.
The quote arrives before anyone has looked at your systems.
The proposal replaces systems that work.
Results are described as "up to" with no method or source.
The contract is long and the exit terms are short.
Nobody can name a customer in your industry, even by description.

One sign is a prompt to ask. Three signs are a reason to walk away.

How do you prepare so the first conversation is short?

An hour is enough if you bring the right material. Prepare before the call.

A list of your systems, with who uses each.
One report you trust and one you do not.
Three tasks that repeat weekly and waste time, with a rough count of hours.
A note on anything sensitive: customer data, health data, payment details.
The name of the person who will own the result.
Your budget range for the first project, or the reason you have none.

If you cannot fill the list, the first step is to build it. That work is the advice, and you can do it yourself in a week.

What happens after the advice?

You have four options, and any is acceptable.

Do nothing yet. The problem is not costly enough or the data is not ready.
Fix the data. Clean fields, agree formats and remove duplicates. Often the right next step.
Buy a tool. One fits the problem and the term is short.
Build. No tool fits, or the system must run in your own account.

Whichever you choose, set a date to review it. Review against the number from the problem statement. If the number moved, continue. If it did not, find out why before spending more.

Three illustrative scenarios

These scenarios are invented. Each shows a different outcome of the same first conversation.

Scenario 1: buy a tool. A consultant wants first drafts of proposals. The task is one person's, the data is not sensitive, and the tool works alone. The adviser says to buy a writing aid on a monthly term, test it for a month and compare drafting time before and after. No build is justified.

Scenario 2: fix the data first. A retailer wants an AI report on customer value. The adviser samples the customer list and finds the same customer under four spellings, with phone numbers in three formats. Any report would double count. The recommendation is to agree the formats, merge duplicates with a person's approval and then revisit. The saving from this step alone is cleaner mailing lists and one customer view.

Scenario 3: build a layer. A manufacturer has stock in one system, production in a spreadsheet and cash in an accounting package. No tool joins them, and the owner wants a morning report in WhatsApp. The adviser recommends a custom layer that reads all three and sends a short report to each department, with changes to production data waiting for a named person.

The three scenarios share one feature: the advice followed the problem and the data, not a preference for a product.

What myths lead people to skip advice?

"AI tools are plug and play." Some are. Those that read your systems need setup, data work and a review process.
"Advice is just a sales call." Some is. Ask for a written recommendation and a second opinion to tell the difference.
"We are too small." Small companies have fewer people to absorb a wasted subscription, which makes a wrong purchase more costly in proportion.
"We need a strategy first." A strategy document is not required. A one-sentence problem and a data map are.
"The vendor will help us figure it out." The vendor knows the product. They have less reason to say it does not fit.

Does funding change the order of steps?

Funding schemes in Singapore may offset part of the cost of a qualifying project, but they should not decide the project. Pick the problem first. Then check whether a scheme covers it. See grants for a starting point, and confirm current terms on the official agency page before you rely on any scheme, because eligibility and rates change.

Two cautions apply. First, some schemes require that you apply before work starts, so ask before you sign. Second, a scheme that covers a particular product can push you toward that product. Compare it with the alternative on the scorecard above, with the subsidy included, over three years.

How do you review the advice after three months?

Set a date in the calendar when the advice is given. At the review, check four things:

Did the number from the problem statement move?
Did the cost match the scope outline?
Did the people who use the result keep using it?
Did anything the system did alone need to be undone?

If the answers are mixed, write down what to change. If they are poor, ask the adviser to explain why before any further spending. Advice that cannot survive a review against its own numbers was not advice.

Where KYN fits

KYN is a Singapore company that builds custom systems for SMEs in Singapore, Hong Kong, Indonesia and Malaysia. Systems run in a cloud account in the client's name. Rights in the custom code and configuration pass to you on full payment, and you can revoke KYN's role while everything keeps running.

The first 14 days are a review: an inventory of your stack and interviews with your team. A full build ships in under 90 days. Pricing is a one-off build fee plus a monthly retainer of at least twelve months, quoted after the review. See pricing and case studies.

Frequently asked questions

Do I need an adviser if I only want one AI tool?

No. If the tool does one task alone, try it on a month's notice. Advice pays off when the tool must read your accounting, CRM or bank data.

Is advice free?

It varies. Some vendors give a free call and expect a sale. Ask up front whether the review is paid, and what you keep if you do not proceed.

What should I bring to the first conversation?

A list of the systems you use, one example of a task that wastes time each week, and the person who owns it. A recent report you distrust is useful too.

Checked 30 October 2026.

Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.

Start Building →