See where your attention would pay most.
Programmes rarely stall for technical reasons. They stall at a small number of predictable points. Answer seven questions honestly and you will see which one is worth your attention first, and three things that tend to move it. Nothing is gated and there is no form in the way.
The seven gates
Answer for one specific initiative rather than in general. The gates build on each other, so the first one you cannot answer comfortably is usually where attention repays soonest.
0 of 7 answered
-
Gate 01 · DirectionCan you name the company or business unit goal this serves, and roughly how much it moves it?
-
Gate 02 · DecisionCan you name the decision this changes, say how often it happens, and what a bad one costs?
-
Gate 03 · DataIs there a named human who owns the data this depends on, and would they agree they own it?
-
Gate 04 · DesignDid you price buying or configuring as seriously as you priced building?
-
Gate 05 · DeploymentDoes the output reach the workflow, and are people still using it in month six?
-
Gate 06 · DisciplineCould you evidence how this is governed to a regulator, an insurer or a major customer tomorrow?
-
Gate 07 · DividendWas a baseline measured before anything went live?
Answer the seven questions above. Without JavaScript, all seven explanations are listed below instead.
Direction
Without a goal underneath it, nobody can say what winning looks like, so the programme tends to drift. Settling this early makes every later decision easier.
Three things that tend to move it
- Write one sentence: this exists to move X from A to B by when.
- Take it to the person who owns X. If they do not recognise it, you have found the problem.
- If no such person exists, stop. That is the finding.
Decision
The capability arrived before the decision it was meant to change. Nobody sized the prize, so the business case is built on a guess.
Three things that tend to move it
- Name one decision. Not a department, not a theme. A decision somebody makes repeatedly.
- Count it. How many times a week, how long each, what it costs when it is wrong.
- If you cannot count it, you cannot prove you improved it later.
Data
There is no arbiter. When two systems disagree, every number becomes a negotiation, and anything built on top inherits the argument.
Three things that tend to move it
- List the three sources this depends on most.
- Put a name against each, then go and ask that person if they agree.
- The gap between your list and their answer is the size of the problem.
Design
The build decision was taken by people who enjoy building, and was never properly challenged. Recoverable, but the window closes fast.
Three things that tend to move it
- Check what the licences you already hold actually do. This alone resolves it surprisingly often.
- Get one real quote for buying, not an estimate from someone who wants to build.
- Compare total cost over three years, including the maintenance nobody budgets for.
Deployment
Two things sit behind this. Either it has not reached the workflow yet, or it has and usage has drifted since. Instrumenting it makes the difference visible while it is still easy to change.
Three things that tend to move it
- Watch one person do the job for an hour. Note every screen they actually touch.
- Put the output in one of those screens, even crudely, before improving the model.
- Instrument usage before launch, not after, so drift is visible while it is still fixable.
Discipline
Governance is easiest to build in early and hardest to retrofit. Treated as a route to production rather than a brake, it tends to speed things up.
Three things that tend to move it
- Inventory what is actually running, including whatever staff adopted without telling anyone.
- Classify each one by the data it touches, not by the tool it uses.
- Write the policy last. A policy written before the inventory describes an imaginary organisation.
Dividend
This is the one that is hardest to recover later. Without a before, an after is difficult to prove, which is where funding conversations usually get uncomfortable.
Three things that tend to move it
- If it has not gone live, stop and measure now. Two weeks of observation beats any estimate.
- If it has, find the closest proxy you can defend and be honest about the confidence.
- Get a named budget holder to accept the benefit into their plan, in writing.
Nothing obvious is broken
If all seven hold, the problem is unlikely to be structural. It is usually pace, capacity or a dependency outside your control, and that is a different conversation.
Worth checking
- Whether the objective has changed since you started.
- Whether the named owners still hold those roles.
- Whether the baseline is still being measured, or quietly lapsed.
Go deeper: take the full baseline
The seven questions above point at one gate. The baseline looks at all of them. Stage two asks four questions per gate, 28 in total. Each offers four descriptions and you pick the one that is true today. It gives you a picture of the whole chain rather than a single answer. About eight minutes.
- A score out of 84 with a band, so you can see the overall shape rather than one problem.
- Every gate scored separately, so you can see which parts are carrying the programme and which need more attention.
- A suggested order of work for the three gates scoring lowest, based on how you answered.
- A one page report with your organisation on it, ready to take to whoever needs to see it.
Answer for one specific initiative. Pick the description that is true today rather than the one you are working towards.
Answer all four to continue.
The Modern Analytics
Where you are starting from
The 7D Baseline
How ready your data, analytics, BI and AI work really is. Scored across seven gates, from the goal you set to the money you can actually show for it.
Early
Summary
What this looks at
The 7D method scores the whole chain from a company or business unit goal through to value booked in the accounts. Seven gates, each out of 12, taken in order because each one depends on the one before it.
This is not an AI readiness review. Most value is lost in the data and analytics layers, long before anyone reaches a model. That is why gates three and four carry as much weight as anything later.
Self scored against the 7D method. Prepared by The Modern Analytics, themodernanalytics.com
7D Baseline ·
Your programme
Data, analytics and AI readiness across seven gates
Early
How to read this. Each gate is scored out of 12. The gates run in order and depend on each other, so a low score early costs you more than a low score late. The shorter bars are where attention tends to repay soonest. Where gate order and score order disagree, gate order usually wins.
Where this sits
Four bands. A number on its own is not a position, so here is the position.
Where the chain breaks first
The gates depend on each other, so a low score early costs more than a low score late. Everything downstream of the first break inherits it, whatever it scores on its own.
Gate 01 Direction
The next ninety days
Three moves in gate order rather than score order, because a gate cannot be judged until the ones before it hold. Each says what it releases, which is the question worth asking before starting any of them.
-
Move 1
Gate 01 Direction
Agree the objective in writing with whoever owns the number, with a figure and a date, before the next funding decision.
-
Move 1
Gate 02 Decision
Build a decision register for the top three candidates: volume, cycle time, error rate and cost to serve. Interview the people who make the decision.
-
Move 1
Gate 03 Data
Name owners for your critical sources and confirm with them directly. Measure quality against a threshold before building anything on top.
-
Move 1
Gate 04 Design
Audit the licences you already hold, then obtain one genuine external quote. Compare over three years, including the people to run it.
-
Move 1
Gate 05 Deployment
Watch the work for an hour. Put the output into a screen people already open, crudely, before improving the model. Then instrument usage so drift is visible before it is terminal.
-
Move 1
Gate 06 Discipline
Run the inventory before writing policy. Classify by data touched. For anything that acts autonomously, define authority, rollback and a named owner.
-
Move 1
Gate 07 Dividend
If it has not gone live, measure the baseline now over a real period. Get a named budget holder to accept the benefit in writing.
Then run this again. Ninety days is long enough for a real change to show and short enough that you can still remember what you changed.
Gate by gate
Each gate read against your answers, with the target state and the first move. Read them in gate order rather than score order.
Direction
0 of 12 EarlyWhich company or business unit goal does this serve, and by how much?
The objective is written down, owned by the person whose number moves, and carries a figure and a date. Everything downstream can be tested against it, which is what makes the rest of this assessment meaningful.
There is an objective, but it is missing either the number, the date, or the agreement of the person who owns it. That gap is small to close now and expensive to close after funding, because by then the objective has been quietly reinterpreted three times.
No agreed objective that a finance director would recognise. Every gate after this one inherits the ambiguity, which is why work like this tends to be judged on delivery rather than on value.
- What good looks like
- One sentence, agreed in writing before any money is committed. It names the goal, the measure, the number, the date, and the person accountable for it.
- First move
- Agree the objective in writing with whoever owns the number, with a figure and a date, before the next funding decision.
- Roughly
- A week, mostly other people's diaries
Decision
0 of 12 EarlyWhich decision changes, how often is it made, and what does a bad one cost?
The decision is named, sized and owned. You will be able to prove whether it improved, which is the only durable defence of the investment.
The decision is understood but not counted. Without volume, cycle time, error rate and cost to serve, the prize is an assertion, and assertions do not survive a second budget round.
No named decision. A capability is being built and a use will be found for it afterwards. This is the most common reason technically successful work produces nothing anybody can point at.
- What good looks like
- A short register of the decisions in play. For each one: how often it is made, how long it takes, how often it goes wrong, and what a wrong one costs.
- First move
- Build a decision register for the top three candidates: volume, cycle time, error rate and cost to serve. Interview the people who make the decision.
- Roughly
- Two weeks, including watching the decision made
Data
0 of 12 EarlyDoes trustworthy ground truth exist, and does a named human own it?
Ground truth exists and a named human owns it. This is the gate that cannot be retrofitted cheaply, and you are past it.
The data is usable but ownership or quality is informal. It holds until three systems disagree, at which point there is no arbiter and the argument is settled by seniority rather than evidence.
No trustworthy ground truth with an owner. Anything built on this will be correct in demonstration and disputed in production, usually around month four.
- What good looks like
- A named person owning every source that matters. Not a team, a person. Quality measured against a standard you agreed, rather than one you inherited.
- First move
- Name owners for your critical sources and confirm with them directly. Measure quality against a threshold before building anything on top.
- Roughly
- Three to four weeks, longer if ownership is contested
Design
0 of 12 EarlyBuy, configure or build, with the third option honestly priced?
Buy, configure and build were compared honestly, including the people cost of running it. That comparison usually saves more than any single delivery decision.
The option was chosen before it was tested. Often the answer is still right, but nobody can show the working, which makes it hard to defend when the bill arrives.
Building by default, or buying by default. Either way the third option was never honestly priced, and the licence you already hold may well do the job.
- What good looks like
- Buy, configure and build priced side by side over three years. Include the people needed to run it. Get at least one real quote from outside.
- First move
- Audit the licences you already hold, then obtain one genuine external quote. Compare over three years, including the people to run it.
- Roughly
- Two weeks, one of them waiting on a quote
Deployment
0 of 12 EarlyDoes it land in the workflow people already use, and are they still using it in month six?
It reaches the workflow people already use, and they are still using it. This is the most commonly skipped gate and the hardest to fake.
It reached the workflow, but usage is not instrumented. You will find out it stopped being used some months after it stopped being used, which is the expensive way to find out.
It has not landed in the workflow, or it landed and people went back to the spreadsheet once attention moved on. Adoption is the gate a supplier has least incentive to test, because the invoice clears at go live.
- What good looks like
- The output lands in a screen people already open. Usage is measured from the first release, so you can see it slipping while there is still time.
- First move
- Watch the work for an hour. Put the output into a screen people already open, crudely, before improving the model. Then instrument usage so drift is visible before it is terminal.
- Roughly
- A fortnight to instrument, a quarter to know
Discipline
0 of 12 EarlyWho governs it, and can you evidence that to a customer or a regulator?
Governance is documented, classified and evidenced. You could answer a regulator or a large customer today rather than assembling something over a weekend.
Policy exists but the inventory behind it does not, or the boundaries for anything that acts autonomously are undefined. This holds until the first review that asks for evidence rather than intent.
Nothing classified, nothing documented, no authority boundaries. This is where go live gets blocked in week ten, after the money is spent and the date has been promised.
- What good looks like
- A list of what is running and how risky each item is. Anything that acts on its own has a defined limit, a way to stop it, and a named owner.
- First move
- Run the inventory before writing policy. Classify by data touched. For anything that acts autonomously, define authority, rollback and a named owner.
- Roughly
- Two weeks for the inventory, then proportionate
Dividend
0 of 12 EarlyIs the value booked against a named budget holder, with a baseline?
The value is booked against a named budget holder with a baseline behind it. Almost nobody can do this, and it is the single best protection against a year two cut.
There is an intention to measure but no captured baseline. A baseline cannot be recovered after go live by anybody, at any price, which makes this a deadline rather than a task.
No baseline and no named budget holder. Whatever this delivers will be real and unprovable, and unprovable benefits are the first thing cut when budgets tighten.
- What good looks like
- A baseline measured before go live. A budget holder who has accepted the benefit in writing. A review in the diary for ninety days out.
- First move
- If it has not gone live, measure the baseline now over a real period. Get a named budget holder to accept the benefit in writing.
- Roughly
- Days, and it has to be before go live
Where attention would repay first
- Gate 01 DirectionAgree the objective in writing with whoever owns the number, with a figure and a date, before the next funding decision.
- Gate 02 DecisionBuild a decision register for the top three candidates: volume, cycle time, error rate and cost to serve. Interview the people who make the decision.
- Gate 03 DataName owners for your critical sources and confirm with them directly. Measure quality against a threshold before building anything on top.
- Gate 04 DesignAudit the licences you already hold, then obtain one genuine external quote. Compare over three years, including the people to run it.
- Gate 05 DeploymentWatch the work for an hour. Put the output into a screen people already open, crudely, before improving the model. Then instrument usage so drift is visible before it is terminal.
- Gate 06 DisciplineRun the inventory before writing policy. Classify by data touched. For anything that acts autonomously, define authority, rollback and a named owner.
- Gate 07 DividendIf it has not gone live, measure the baseline now over a real period. Get a named budget holder to accept the benefit in writing.
What is already working well
- Gate 01 DirectionYou know what this is for and who owns the outcome. That is rarer than it sounds and it protects everything downstream.
- Gate 02 DecisionThe decision is named and sized. You will be able to prove whether it improved.
- Gate 03 DataYour foundation holds. Ownership and quality are the two things that cannot be retrofitted cheaply.
- Gate 04 DesignThe build or buy call was made honestly. That usually saves more money than any single delivery decision.
- Gate 05 DeploymentIt reaches the workflow and people are still using it. This is the most commonly skipped gate and the hardest to fake.
- Gate 06 DisciplineYou could answer a regulator or a large customer. Most organisations discover they cannot at the worst possible moment.
- Gate 07 DividendYou can prove what it was worth. Almost nobody can, which is why budgets get cut in year two.
Keep this baseline
Your answers are on this device and nowhere else. If you want to come back to them, or run this again in ninety days and see what moved, I have to store them. That is your call, not mine, so it is a checkbox rather than a default.
- Stored: your twenty eight answers, the gate scores and the date. Held in the Mumbai region.
- You get: a private link by email. No password, no account to remember.
- Deleted: whenever you ask, no reason needed, same week.
Saves as a three page PDF: a cover with your organisation and score, the gate by gate detail, then where to start. Prints cleanly in black and white.
Why take my word for any of this
A fair question, and the honest answer is that you should not, yet. So here is what I would want to know if I were you.
The Modern Analytics is new, and I am not going to pretend otherwise. Behind it are thirteen years leading data, analytics and AI at TE Connectivity, Silicon Valley Bank, Novartis, JP Morgan and TCS.
Across 20+ AI and analytics products covering churn, pricing, segmentation, forecasting and opportunity scoring. Verified means measured against a baseline and accepted by finance, not modelled in a slide.
Built and led a 50+ person team across India, the US and Europe. Took analytics maturity from 46% to 88% in a year. Owned platforms serving 25,000 users.
Conversational AI for 10,000 users, agentic workflows that cut manual monitoring by 60%, model governance across eight production models, and three platforms shipped for clients.
The seven questions above are the actual first pass of a paid engagement. If the answer you got was useful, that is the evidence. If it was not, you have lost four minutes and owe me nothing.
Clients are described rather than named because the work sits under commercial confidentiality. Named references come once we are talking properly, not on a public page. Completed the Chief Digital and AI Officer programme at the Indian School of Business. Certified across Databricks, AWS and Alteryx. All of it checkable on LinkedIn.
If it was useful, the next hour is free too
The questions above give you a diagnosis. A conversation gives you the reasoning behind it, applied to your actual situation rather than a generic gate. Forty five minutes, no deck, no pitch, no follow up sequence.
If I cannot help, I will say so on the call and point you at someone who can.
Chinmaya Dash, The Modern Analytics. Data modernisation, analytics and BI, AI and agentic transformation, legacy modernisation, platform development, governance and adoption. More about the practice or read how the seven gates work.