AI Transformation
From scattered pilots to a governed AI capability that ships. An operator’s account, not a slide deck: we have put AI into a client’s business, into our own products and into our own company — and the programme is built from what went wrong along the way.
Why most AI ideas never go live
When a leadership team scores every place AI could help, most ideas fall out — and for the same few reasons.
- The AI would hand a decision straight to a customer or a dealer, with no test to catch a wrong one.
- The data it needs lives in spreadsheets.
- Nobody will own the number.
- The ERP or the CRM already does it.
- There is no agreed way to rank what is left, so the list never becomes a sequence.
The ideas that survive share one shape: the AI reads, finds or drafts, and a person still decides. A workshop does not find those. A sequence does.
A point of view, researched before we arrive
A value case built from your own numbers
A scored portfolio, then forced choices
Owners before builds
1. Guardrails and data security
2. Telemetry
3. What kind of AI it is
4. Who builds it
5. Model choice and cost
The hands — your functions build these
The spine — built once
The brain — built once
Your systems — nothing changes
Start with the AI you already own
Most enterprises already pay for AI inside their core platforms. Switch it on where the work lives in one system. Build only where the work crosses systems.
- ERP AI, such as SAP Joule — orders, invoices, disputes and cash. It stops where the work crosses to email, the warehouse or the storefront.
- Copilot on your data platform — plain-language questions, reports and summaries. It stops at the edge of Microsoft.
- CRM AI, such as Salesforce Einstein — cases, replies and service history. Stock, credit and delivery live somewhere else.
What none of them can do alone is the work that crosses two or three systems: is this claim genuine, ship now or hold for credit, what do we say to this dealer. That needs one layer, built once.
Where to start: six weeks, beside your work
One workflow and one owner, run next to the way you work today. Nothing in your operation changes until you decide.
- Week 0 — pick one workflow and measure today’s number.
- Weeks 1–2 — connect the data, read-only, and write the rules: what the AI may do, and who approves.
- Weeks 3–4 — run it beside today’s way. The owner sees both, every day.
- Week 5 — the owner judges whether it beat today.
- Week 6 — scale it, fix it, or stop. It is cheap to be wrong.
What we need from you. A named sponsor. One owner per workflow — the person who owns the number — for two hours a week. Read access to your ERP, data platform and CRM. And your own rules: credit limits, content standards, service promises. They become the check the AI is tested against, and no partner can supply them.
Across the six weeks you also get a scored portfolio with an agreed Wave 1, a value case for each use case, an architecture recommendation, and a decision log.
Private Enterprise AI
Where the data cannot leave, the platform comes to it. A governed AI environment inside your own perimeter: enterprise sign-on, role-based access, an AI gateway, your choice of commercial or open-weight models, and retrieval grounded in your own documents. AWS-native, live in six to eight weeks.
It exists because the alternative — staff quietly pasting company data into public tools — is already happening in most organisations we walk into. We have delivered it company-wide for an Indian FMCG group.
Where we have done it
Three places, at increasing distance from someone else’s rules. The further from a client’s own business, the more freedom the AI gets — because freedom is not a measure of how good the AI is. It is a measure of whose rules apply, and how fast a mistake can be caught.
Inside a client’s business
A global industrial manufacturer. Its leadership scored every place AI could help — 48 opportunities were scored — and the ones that went live all share one shape: the AI reads, finds or drafts, and a person makes every decision that reaches a dealer or a customer. The boundary was set by their CIO and audit committee. An Iksula client since 2022.
Inside our own products
Our product-content and data-quality accelerators, rebuilt so the AI does the work and the rules check it. A reviewer sees the exceptions, not the batch. Production AI on product data also runs for a large North American distributor and a global home-improvement retailer.
Inside our own company
We run parts of Iksula on agents built over one governed company brain — sales, hiring, budgeting and delivery. Several are maintained by people who are not engineers, on rails an engineer owns. Money, brand and hiring always stop for a person.
Frequently asked questions
We have already run pilots. Where does this start?
Usually with the pilots. They tell us what your data can actually support and where the organisation already has energy. The first job is turning a pile of experiments into a ranked portfolio with an agreed Wave 1.
Is this strategy work or engineering work?
Both, and the handover between them is the point. The programme is run by people who also build and operate the systems, so the roadmap is constrained by what can genuinely be delivered.
How long is the programme?
Six weeks for the first use case. It runs as a test beside your current process. At the end you scale it, fix it, or stop. Nothing in your operation changes until you decide.
Do we have to buy new AI tooling?
Often not. We start by exhausting the AI you already own inside your ERP, data platform and CRM. We build only where the work crosses systems, because that is the part none of them can do alone.
Our data is not ready. Is this premature?
Data readiness is one of the three scores in the portfolio, so it shapes sequencing rather than blocking the programme. Some Wave 1 choices are deliberately picked because they work on the data you already trust.
Can our business teams build their own AI?
Yes — once the rails exist. Letting business users build before there is one gateway, data labels and a check is how company data ends up in public tools. What scales safely is the process owner paired with an engineer who owns the guardrails.
Who needs to be in the room?
A named sponsor who clears the blockers and does not pick the use cases, the executive team for the decision-forcing sessions, one owner for each workflow, and the people who run your systems for the architecture work.
What happens after the six weeks?
Wave 1 moves into build, and anything that reaches production moves into managed AI operations. Both are described on their own pages.
Three places, at increasing distance from someone else’s rules. The further from a client’s own business, the more freedom the AI gets — because freedom is not a measure of how good the AI is. It is a measure of whose rules apply, and how fast a mistake can be caught.
Bring us your hardest use case
Not the easiest one — the one your team keeps arguing about. We will tell you whether it is worth doing, what it depends on, and what a six-week test would show. It is cheap to be wrong. Not starting is the expensive option.
Connect with our experts
DJ Basumatari
Chief Executive Officer
Abhishek Jain
Director - Solutions & Innovation
