AnswersData strategy
What a data strategy consultant actually does
Kristi CantorDigital Content Manager, P3 Adaptive
Most of them hand you a plan. The good ones hand you working data first. The work itself is real: somebody has to look at what you have, what it’s for and what’s in the way. The trouble is the shape the answer usually arrives in. An assessment, a roadmap, a target architecture and a governance model make a persuasive document and leave your company exactly where it was on the day you hired somebody.
A data strategy consultant works out what data a company has, what the business needs to know, where the data should live, and who owns it. Most deliver that as a plan. The better ones get working data in front of people first and let what they use shape the plan.
You searched for this because somebody in your company said the words “data strategy” out loud, and now you’re trying to work out what buying one involves.
I should say up front that we sell this. P3 has been doing it since 2013, so read everything below with that in mind. But I’d rather tell you what the job is actually like than sell you the version that closes faster.
What does a data strategy consultant do?
Strip out the language and a data strategy engagement is four activities. Somebody inventories what data you have and what shape it’s in. Somebody talks to the business about what it wishes it knew. Somebody designs where the data should live and how it should move. Somebody writes down who’s allowed to touch what, and what the rules are.
All four are worth doing. We’ve never once walked into a company and found those four already answered. The problem isn’t the questions.
What do you get from a data strategy engagement?
Proposals for data strategy services describe this in outcome language: alignment, enablement, a data-driven culture. Here’s the same thing in artifacts, which is what really lands in your inbox:
- A current-state assessment. An inventory of systems, sources and reports, usually with an interview summary attached. Genuinely useful the first time anybody does it, and the part most likely to be out of date within a year.
- A maturity score. Your organization rated one to five across six or eight dimensions, against a model the firm wrote. Nobody has ever scored a two and been surprised.
- A target-state architecture. A diagram of the platform you don’t have yet, with the boxes named after products.
- A phased roadmap. Three or four waves over eighteen to twenty-four months, with the platform work in wave one and anything a business person would recognize in wave three.
- A governance charter and a RACI. Who owns what, who approves what, and which committee meets monthly. This one survives longest, because it’s the only artifact that doesn’t depend on the build happening.
Look down that list and notice what’s missing. Not one of those artifacts answers a question anybody in your company actually has. They describe the conditions under which questions could one day be answered.
Why the deliverable is always a document
Then it goes in a folder, because the roadmap’s first phase is a platform decision, and the platform decision needs a business case, and the business case needs the numbers the platform was supposed to produce.
I want to be fair about why this keeps happening, because it isn’t laziness and it isn’t a racket. It’s an incentive that runs the same direction for everybody in the room, including the person who signed the contract.
A plan can’t be wrong yet
Nobody has ever been criticized for a roadmap. A working report can be judged the afternoon it ships: the number is right or it isn’t, and somebody who knows the business will tell you within the hour. Assessment is the safe half of the work, for the consultant and for the person who hired them.
The people you interview have already been trained
Requirements gathering asks the business what it needs. But everybody in your company learned years ago what’s realistic to ask for around here, and they answer inside that. You get back a tidy list of things that were already nearly possible, and none of the things people quietly gave up on.
The sequence puts the payoff last
Platform, then pipelines, then models, then finally something a human being looks at. Every month of that is spend with no feedback attached, which is also why these programs are so easy to cancel in month nine.
What about Deloitte, PwC and Accenture?
They come up in every one of these searches, so let me answer it straight rather than pretend we are in the same conversation.
They are genuinely good at something specific: enormous, multi-year, multi-country programs with regulatory exposure, where the real difficulty is coordinating a thousand people and surviving an audit. If you are carrying a merger of two global manufacturers onto one platform, that is the right kind of firm and I would tell you so.
Most companies buying data strategy are not in that situation, and buy that shape of program anyway, because it is the shape the market sells. The assessment is thorough, the deck is excellent, the transformation is eighteen months, and the questions your people actually have are the same ones they had at the start.
A plan can’t be wrong yet. Something running can be wrong by Thursday, which is the entire reason it’s worth more.
Faucets first
The name is ours and it’s meant literally. Before anybody designs a new pipe, open a faucet on the plumbing you already have and see what comes out.
What “usable” means, as opposed to clean
Almost every company we walk into is sitting on more usable data than it believes. Not clean data. Usable. In practice that means at least one of these is already true, and usually two or three:
- A finance model somebody maintains properly, because the numbers in it get audited and there was never an option to let it rot.
- An operational extract that is basically right, running on a schedule nobody has touched in years, feeding one report a team depends on.
- A warehouse two teams trust for two specific things and distrust for everything else, and they’re usually right about both.
- A spreadsheet that is load-bearing. Somebody rebuilds it by hand every month. It’s one of the most accurate artifacts in the company and it lives on one laptop.
That isn’t the estate anybody would design on purpose. It’s also enough to answer real questions this month, which is the only property that matters right now.
What two weeks of this actually produces
The order goes backwards from the standard one. Get something in front of people who have questions, watch which questions they actually bring, and let that decide what gets built properly. Concretely, you come out of a fortnight holding:
- A handful of answers somebody wanted and couldn’t previously get, which is the only item on this page that pays for itself immediately.
- A list of every question people tried to ask and couldn’t. It’s the most valuable thing in the exercise, and the one no interview produces.
- A concrete modeling backlog: the specific tables, relationships, definitions and names that got in the way, each one attached to the question it blocked.
- Evidence about which sources are load-bearing, measured by what people reached for rather than by what the inventory says is important.
The strategy falls out of that evidence instead of out of interviews. It’s a shorter document and a much harder one to argue with.
Why this produces a better plan, not just a faster one
This is the part that sounds like an excuse for skipping the thinking, so let me be precise about the mechanism. It isn’t that planning is bad. It’s that a conference room is a terrible instrument for measuring demand, and usage is an excellent one.
You find out which data is load-bearing
Not which data somebody said was important. Which tables people keep returning to when they have a real question in front of them. That list is always shorter than the inventory, and never quite the same as the one in the assessment.
You find out what your models are actually missing
Gaps in a semantic model are invisible until somebody without the institutional instinct runs a question through it. Two weeks of real use will surface more concrete modeling work than a month of documentation review, and every item on that list arrives attached to a question somebody wanted answered.
You find out what nobody was asking for
The invisible backlog: the weekly view a regional manager would genuinely use and has never once requested, because everybody learns what’s worth asking for around here. Drop the cost of asking and it all shows up at once. None of it was ever going to come out of a requirements interview.
Now write the roadmap. It’ll be shorter, it’ll be about different things, and you’ll be able to defend every line of it by pointing at something that happened.
Why the old order made sense, and what changed
Sequencing everything behind a platform build wasn’t a mistake when the rules were different. Producing anything a business person could look at used to take a specialist, a desktop tool, a modeling language and a visualization grammar. When each answer is that expensive, batching them behind one big build is rational.
That stopped being true recently enough that most strategy methodology hasn’t caught up with it. When somebody can describe what they want to see and have it, the cheapest way to find out what your company needs is to let it ask, and the long discovery phase becomes the slow way round to a worse answer. Both halves of that shift have their own page: why reports have looked the same for ten years and why your AI can’t answer questions about your own data.
Do you need data strategy consulting at all?
Sometimes not, and we’d rather say so than take the engagement. If you have somebody in-house who knows your business and knows the modeling, give them a fortnight and the mandate to put something in front of people. You’ll learn the same things and keep the money.
Bring in outside help when one of these is true:
- Two reports disagree and nobody can reconcile them. This is almost never a tooling problem. It’s two definitions of the same word, and it usually needs somebody with no stake in which definition wins.
- The person who understood the model has left. Nothing is documented, everything still runs, and nobody will touch it. Common, and genuinely hard from inside.
- The work is stuck for a reason nobody inside the building will say out loud. More common than the first two put together, and the only genuinely good argument for an outsider.
What to ask before you sign anything
Ask whoever you’re talking to, including us. The answers tell you more than the proposal does, because a proposal is written to be agreed with and these aren’t.
- “What will we have running at the end of week two?” If the honest answer is a workshop schedule, you’re buying a document. That may be what you want. Just know which one you’re buying.
- “Which of our existing sources would you use, and which wouldn’t you?” Anybody who’s looked properly has an opinion by the second conversation. A firm that needs a discovery phase to have any opinion at all hasn’t looked.
- “Who’s actually doing the work?” Ask for the people who’ll be in your data, not the practice lead who’s in the room today.
- “What would make you tell us not to do this?” Everybody has a version of that answer if they’ve done the work. Nobody has one if they haven’t.
- “What happens to this if our priorities change in six months?” They will. A plan whose value depends on eighteen stable months hasn’t priced in your actual company.
What to walk away from
- A first phase that’s another assessment, when you’ve already paid for one. Ask what shipped from the last one before you buy a second.
- A maturity score used as the argument. It’s a sales instrument. Every model is calibrated so the answer is “you have work to do,” and it’ll say that whatever is true.
- A platform decision in wave one. Not because the platform will be wrong, but because it’s the most expensive and least reversible call, made at the moment you know the least.
- Nobody willing to name a thing you shouldn’t do. A strategy that’s entirely additive hasn’t been thought about hard enough.
Four things you can check before you hire anybody
Diagnostic
Is this a strategy problem or a delivery problem?
Every one of these you can do this week, without buying anything.
Find the last data strategy document your company paid for. Ask what shipped because of it.
Most companies have one. If nothing shipped, a second document isn’t the missing ingredient, and you should be very suspicious of any proposal whose first phase is another assessment.
Ask three managers what they would look at every week if it existed. Count how many exist.
The gap is your real backlog, and it’s the only requirements document with evidence behind it. Notice how many of them never became a request at all.
Ask what would have to be true for a business person to get a new answer without filing a ticket.
Whatever comes back is your actual data strategy, whether or not anybody ever wrote it down. If the honest answer is “that isn’t how it works here,” you’ve found the thing worth fixing first.
Count the spreadsheets attached to last month's leadership email.
Every one is a question somebody needed answered, couldn’t get answered, and rebuilt by hand. Now it’s circulating outside your security model entirely. It’s the most honest measure of unmet demand in any company I know of.
Where should a data strategy start?
Take one subject area your people already argue about: the one where two reports disagree and everybody has a theory. Don’t pick the cleanest one. The argument is the signal.
Open it up to ten or twenty people who have real questions and no way to answer them. Not a steering committee, and not your power users, who’ve already bent themselves around the limitations and will ask for more of what they already have. Then watch for two weeks and write down what they reach for.
That list is your data strategy. Everything after it is sequencing, and sequencing is the easy part once you know what’s worth building.
Start with one question, not a program.
Tell us the thing your data won’t tell you. We’ll say straight away whether it’s a modeling problem, a plumbing problem or nothing much at all. And if you don’t need us for it, we’ll say that too.