Research automation

On this page

An industry research body funded by grower levies and matched government money, an independent research institute or a research-heavy not-for-profit spends most of its budget through other people. Universities, government science agencies, consultants and growers’ groups do the research under contract, and the organisation’s own staff manage the portfolio: contracts, milestones, payments, reports to funders, and the governance that sits over all of it. Research managers, contract and finance officers, and a compliance and governance officer spend much of the day moving information between the project system, the finance system, the records system and the board portal: chasing milestone reports, keeping the obligations register current, preparing board papers, filing records, reconciling income. Those systems already do the core work, and we build around them rather than replace them. Automation takes the tracking, checking, chasing and filing around them and runs it as fixed, tested software. AI is used only where something has to be read, such as a milestone report, a research agreement or a set of minutes, and its output is checked before anything is approved, paid or filed.

Your process gets its own map in the assessment.

Research milestone reports and payments

A research portfolio can hold dozens or hundreds of active projects, each with a schedule of milestones written into its agreement: a progress report, a data set, a field trial result, a final report, an extension activity. Payment to the research provider is usually tied to each milestone being met. A research manager or contract officer keeps the dates in a spreadsheet, emails the principal investigator when something is due, reads the report when it arrives, decides whether it meets the deliverables and then asks finance to release the payment. Tracking the schedule follows rules, so it runs as fixed software. Milestones are loaded from the executed agreement, reminders go to the provider before each due date and again when it passes, the report is logged against its milestone in the project and records systems, and the payment is queued in the finance system only once the milestone is accepted. AI reads the submitted report against the deliverables listed for that milestone and drafts a short note on what appears to be covered and what appears to be missing. The research manager reads the report and decides whether the milestone is accepted, and any late, partial or changed deliverable stays with them to resolve with the provider.

The compliance obligations register

A research organisation answers to several sets of rules at once. There is the funding agreement with government, with its plans, reports and use-of-funds conditions. There are statutory obligations under the legislation that applies to the organisation, whether that is corporations law, charity reporting, a statutory authority’s own Act, privacy, work health and safety or records legislation. And there are the constitution, the board’s own policies and the conditions in major research contracts. The compliance and governance officer keeps a register of these, chases each owner for an update before the audit and risk committee meets, collects evidence from inboxes and shared drives, and redoes the register when an agreement is renewed. Holding and chasing the register follows rules, so it runs as fixed software. Each obligation is held with its source, its owner, its due date and the evidence that it was met, reminders and attestation requests go out on schedule, evidence is filed against the obligation in the records system, and the committee’s status report is built from the register on a set day. AI reads a new or amended agreement, policy or guidance note and suggests obligations that may need adding or changing, and the compliance officer confirms every change before it enters the register. Whether an obligation has been met, and whether anything needs to be reported to a funder or regulator, is decided by the accountable owner and the executive responsible, never by the software.

Research contracts and projects

Each project needs an agreement with the research provider, and most organisations work from a set of templates that cover budget, milestones, intellectual property, data, publication and reporting. A contract officer builds each agreement from the template and the approved proposal, sends it out, tracks the changes the university’s research office asks for, and later handles variations when a project needs more time or a different budget. Building the draft from the template and the approved project details, tracking each version and signature, and loading the executed terms into the project and finance systems all run as fixed software. AI reads the provider’s marked-up version and lists where it departs from your standard terms. The contract officer decides what is acceptable within your policy, anything outside it goes to the legal or executive contact you name, and signing stays with the delegate who holds that authority.

When a call for research proposals or a grant round opens, applications arrive through a form, a portal or email, often with attachments in every format and some sections missing. Someone checks each one for completeness and eligibility, sets up a record, collects conflict-of-interest declarations from assessors and assembles the papers for the assessment panel. Checking applications against the call’s stated criteria, chasing missing sections before the closing date, setting up records and assembling the panel pack all follow rules, so they run as fixed software. AI reads each proposal and drafts a summary in the panel’s format for the program manager to check. Eligibility questions that are not clear-cut go to the program manager, and which projects are funded is decided by the panel and the delegate under your own process.

A project is not finished when the research stops. The final report has to be received and accepted, the last payment made, any unspent funds returned or agreed, intellectual property and data recorded, and the outputs published or passed to extension staff. Closure steps are easy to leave half done when the research manager has moved on to new projects. Each closure runs as a fixed checklist, with the final report logged, the final payment and any return of funds checked against the budget in the finance system, and the project record closed in the project and records systems. AI reads the final report and drafts the plain-language summary for the research manager to edit before anything is published. Whether the project met its objectives, and how its outputs are used, stays with the research manager.

Governance, board and records

The board, the audit and risk committee and any research advisory committees each meet on a cycle, and each meeting needs an agenda, papers from several authors, minutes and an action register. The governance officer chases authors for papers, formats them into the pack, loads them to the board portal, drafts the minutes and then chases the actions until the next meeting. Building the agenda from the annual cycle, requesting papers by a set date, assembling the pack in your template and chasing open actions before each meeting all run as fixed software. AI drafts the minutes from the governance officer’s notes or the recording, in your format, for the company secretary or governance officer to correct. The chair approves the minutes, and the board’s decisions are recorded as the board made them.

Directors, committee members, staff and external assessors all hold interests that can touch a research decision, from a university appointment to a farm or business in the industry. Declarations are collected on appointment, at each meeting and before each assessment panel, and someone keeps the register and checks it against each agenda. Sending declaration requests on schedule, holding the register, and matching declared interests to the organisations and projects on each agenda all follow rules, so they run as fixed software. AI reads free-text declarations and suggests which organisations and projects they relate to, and the governance officer confirms each match. How a conflict is managed is decided by the chair or the delegate under your policy.

Contracts, milestone reports, board papers, correspondence and financial records all need to be filed against the right project or function and kept for the period your retention schedule sets. In practice much of it sits in inboxes and personal drives, and the records officer works backwards to find what should have been captured. Filing documents from the project, finance and governance processes on this page into the records system with the right classification and metadata runs as fixed software, and so does flagging records as they reach the end of their retention period. AI reads loose documents and emails and suggests a classification, and anything it is unsure of goes to a review queue. The records officer approves every disposal, and nothing is destroyed without that approval.

Finance and the back office

Income arrives from several directions: levy remittances collected on the organisation’s behalf, matching contributions from government, member subscriptions, co-investment from research partners and grant payments. Each comes with its own statement and timetable, and finance matches the bank to each statement, records the income against the right fund or program and works out which contributions can be claimed. The matching follows rules, so it runs as fixed software, with each receipt matched to its statement, income coded to the fund or project, and unmatched items listed with their age. AI reads remittance advices and statements that arrive as documents rather than files. The finance officer approves the reconciliation, and any claim for matched funding is prepared for the person accountable for it to check and lodge.

Alongside milestone payments, a research organisation pays for laboratory and field services, analysis, consultants, events and a steady stream of travel to trial sites, field days and meetings. Invoices arrive by email in every format, card transactions need receipts and a project code, and each cost has to land against the right project or fund. Matching invoices to purchase orders or approved budgets, routing approval to the budget holder under your delegations, and chasing missing receipts for card transactions all run as fixed software. AI reads invoices and receipts that capture tools cannot and suggests the project code. The budget holder approves each cost, and anything over budget or without a clear project goes to them to decide.

Each month finance closes the ledger, accrues milestone payments that are due, and reports spend against each program and project. Each year, and at the end of each grant, the organisation acquits its funding, showing how the money was spent against the agreed budget and conditions. Much of this is the same each cycle, so the extraction, the control totals, the budget-to-actual comparison and the report templates run as fixed software on a set day, and every variance is listed rather than netted off. AI drafts a few lines on what changed for the finance manager to edit. The figures come from your systems, not from AI, and every acquittal or report to a funder or the board is reviewed and signed off by the person accountable for it.

Retyping between the project, finance, records and board systems

A new project is entered in the project system, again in the finance system when the purchase order is raised, again in the records system when the agreement is filed, and again in the board paper that reports on the portfolio. A milestone accepted by the research manager is typed into an email to finance, keyed again as a payment and summarised again for the committee. A change made in one place rarely reaches the others, so a varied milestone date can still trigger a reminder, or a payment can be released against a superseded budget. Each hop looks small, so nobody measures it as one process. Almost all of it is fixed software, with field mappings between systems, a scheduled sync, reconciliation checks and an exception queue for records that do not line up. Where a system has no way to exchange data, the entry stays with a person, and we prepare it so it can be checked and entered quickly. AI is used only to map free-text notes into the right field. We measure the whole chain as one process first, so the cost is visible before anything is built.

If one of these costs you real money every month, describe it in four answers.

Thinking about what a process actually costs you?

The assessment is small, fixed-price, and tells you the real number, whether or not you ever build.

Describe the process

Worth a 30-minute conversation.