AI Readiness & Opportunity
Not sure where AI fits? Start with the work, not the model.
You keep hearing about AI, a few people are already trying it, and someone has asked for a strategy. Northvian maps the work your organization actually does, finds where AI would help, and tells you where it wouldn't.
01
Is this you?
-
“We keep hearing about AI and we're not sure where it would actually help our business.”
-
“A few people on the team already use ChatGPT, Copilot or Claude for drafting, on their own, and nobody has compared notes.”
ChatGPTCopilotClaude
-
“We have years of documents in SharePoint and Google Drive, and people still ask each other where things are.”
SharePointGoogle Drive
-
“A vendor is pitching us an AI feature and we can't tell whether to buy it or build something ourselves.”
-
“The board, or a partner, has asked for an AI strategy, and we'd rather have a plan than a slide.”
02
What this work is
An AI readiness and opportunity assessment is a short, structured look at the work your organization does, to find the few places where AI would make a real difference and say which approach fits each. It starts with the work: who does what, with which information, how often, and what goes wrong. The model comes last.
Ready has a plain meaning here. Can the work be described? Does the information it needs live somewhere a system could reach, or in people's heads and inboxes? Are the rules predictable, or does each case need judgment? Is anyone allowed to use this data this way? Would a person check the result before it matters? Those five questions decide more than any tool.
The output is not a strategy deck. It's an opportunity map: candidate pieces of work, weighed on value, effort, data availability and risk, with a recommendation for one pilot and a sensible order for the rest. Some candidates come back marked not AI: better search, a simpler automation, a form, a process change. Deciding what not to build is part of good AI strategy, and we write those down too.
Where a candidate does need AI, we say which kind. Retrieval, a system that finds the right passages in your own documents and hands them to the model before it answers, is a different thing from an agent, an AI system that takes steps and uses tools to complete a task rather than answering one question. Both are different again from a person using Copilot with a good prompt. The assessment names the fit.
03
What we actually do
- 01
Inventory the work
We talk to the people who do the work, across roles, and list what repeats, what is slow and what goes wrong. You get a written list of candidate work, in your words.
- 02
Check the information
For each candidate: where the information lives (SharePoint, Drive, a line-of-business system, a mailbox, a spreadsheet), who may access it, how current it is, and whether personal information is in it. You get a data availability note each.
- 03
Map the workflow
For the strongest candidates, the steps as they actually run, exceptions included, and which steps are predictable versus judgment. You get a one-page workflow map for each.
- 04
Test feasibility
A quick, hands-on trial for the top candidates, with your documents and your questions, so the recommendation rests on evidence rather than a vendor demo. You get a short feasibility note: what worked, what did not, what it would take.
- 05
Weigh privacy and risk
Which information would leave your environment, which decisions the system would touch, and where a person must stay in the loop. You get risk observations you can hand to whoever owns privacy.
- 06
Score and choose
Value, effort, data, risk, and how visible a win would be, per candidate. You get a ranked list and a pilot recommendation with a single success measure.
- 07
Write the roadmap
The pilot, what follows, what to buy, what to build, what to drop. You get a roadmap you can act on without us.
04
What we might tell you not to do
-
Don't start with an AI strategy document
A strategy written before anyone has looked at the work is a list of hopes. Write the list of work first. The strategy is what the list tells you.
-
Don't build a chatbot over your whole document library first
It's the most requested project and the one most likely to disappoint: unclear permissions, stale documents, and questions nobody agreed it should answer. A narrow retrieval pilot on one team's documents teaches you more, faster.
-
Don't buy a platform before you have a use
A licence for everyone, in the hope that uses appear, mostly produces a renewal conversation. Run one workflow properly, then decide what to buy.
-
Don't automate the judgment step first
The step that needs an experienced person should stay with them for now. The predictable steps around it are where the time goes anyway.
05
Illustrative scenario
A firm with twenty years of documents and three people using ChatGPT
Situation
We're a mid-sized law firm. Precedents live in a document management system and on a shared drive, and associates search by memory. Three of our lawyers had started using ChatGPT to summarize clauses, on personal accounts, which made the managing partner nervous. We'd been asked by the partnership whether we needed an AI strategy.
What the work looked like
The assessment interviewed partners, associates, clerks and the office manager and listed the work that repeated: locating precedents, first-draft summaries, intake conflict checks, engagement letters. Feasibility trials ran retrieval over one practice group's precedents, using real questions from the associates. The privacy review flagged which documents could not leave the firm's environment and which tools' terms allowed client material at all.
What changed
The recommendation was narrow: a retrieval pilot for one practice group, with permissions mirrored from the document system; a short approved-tools rule so the personal-account drafting stopped; and a clear not-now on client-facing chat. The strategy question answered itself. The roadmap was the strategy.
06
Illustrative scenario
An operations team re-typing the same request into three systems
Situation
I run operations for a distribution business. Every order change arrives by email, gets re-typed into the ERP, then into the carrier portal, then confirmed back by email. We'd been told an AI agent could do all of it. I wanted to know whether that was true before we spent anything.
What the work looked like
We mapped the workflow with the two coordinators who ran it, exceptions included. Most of it was predictable: the same fields, the same three systems, the same confirmation. The judgment sat in one place, deciding whether a change was allowed under the customer's terms. The feasibility trial used a month of real emails to test how accurately the fields the ERP needed could be extracted.
What changed
The map said plain automation for the re-typing, an AI step only for reading the unstructured emails, and the coordinator kept on the terms decision with a one-click approve. No agent. The pilot recommendation was one customer's orders for a few weeks with a single measure: minutes per change. What came after was a hand-off to the automation work, with the judgment step still human.
07
How an engagement runs
- 01 1 to 2 weeks
Discover
Interviews across roles, the inventory of work, and a first list of candidates.
- 02 1 to 2 weeks
Assess
Data and workflow checks, feasibility trials on the top candidates, privacy and risk notes.
- 03 About 1 week
Decide
Scoring, the pilot recommendation, the roadmap, and a readout with the people who will own it.
Formats
- A fixed-scope readiness assessment, typically three to five weeks end to end.
- A shorter opportunity workshop for a single team or a single workflow.
- Advisory time by the month, for organizations that want a second opinion as they go.
Bands are typical, not promises. Access to people and documents sets the pace.
08
What you receive
-
Opportunity map
Every candidate piece of work, weighed on value, effort, data availability and risk.
-
Prioritized use cases
The candidates worth pursuing, in order, with the kind of solution each needs.
-
Feasibility notes
What the hands-on trials showed on your own documents, including what did not work.
-
Data and integration considerations
Where the information lives, who can reach it, and what it would take to connect.
-
Risk observations
Privacy, data-handling and decision risks per candidate, written for whoever owns them.
-
Pilot recommendation
One piece of work, one success measure, one owner, and what a good result would look like.
-
Roadmap
The sequence after the pilot, including what to buy, what to build and what to drop.
09
Technical and risk notes
What we look at
- Where the information lives: document stores, line-of-business systems, mailboxes, spreadsheets, and how each is accessed and refreshed.
- Identity and permissions on those sources, and whether a system could honour them per user.
- Integration surfaces: an API, an export, a webhook, or nothing but a screen.
- Data sensitivity: personal information, client-confidential material, contractual restrictions on where it may go.
- Predictable versus judgment steps in each candidate workflow.
- The terms of the tools already in use: retention, training use, region, admin controls.
- Who would own a pilot, and how much time they actually have.
What can go wrong
- The strongest candidate has no reachable data; the information is in people's heads and inboxes.
- Permissions do not exist at the document level, so any retrieval system would see everything.
- The vendor's AI feature is a consumer-tier tool with different data terms from the one you thought you were buying.
- Value is measured in hours saved that nobody reclaims.
- The workflow map skips the exceptions, and the exceptions turn out to be most of the work.
- Nobody owns the result after the readout, and the roadmap becomes a document.
10
Questions people ask
What exactly is assessed?
The work, first: what repeats, what is slow, what goes wrong. Then the information behind it, the people who do it and who would own a change, the rules already in place for tools and data, and what you are prepared to spend. Technology is assessed last, and only against the work that came out on top.
How long does it take, and what do we get at the end?
A full assessment typically runs three to five weeks, depending on how many teams are in scope. You receive the opportunity map, the prioritized use cases with feasibility notes, the data and risk observations, a pilot recommendation and a roadmap, all written to be acted on without us.
How does pricing work?
As a fixed scope, quoted after a first conversation, once we know how many teams and workflows are involved. There is no hourly meter running during an assessment. If the scope changes, we say so before it does.
What happens if the answer is that we do not need AI?
Then that is the answer, and the roadmap says what you need instead: better search, a simpler automation, a form, a process change, or nothing yet. You still have the map of your work and the reasons. We would rather tell you now than after a build.
How is this different from the free AI check?
The check is fourteen questions you answer about yourself in five minutes, and it tells you your starting point and your biggest gap. The assessment is us talking to your people, looking at your documents and systems, and trying things on your real examples. One is a compass; the other is a map.
Who is it for?
Organizations from around ten people to a few hundred, where the work is knowledge-heavy: law and accounting firms, professional services, agencies, and operations teams in businesses of any kind. If your staff spend their days finding, reading, drafting and re-typing, there is usually something here.
What do you need from us?
Time with the people who do the work, usually an hour each. A sample of the documents and systems involved, under whatever confidentiality terms you need. And one person on your side who can decide scope as we go.
What comes after the assessment?
Usually a pilot. A workflow becomes an AI Automation & Agents engagement; a product or an internal assistant becomes AI Product Development; rules for tools and data become AI Governance. Some organizations take the roadmap and run it themselves, which is fine by us.
11
Related
Bring the work. We'll find where AI fits, and where it doesn't.
One conversation is usually enough to tell whether an assessment would help you.