Skip to main content
Productivity•8 min read

Statement of Work: Where Projects Go Wrong Early

November 6, 2026•By Ajan Kanagalingam

Integration with your existing systems. Six words, signed by both sides, each of them certain they knew what it meant. One party was thinking of two systems and the other of nine, and the difference became somebody unpaid overtime in month three.

Ambiguity is not neutral

A vague line feels like flexibility when you are signing. It reads as room to be reasonable, and both parties sign assuming their own interpretation will hold.

It is not neutral, and it does not stay unresolved. It gets resolved later by whoever can afford to push, which in a client relationship is rarely the supplier halfway through delivery with an invoice outstanding.

This is why pinning down specifics is a kindness to both sides rather than a sign of distrust. A client who knows exactly what they are getting and a supplier who knows exactly what they owe have no reason to argue in month three.

The sections that carry the risk

Most statements of work are long in the places that are easy to write and short in the places that decide the outcome.

SectionCommon versionWhat it should say
DeliverablesA list of nounsEach one with a format and a quantity
AcceptanceMissing entirelyWho signs off, against what, in how many days
Out of scopeOmitted as impoliteThe five things people will assume are included
AssumptionsGeneric boilerplateWhat you need from them, by when
Change processBy mutual agreementWho approves, how it is priced, in writing
Client obligationsUnmentionedAccess, data, decisions, and the cost of delay

Suppliers leave client obligations out because it feels presumptuous, and that omission is what sinks timelines. If your work depends on a system login, a data export and one decision from a named person, and any of those arrives three weeks late, the timeline moves. Writing that down in advance converts an awkward conversation into a clause.

An out-of-scope list feels similarly impolite and does the same job. Naming the five things a reasonable person might assume are included is far cheaper than discovering the assumption later, which is the same logic as the specificity in a service level agreement that means something.

Deliverables, not activities

There is a real difference between we will run discovery workshops and you will receive a written requirements document of roughly twenty pages covering these four areas. The first is an activity you can perform fully and still disappoint. The second can be handed over and judged.

Activities are easier to write because they describe what you intend to do, and they are what most statements of work contain. Deliverables are harder because they commit you to an artefact, which is exactly why they protect both sides.

Then attach acceptance to each one. A named person, a stated test, and a review window after which it is deemed accepted. Without that last clause a deliverable can sit unreviewed for two months while the project cannot close.

What AI should and should not draft

The useful job is making the implicit explicit. Give it the meeting notes and the email thread and ask for the scope, the deliverables and, most valuably, a list of assumptions nobody wrote down. The things both sides took for granted are where the risk is, and a model reading the correspondence will surface several you had stopped noticing.

It is also good at the out-of-scope section, which is tedious to write and easy to leave thin, and at checking a draft against your own previous statements of work for sections you have simply forgotten this time.

Where it should stop is anywhere describing what happens when things go wrong. Payment terms, liability, indemnity, intellectual property ownership and termination are commercial and legal positions, not descriptions of work. A model will write all of them fluently, reflecting no particular jurisdiction and no particular appetite for risk, which is the same failure mode as generating contract templates and the reason AI contract review misses what it misses.

One practical check before sending: read the draft looking only for sentences that could be read two ways. Numbers, counts and names are where to look, because every unquantified plural is a future argument.

Change is not the enemy

Projects change. A statement of work written to prevent that gets quietly ignored by both sides, which leaves you with no document at all.

What works is making change visible and priced. A named approver, a written request, and a stated basis for cost means a change is a decision rather than something absorbed in silence until somebody notices the margin has gone.

Where you are the one subcontracting, the same clauses need to flow down, because an obligation you accepted upstream and did not pass on is a gap you now own, which is the point of keeping control of work you do not do yourself. And if a scope keeps arriving vague from the client side, the fix is usually earlier, in how the request was written.

Frequently Asked Questions

What is a statement of work?

The document that says what will be delivered, by when, for how much, and how both sides will know it is finished. It usually sits under a master agreement that handles the legal terms, which leaves the SOW to do the harder job of describing the actual work precisely enough that two parties reading it separately reach the same conclusion.

Why do most disputes trace back to the SOW?

Because ambiguity has a default owner, and it is whoever has less leverage when the argument happens. A line reading integration with your existing systems does not say how many systems, which ones, or what integration means. Both sides read it favourably at signature, and the gap surfaces in month three when somebody has to absorb the extra work.

What is the most commonly missing section?

Acceptance criteria. Plenty of statements of work describe deliverables in detail and never say how a deliverable is judged complete. Without that, the project ends when the client stops asking for changes, which can be a long time after the budget ran out. Name who signs off, against what test, and within how many days.

What should AI draft in a statement of work?

The structural and descriptive parts. Turning meeting notes into a first draft of scope and deliverables, listing assumptions that were implied but never written, producing the out-of-scope section, and checking a draft against your own previous statements of work for missing sections. It is good at making an implicit understanding explicit, which is most of the work.

What needs a human signature rather than a draft?

Payment terms, liability and indemnity, intellectual property ownership, termination, and anything describing what happens when the project goes wrong. These are commercial and legal positions rather than descriptions, they are where the money is, and a model will produce confident language for all of them that reflects no particular jurisdiction and no particular risk appetite.

How does a good SOW prevent scope creep?

By making the change visible rather than by forbidding it. Projects change, and a document pretending otherwise just gets ignored. What stops uncontrolled drift is an explicit out-of-scope list, a named change process with who approves and how it is priced, and assumptions written down so that when one turns out to be false there is a clear trigger for a conversation instead of silent absorption.

Scope it before you sign it

We help Canadian services businesses tighten the sections that decide who pays when something is ambiguous, and automate the drafting that is safe to automate.

Related Articles

Productivity

Root Cause Analysis: Where AI Helps, Where It Guesses

October 4, 2026Read more →
Productivity

Ask Your AI What It Is Likely to Get Wrong

August 25, 2026Read more →
Productivity

“Workslop”: When AI Output Creates More Work

July 24, 2026Read more →
AK
Ajan Kanagalingam
Founder & ChatGPT Consultant, ChatGPT.ca

Ajan leads the ChatGPT.ca team: 200+ custom GPT builds and automation projects for 50+ businesses across 20+ industries. Based in Markham, Ontario. PIPEDA-compliant solutions.

Stay ahead of AI in Canada

Weekly case studies, new tools, and ROI playbooks for Canadian SMEs. One email, zero spam.