Notes from the work
Short, practical reading for decisions about AI, data and your own team.
Before your first AI pilot
Six questions to answer before your first AI pilot, the reasons pilots stall, and how to choose the smallest useful next step.
By Akshay Dodhiwala
Read the article
Building a capability you can own
How build-operate-transfer works for an India-based capability centre, how it compares with outsourcing or building alone, and what to agree before you start.
By Akshay Dodhiwala
Read the article
Agentic AI in the enterprise: where it works and where it fails
An agent is a model that plans, calls tools and acts across several steps. It suits bounded work with checkable steps, and it needs narrow permissions, approvals, full logs and a route to a person.
By Ketan Nandani
Read the article
How to set up a global capability centre in India
What has to happen, and in what order, to open a capability centre in India that your company owns and that does real work from its first months.
By Akshay Dodhiwala
Read the article
How to choose your first process to automate
Score candidate processes on seven factors, take a baseline before you build, and avoid the first choices that most often stall.
By Akshay Dodhiwala
Read the article
Build, buy or partner: how to source AI capability
Four routes to AI capability and four questions that choose between them: how central it is, how fast you need it, whether you can hire the talent, and what you want to own at the end.
By Akshay Dodhiwala
Read the article
Co-op marketing funds: where the money and time go
Co-op funds pay for local marketing, but late claims, missing proof and slow approvals waste much of the effort. Here is the cycle, the leaks and the controls.
By Akshay Dodhiwala
Read the article
RPA, intelligent automation and agentic automation compared
Three kinds of automation, separated by how much judgement the software is asked to exercise. Rule-based work often needs no AI at all.
By Akshay Dodhiwala
Read the article
Choosing a contract model for AI, data and automation work
Three contract models, each placing risk differently between client and provider. Choose by how well the work can be defined and what you want to own at the end.
By Akshay Dodhiwala
Read the article
What an AI centre of excellence is, and when you need one
An AI centre of excellence is a permanent team that owns how your organisation builds and runs AI. Here is what it does, who is in it, and how to tell whether you need one.
By Akshay Dodhiwala
Read the article
How to evaluate an LLM application before you launch it
A fluent demo proves little. Build a test set from real tasks, define a good answer, score retrieval and answers separately, rerun on every change and keep watching after launch.
By Ketan Nandani
Read the article
An assistant is only as useful as its sources
An assistant that answers from your documents can only be as good as those documents. Here is how RAG works, and how to choose, prepare, permission and test the source set.
By Ketan Nandani
Read the article
Running an AI pilot that can actually reach production
Pilots stall when they are run as technology experiments. Run yours as a project: one scope, one owner, a baseline, a small team, weekly checkpoints and a decision to scale, change or stop.
By Ketan Nandani
Read the article
Why AI projects stall on data, and a checklist to run first
A prototype works on a hand-cleaned export. Production needs data that is owned, accessible, current and permitted. Seven checks tell you where you stand before you build.
By Akshay Dodhiwala
Read the article
What a forward deployed engineer does, and when you need one
A forward deployed engineer works inside your team and systems until a product or model solves your real problem. Here is how the role works and how to judge whether you need it.
By Akshay Dodhiwala
Read the article
Voice AI for customer and staff calls: what works, what is hard
Voice AI handles routine, structured calls well. Accents, noise, interruptions and delay make it harder than chat. Ground the answers, keep a route to a person, and judge it on resolved calls.
By Ketan Nandani
Read the article
Digital asset management for dealer and partner networks
Partners use whichever file is easiest to reach. Good asset management makes the approved file the easiest one, then checks that it is really used.
By Akshay Dodhiwala
Read the article
UiPath or Power Automate: how to choose
Both tools can be the right answer, and some organisations use both. Choose on fit with your systems, your work and the people who will maintain it.
By Akshay Dodhiwala
Read the article
Handwritten forms and regional languages: what document AI can do
Document AI can read handwritten forms in regional scripts well enough to automate real work, provided you control scan quality, validate every field and send doubtful cases to people.
By Ketan Nandani
Read the article
How to plan a legacy database migration without downtime
Most migration downtime comes from one big switch. Plan the inventory, parallel run, reconciliation, cutover and rollback well, and the switch becomes small and dull.
By Ketan Nandani
Read the article
Adding AI features to a product your customers already use
A first AI feature works when it removes effort from a task users already do, inside the screens they already use. Here is how to choose, test, launch and maintain it.
By Akshay Dodhiwala
Read the article
Custom ERP or off-the-shelf: how manufacturers should decide
A standard package suits standard processes. When the way you make things is what differs, map the plant first, then choose between a package, a customisable platform and a custom build.
By Akshay Dodhiwala
Read the article
Computer vision for site safety: what works on real sites
Cameras can flag missing protective gear, restricted-area entry and known defects. Site conditions, alert quality and human review decide whether it helps.
By Akshay Dodhiwala
Read the article
Adding AI inside an EHR without making clinicians switch
AI that lives in a separate window gets dropped. Put each module inside the screen clinicians already use, keep every clinical output under review, and roll out one module at a time.
By Ketan Nandani
Read the article
Let’s talk about your next step.
The workflow, the data and what needs to change.