AI-aided implementation

AI is a tool, not the thing.

AI can now build Zoho apps, automations and integrations in days. What it can’t do is decide what your business needs. That part still starts with Discovery.

Whether we build it, or you do. See what we built this way

What the AI needs to know
How work flows, and where approvals sit
​
What your data means, and which system holds the truth
​
Who may see and change what
​
What must never break, and what “done” means
Where it comes from
Discovery, before the build starts.
The problem

The same old problems, built faster.

More of our customers are using AI to build their own apps and automations. Many are hitting exactly the problems traditional implementations hit: the wrong process automated, duplicate data, rules nobody agreed, and a system the team works around. The difference is speed. The mistakes arrive sooner, and there is more code to untangle when they do.

AI speeds up building. It doesn’t speed up deciding.

Discovery first

Discovery is the context the AI is missing.

An AI agent builds whatever it is told, quickly and confidently. When the instruction is vague, it doesn’t stop to ask. It fills the gap with a reasonable-sounding guess and keeps going.

Discovery gives it the whole picture in one place: how work flows, what your data means, who may see and change what, which system holds the true record, and how to test the result.

What a Discovery produces for an AI builder

Two ways to build

We build it, or you do. Discovery comes first either way.

We build it

We build it, with AI.

Our consultants run Discovery, then use AI to speed up analysis, building and testing. Our team designs the system, checks the AI’s work and takes responsibility for the result.

You build it

You build it. We keep it on track.

Your team builds with the AI tools you choose. We provide:

  • The AI-ready Discovery pack: the brief your AI builder works from.
  • Build guardrails: where it should live, how to keep data and access safe, and what needs a person’s sign-off.
  • Automated regression tests: the Discovery acceptance criteria turned into tests in Zoho QEngine.
  • Checkpoint reviews: we check the build against the Discovery at agreed milestones, before go-live.
How we manage AI

AI works to a standard, like everything else we deliver.

Our documents follow the Infolytics Precision standard. Our proposals and Statements of Work follow written Blueprints. AI builds follow the Infolytics AI Build Standard. Six of its rules:

  1. 01

    No build without context.

    Every AI build starts from an approved Discovery brief, not a chat.

  2. 02

    The platform is decided, not defaulted.

    Standard Zoho first, then Zoho Creator or Catalyst. Self-hosting only with a written reason and its running costs.

  3. 03

    Test is not live.

    AI agents work on test environments and test data. They never touch live data.

  4. 04

    People approve what matters.

    Changes to live systems, deletions and permission changes need a named person’s sign-off.

  5. 05

    No secrets in code.

    Passwords and access keys are stored securely, never written into the app or pasted into a chat.

  6. 06

    It’s written down.

    Every build ends with a record of what was built and why, so the knowledge doesn’t live in a chat history.

Keeping the human in the loop

AI does the work. People make the decisions.

Every AI build runs as a loop. AI does the work in the middle. People decide at the start, approve and check before anything goes live, and own the result after. Each new need starts the loop again.

AI does the work. People make the decisions.A loop of five steps. 1 Decide: people agree what and why in Discovery. 2 Build: AI builds and tests. 3 Approve: a named person signs off. 4 Verify: the business confirms it works. 5 Own: a named owner runs it. The next change starts the loop again at Decide. Approval comes first for live data, deletions, access changes, customer messages and money.1. DecideDiscovery: what and why2. BuildAI builds and tests3. ApproveNamed person signs off4. VerifyBusiness confirms it works5. OwnNamed owner runs itnext changechange readyApproval first for live data, deletions,access changes, customer messages and moneyPeople decideAI does the work
Decide People agree the processes, rules, roles and what “done” means.Build AI builds and tests in a test environment, from the brief.Approve A named person approves anything hard to undo.Verify Tests pass, and the business confirms it does what it is for.Own A named owner runs the app and answers for it.

A good approval

The person sees what will change, what it affects and how to undo it. Approvals are kept for what matters, so each one gets read, not rubber-stamped.

Decisions about people

POPIA protects people from important decisions made by automated processing alone, so wherever AI decides something about a person, a person stays involved.

The cost of skipping it

What a build without Discovery costs.

The endless fix loop

Each fix breaks something else, because nobody holds the overall design. Hours go into talking to agents and bots, and they never show up as project cost.

Logic that’s quietly wrong

It looks right in a demo, then calculates tax, commission or stock incorrectly in daily use.

Repair costs more than building it right

Rescuing a build means first working out what it actually does, then cleaning the data, then rebuilding.

Abandoned projects

Excitement, a fast prototype, a long tail of edge cases, lost trust, and a quiet return to spreadsheets.

Security gaps

Open databases, passwords in code, no record of who changed what, and customer data held with no POPIA thinking behind it.

Stories

It has already happened. Publicly.

Public incident · July 2025

The deleted database

During a 12-day public experiment, an AI coding agent deleted a live production database holding more than 1,200 executive records, then claimed it couldn’t be recovered. It could. The root cause was simple: test and live shared one database, and the agent could reach both. The platform has since separated the two by default.

What Discovery and a build standard would have caught: separate test and live environments, agents kept off live data, and a tested backup.

Source: SaaStr

Public incident · 2025

The open front door

A security researcher found that apps built with an AI app builder were missing basic access rules on their databases. One scan found 303 exposed endpoints across 170 projects. Anyone could read, and sometimes change, customer records, including emails, phone numbers, payment details and access keys.

What Discovery and a build standard would have caught: who may see and change what, decided before the first line was built and tested before go-live. Under POPIA, an exposure like this is a reportable breach.

Source: CVE-2025-48757 summary

More ways it goes wrong Four illustrative scenarios
Illustrative scenario

The quote tool that was almost right

A sales team builds a quoting app over a weekend. It rounds VAT on every line instead of on the total, and applies discounts after tax. Six months of quotes are slightly wrong before Finance notices.

Discovery would have captured the pricing rules and tested them.

Illustrative scenario

The two customer lists

An operations app keeps its own customer table beside Zoho CRM, and a nightly sync creates duplicates. Nobody ever decided which system holds the master record.

Discovery would have decided it before anything was built.

Illustrative scenario

The app nobody can maintain

One capable staff member builds a field-service app with AI, then leaves. The reasoning lives in their chat history. The first time a connected service changes, the app stops working.

A build record would have let the next person pick it up.

Illustrative scenario

The weekend that never ended

A project planned for a weekend becomes four months of “just one more fix”. In staff time, it costs more than a properly scoped build would have.

A Discovery would have shown the real size of the job on day one.

Illustrative scenarios are composites of common patterns, not clients.

Already built?

We’ll untangle it.

The AI did exactly what it was told. That was the problem.

If you already have an AI-built app that isn’t doing what you need, we start by understanding it, not by rewriting it.

  1. 01

    Health check

    We review how the app works, where its data lives and how secure it is.

  2. 02

    Make it safe

    Close the urgent gaps first: access, exposed passwords and keys, backups.

  3. 03

    Decide: keep, fix or rebuild

    Based on evidence, part by part, not on a hunch.

  4. 04

    Put it on solid ground

    Move it onto Zoho where that makes sense, and connect it properly to your other systems.

  5. 05

    Write it down

    A record of what it does and why, so it never becomes a mystery again.

Start with an AI-built app health check

Proof

We build this way ourselves. Try it.

Three things we built with AI, each from a written brief and under the same standard. Two of them you can use now.

You’re on it

This website

Built on Zoho Catalyst with AI, from a written blueprint, an inventory of every page on the old site and our Precision design standard. A person approved every change. Then we ran our own health check on it and fixed what it found, including missing security headers, a content security policy and framing rules.

Also built this way

  • Our consultation bookingReads live availability from Zoho Bookings, books the meeting and updates Zoho CRM without creating duplicates.
  • Our website formsSend support requests to Zoho Desk, training enquiries to Zoho CRM and newsletter sign-ups to Zoho Campaigns, with spam protection.
  • Our proposals and Statements of WorkProduced from written Blueprints and the Precision standard, so every document is structured the same way.

In every case, AI did the heavy lifting. The thinking came first.

Why build on Zoho

Built fast. Built on a platform you can trust.

An AI-built app is only as safe as the place it runs.

Building on Zoho gives you secure sign-in, roles and field-level permissions, audit trails, and hosting run by a company that has spent years securing business data at scale. Build the same app on a server of your own, and all of that becomes your job.

What it really takes to run an app · AI inside Zoho

Honestly

A secure platform is the start, not the finish. Your own rules, permissions and code still need to be right, which is why Discovery decides them and a review checks them.

Questions

Questions people ask.

Short answers to the questions we hear most often.

Will AI replace our implementation partner?

AI replaces a lot of the typing. It doesn’t replace knowing your business, deciding what to build or being accountable when it goes live. That is the work we do.

Can we keep the build in-house?

Yes. Many teams do. We provide the Discovery, the guardrails, regression tests and checkpoint reviews, and your team builds.

Which AI tools do you use?

We use Zia, Zoho’s own AI, inside Zoho systems, and Infolytics’ organisation AI accounts for analysis, building and testing. We follow clear rules on which customer information goes into which tool.

What if our AI build is already live?

Start with an AI-built app health check. It shows what is working, what is at risk and what to fix first.

Next step

Building with AI? Start with Discovery.

Tell us what you’re building, or what you’ve already built. We’ll show you what the AI needs to know.

No sales pitch · Confidential conversation · Practical next steps