Construction WIP Schedules: Software vs Excel vs AI Agents

By Jude Lee · · Comparison

Two accountants reviewing a construction work-in-progress schedule with job cost reports on a laptop

Why WIP schedules consume so much firm time

If your firm serves contractors, you already know the pattern. Every month (or quarter, or at year-end for the bonding agent), someone exports job cost detail, re-keys contract amounts and change orders, chases project managers for estimates to complete, recomputes percent complete, derives earned revenue, and books over- and underbillings. Then a reviewer re-performs most of it because the stakes are high: these schedules feed the financial statements a surety underwriter uses to set bonding capacity.

The work is repetitive but not trivial. Under FASB’s ASC 606, revenue on many construction contracts is recognized over time, with progress commonly measured using a cost-to-cost input method — the mechanics most WIP schedules encode. The AICPA publishes industry guidance for construction contractors covering how that plays out in practice. Confirm the specific accounting treatment for any client against the standard itself and with a qualified accounting professional.

What the schedule actually needs — and which parts a machine can get

Split a WIP schedule into two buckets and the automation question answers itself.

Retrievable facts (a system of record knows them): original contract value, approved change orders, job costs to date, billings to date, retainage, prior-period WIP figures, GL job-cost totals for tie-out.

Judgment inputs (no system reliably knows them): estimated cost to complete, whether a pending change order is probable enough to include, whether a job has turned into a loss contract requiring immediate recognition, how to treat claims, uninstalled materials, or a disputed back-charge.

Our own framing of the distinction, which the rest of this piece rests on:

Any tool that claims to “generate” a WIP schedule is really generating the retrievable half and borrowing last month’s guesses for the rest.

Construction ERP and WIP modules

Purpose-built construction platforms differ more than vendor comparison charts suggest. Sage’s contractor products and Foundation are full accounting systems — contract, cost, and GL data live together, so a WIP report comes out natively. Knowify and Procore are project- and operations-first and commonly pair with a separate ledger such as QuickBooks Online, which means the WIP math may happen downstream in the paired system or back in your template. Check current product documentation before assuming any of them hands you a bonding-ready schedule; these capabilities change.

When a client is fully on one of these and maintains it, this is the cleanest answer and you should use it. The constraints are real, though. The data is only as good as the field discipline behind it; a PM who never updates the cost-to-complete field produces a confidently wrong report. And for most small and mid-sized firms, the bigger issue is heterogeneity: one client is on a full ERP, three are on QuickBooks Online with job costing, one runs a shared spreadsheet. You cannot standardize your firm’s process on a platform only one client owns.

Excel templates

Excel is still where most firm-prepared WIP schedules live, and there’s nothing embarrassing about that. A good template is transparent, auditable, reviewable, and easy to adapt to a weird contract. It’s also where errors hide — broken links, prior-year rows carried forward, a job that quietly fell off the schedule.

Excel scales with headcount and nothing else. That’s the whole tradeoff. If you prepare a handful of WIP schedules a year, stop reading here; a tight template plus a review checklist is the right call. The automation argument only turns on when volume and client-system variety both rise. Our broader piece on Excel automation versus accounting software versus AI agents walks through where that line typically sits.

Where an AI agent actually fits

An AI agent here is not a chatbot that answers WIP questions. It’s an assistant with governed, least-privilege access to the systems you already use — via MCP, the open Model Context Protocol that lets an AI assistant like Claude call defined tools against your data — running a packaged skill that encodes your firm’s WIP procedure the same way every time.

Concretely, for one contractor client on QuickBooks Online:

  1. Pull the retrievable half

    Read-only access to the GL and job cost detail for the period, plus prior-period WIP workpapers from your document store. See connecting an AI assistant to QuickBooks or Xero via MCP for how that connection is scoped.

  2. Tie out before computing

    The skill requires job cost totals to agree to the GL control account and billings to agree to AR activity. Mismatch stops the run and produces a reconciling-item list — it does not proceed with a plug.

  3. Draft the schedule

    Compute percent complete, earned revenue, over/underbillings and gross profit using last period’s estimates to complete, clearly marked as stale. The output is a dated workbook — summary tab, job detail tab, and an assumptions tab listing every stale ETC — written to that engagement’s period folder in your document management system as client-period-v1-agentdraft. The agent writes v1 and nothing else; the reviewer opens it, saves a reviewer copy, and that copy becomes the controlling version. A re-run creates v2 alongside, never over, the reviewer’s file.

  4. Flag what needs a human

    Jobs over 100% complete, negative estimated cost to complete, billings exceeding contract value, any job with no updated ETC, any job trending to a loss — plus gross margin movement beyond a threshold your firm picks. As a worked example you can adapt: say you flag any job where gross margin moved two percentage points or more from the prior period on contracts above $250,000. That is a threshold you choose and tune, not an industry benchmark. Start it loose enough that reviewers don’t drown in flags, then tighten once you see what’s noise.

  5. Draft the PM query list and the memo

    A plain-language email to the project manager asking only for the flagged jobs’ updated costs to complete, plus draft variance commentary for the reviewer.

  6. Human sign-off, then post

    A person reviews the flags, enters real ETCs, decides change order and loss treatment, approves the over/underbilling journal entry in the reviewer copy. Only then does anything get written back.

That last step matters more than any other design choice. Keep the agent read-only on client ledgers until your review gates are proven, and treat write access as a separate, later decision.

Build, buy, or leave it alone

Off-the-shelf (ERP or WIP module)
Best when a client is already on one construction platform and maintains it. Vendor owns the math and the updates. Weak when your clients are scattered across systems, or when your firm’s schedule format, flag thresholds, and bonding-package conventions differ from the vendor’s.
Custom agent + skill over your existing stack
Best when you prepare WIP for many clients on different ledgers and the firm’s procedure is the standard you want enforced. You own the skill, the thresholds, and the tie-out rules. Costs real engineering time, needs an owner, and demands audit logging and access governance you must maintain.

A custom MCP server makes sense when the data you need lives somewhere no vendor connector reaches — a firm-built job master, a shared workpaper library, a legacy contractor system with only a database export. Expose narrow tools (get_job_cost_detail, get_prior_wip, get_contract_master) rather than broad database access, log every call, and segregate client data so one engagement’s agent cannot read another’s. Firms handling taxpayer data should map these controls against IRS Publication 4557, Safeguarding Taxpayer Data, and their written information security plan.

And sometimes the right answer is neither. Five WIP schedules a year, one client system, a reliable template: automate nothing, tighten the review checklist.

Modeling the payback without inventing numbers

Don’t take anyone’s published savings figure — including ours, which is why we don’t publish one. Measure your own baseline first, on one real engagement:

3 inputs
No agent should produce: estimate to complete, change order probability, loss accrual

Run a parallel month on two clients: agent draft versus normal process, compare line by line, count the differences. That’s your actual accuracy data, and it beats any benchmark.

What this means for the people doing the work

The recurring question — whether AI puts contractor-focused accountants out of a job — gets more honest when you look at the task split. The retrievable half is genuinely automatable and was never where your value sat. The judgment half is where bonding capacity, revenue recognition, and client trust are actually decided, and it requires sitting with an owner and a PM and asking uncomfortable questions about a job that’s fading. Our read: the preparer role compresses and the reviewer-advisor role expands. The firms that struggle will be the ones whose WIP service was the data assembly.

Start narrow. One client, one skill, read-only access, a parallel run, and a flag list your reviewers trust. If the draft saves your preparer real time and your reviewer catches the same issues they always did, expand. If it doesn’t, you’ve lost a month and learned something concrete — which is a better outcome than a firm-wide rollout built on a vendor’s slide.

Not sure where to start?

Get a free automation audit: we map your bookkeeping, month-end close, client onboarding, document collection, and AP/AR — and show you what's worth automating before you spend a dollar.

Get a free automation audit

Related guides